What is Atlas

Atlas is a filterable, time-indexed dashboard that compares the latency of data providers across different chains and regions. Starting with RPCs, we aim to expand the provider type to API providers and indexers.

Today on Atlas, you can:

• Use the dashboard to get an overview of RPC provider performance over a custom period.

• Filter by chain, region, and provider to see median latency, p95, block propagation delay, and the average latency for a debug endpoint.*

• Navigate to a specific provider to gain further insight into their request latency over the past seven days, the most performant chains and regions, latency across the chains they offer, and the availability of WebSocket vs. HTTP endpoints.

Who is Atlas for?

Atlas can be used by:

• Developers who want to compare RPC and other data providers’ performance for any given chain(s)

• Researchers interested in monitoring data provider performance might use this data to inform reports and articles about providers and chains.

• Providers may use this data to compare the performance of their infrastructure vs. that of others in the industry.

How Atlas works

For each endpoint, multiple independent probes are deployed per region to ping and measure response times from a given provider’s API. A probe scheduler is also deployed per region to spawn, orchestrate, and execute their respective probes. Each endpoint supports the default JSON-RPC scheme provided by the Ethereum Foundation, ensuring compatibility with any EVM-compatible chain.

Implemented in Rust, these probes send all measurements to a single BigQuery database. Daily queries transform the raw data into its final form, ready for analysis in Looker Studio.

Probes

Several probes measure different aspects of RPC latency:

• erc20_balance: Measures the time to query the balance of a wallet for a given ERC20 token.

• eth_balance: Measures the time to query the balance of a wallet for the native chain token.

• eth_block_number: Measures the time to query the current block number of the chain.

• http_basic and websocket_basic: Measure the response times for HTTP and WebSocket requests.

• websocket_block_propagation and websocket_mempool_propagation: Measure block and mempool transaction propagation times over WebSocket connections.

Looker Studio

Data is analyzed on Looker Studio, which converts the raw data into customisable dashboards to provide insights such as:

• Best location per chain: Average request latency per region.

• Lowest request latency by region: Detailed breakdown of latency metrics by region.

• Fastest chains and providers: Overall performance metrics by chain and provider.

You can navigate to Atlas’ full Methodology page for more details.

What's Next for Atlas?

• Submission flow for data providers to add their endpoints. Atlas is launching with 38 RPC providers for comparison, we want a streamlined way for new and existing providers we may have missed to include themselves as well.

• Improving UI/UX on the dashboard. Faster load times, more filters, and customizable historical performance view.

• Expanding from RPCs to API providers and indexers. This includes non-standardized abstracted endpoints such as token, NFT, and transfers

• Adding data for paid endpoints. Atlas currently compares free or public endpoints, but paid endpoints with SLAs can tell a different performance story. We’d like to work with data providers for the best way to incorporate these in the dashboard.

• Adding more networks (Bitcoin, Solana, etc.)

Your feedback is invaluable as we strive to refine and enhance Atlas. Please reach out on Twitter or n@1kx.capital to learn more or if you are interested in helping with any of the above.

Most recently, he spent a year at Lattice, where he worked on strategy for MUD and Redstone and was the product lead for Sky Strife. Prior to that, he spent five years at Dapper Labs, from the early CryptoKitties days to the release of Flow and Topshot. Before web3, Rafael was a game designer with CCP (Eve Online) and Ubisoft.

We sat down with him for an informal Q&A to get to know him better. Here’s that conversation.

You were at Dapper Labs during the last bull run. What was that like?

Intense. When Topshot got full traction, everyone had to level up very fast just to keep up with the market. I never learned as much as I did during that time. It was proof that some of the issues that blocked wider adoption were, within that context, solved.

What issues are you referring to? And do you still think this is the case today?

At that point in time NFTs were really cool, even beyond the expectation of financial returns. There’s a lot of subtlety here in this value prop (being cool) but there was this ‘Supreme drop’ vibe around TopShot drops, and the wallet creation and credit card payment mechanics were stellar. Because consumers were interested, brands also became interested. As we all know, that didn’t last.

In any case, I don’t think big brands are the next necessary step to expand the user base. Small innovative teams that are crypto-native are probably in a better position to generate interest with unique functionality stemming from the technology. Dozens of companies are working to solve UX issues. The value prop side of things is something that will require “shiny playful things” or “super useful things”. While convenience is necessary for mass adoption, there’s a lot of room for growth even as things remain a bit cumbersome.

That’s a perfect segway to our next question. Who are some of those web3 gaming projects & teams you are most excited about right now?

Project Awakening, since CCP is probably the only team that has had the same game running with a proper player-driven economy for more than a decade. The way it works is that all items in the game are made by players. When you want to buy a ship, there are different options/prices because those are all set by the players. This creates a lot of opportunities for players (arbitrage, exploration of rare resources, ambushing convoys, etc). The fact that players can trade amongst themselves is not usual in multiplayer games and most game systems would break down with free trading.

But truth be told, I’m more excited about the ecosystem, innovation and learnings coming out of small projects than I am about any one larger game. Small Brain Games has done words3, an infinite multiplayer scrabble for actual money, which is very interesting. Fomo3D, and specifically its ending, was really fun back in the day. I’m interested in any projects that focus on the fundamentals of blockchain and go beyond things like “own and trade a sword” as player motivation.

What are the main blockers you see to the adoption of web3 gaming?

One of the most fundamental mistakes some teams make is to think that gamers care about new technologies. Traditional gamers are very conservative and don’t tend to try out new things just for fun. It took self-identified gamers way longer than developers expected to start playing games on mobile for instance, as compared to more casual game players.

I also think there are many “bad” games out there, and by bad I mean games that are not focused on longer retention or providing a fun recurring experience. The whole wallet setup to interact, fueling the wallet with gas from an exchange, all of that is really annoying and a significant hurdle, but I believe there’s a lot of room to grow even without completely fixing that.

Technical limitations such as perfect information, no easy randomness function, and slow tick are also frustrating, but I don’t think they are the biggest blockers right now.

In a way, it’s the powerful distribution mechanics that can be the biggest barriers to sustainable adoption, because by creating a spike of interest, which then more often than not plummets, you sacrifice later cohorts and long-term retention.

How were you first introduced to 1kx, and how did this opportunity come about?

I first met Peter Pan at a gathering in Lisbon organized for people interested in Fully-On-Chain applications, which eventually became the Autonomous Anonymous group. Many of the teams I already knew through my work at Lattice, because they were using MUD. After Sky Strife was released, I was invited to take on an EIR position within 1kx. I had always been curious about the VC world and their perspectives on the market and I specifically felt like 1kx would be an ideal ground for experimentation in onchain gaming.

What will you focus on, and how can people get involved?

In the short term, I plan on helping ideate with teams from the current portfolio and vetting and probing new projects. In the longer term, stay tuned! To get in touch, you can reach me at @nagual_ape.

How Prediction Markets Work
Bottlenecks Preventing Greater Adoption of Prediction Markets
- Supply Side
- Demand Side
Solutions to these problems
- Supply Side
- Demand Side
Other ways to increase adoption

Prediction Markets: Bottlenecks and the Next Major Unlocks

Augur, the first on-chain prediction market, was among the earliest applications ever launched on Ethereum. The aspiration was to allow anyone to bet any size on anything. Plagued with issues, Augur’s vision fell short of reality many years ago. Lack of users, poor settlement UX, and high gas fees led the product to shut down. Since then we have come a long way: blockspace is cheaper, and order book designs are much more efficient. Recent innovations solidify that crypto’s permissionless and open-source nature allows for an unconstrained and global liquidity layer in which anyone can become a market participant through liquidity provisioning, market creation, or betting.

Polymarket has been the emergent market leader, with approximately $900 million in volume to date, and SX Bet has amassed $475 million thus far. Despite this, there is substantial room for growth, especially when compared to the massive scale of sports betting, a subcategory of traditional prediction markets. Sportsbooks in the United States alone have handled over $119 billion in 2023. The volume figure is much more prominent when considering land-based and online sports betting volume in all other countries combined and other types of prediction markets such as politics and entertainment.

This article aims to break down how prediction markets work, the current bottlenecks that would need to be addressed for further adoption at scale, and some ways we think they can be solved.

How do prediction markets work?

There are several ways to design prediction markets, and most can be divided into two categories: order book models and concentrated AMM (Automated Market Maker) models. Our thesis is that order book models are the more optimal design choice because they allow for better price discovery, enable maximum composability, and ultimately lead to volume at scale.

For order book models, each market has only two possible pre-defined outcomes: Yes and No. Users trade these outcomes in the form of shares. At the time of market settlement, the correct share is worth $1.00, and the incorrect one will be worth $0.00. Before market settlement, the price of these shares could trade anywhere between $0.00 and $1.00.

In order for share trading to occur, liquidity providers (LPs) must exist; in other words, they must provide bids and asks (quotes). These LPs are also referred to as market makers. A market maker provides liquidity in exchange for a small profit in the spread.

An example of a specific market: if there is an even chance something will happen, such as a coin toss landing on heads, both “Yes” and “No” shares should theoretically trade at $0.50. However, much like in any financial market, there tends to be a spread and thus slippage. If I wanted to buy “Yes” shares, my execution price might end up being closer to $0.55. This is because my counterparty, a liquidity provider, is purposely overpricing the true odds to make a potential profit. The counterparty might also be selling “No” shares for $0.55. This spread of $0.05 per side is compensation that the liquidity provider gets for offering quotes. Spread is driven by implied volatility (expectation of price movement). Prediction markets essentially have guaranteed realized volatility (actual price movement) simply due to the design where shares must eventually reach a pre-destined value of $1 or $0 by some pre-defined date.

To illustrate a market maker scenario for the coin toss prompt above:

1. A maker sells 1 “Yes” share for $0.55 (this is equivalent to buying a “No” share at $0.45)

2. A maker sells 1 “No” share for $0.55 (this is equivalent to buying a “Yes” share at $0.45)

3. The maker now has 1 No Share and 1 Yes Share that it paid $0.90 for in total

4. Regardless of whether the coin lands Heads, the maker will redeem $1.00, earning a $0.10 spread

The other primary way of settling prediction markets is through concentrated AMMs, used by both Azuro and Overtime. For the sake of this article, we will not be touching on these models too much, but the analogy in DeFi is GMX v2. Capital is pooled, and acts as the single counterparty to platform traders, and the pool relies on external oracles to price the odds that are offered to users.

What are the current bottlenecks for prediction markets?

Prediction market platforms have been around and heard about for a long enough time that escape velocity would have happened already if there was true product market fit. Current bottlenecks boil down to the simple conclusion of a lack of interest on both the supply (liquidity provider) and demand (bettor) side.

On the supply side, problems include:

Lack of liquidity due to volatility: Polymarket’s most popular markets tend to be conceptually novel ones with scarce relevant historical data, making outcomes challenging to predict and price accurately. For instance, predicting whether or not a CEO, such as Sam Altman, will return to his position after rumors of potential AGI mishandling is difficult because no past events closely mirror this scenario. Market makers will put up larger spreads and less liquidity on uncertain markets in order to compensate for implied volatility (i.e. wild price action in the Sam Altman CEO market, where consensus flipped 3 times in less than 4 days). This makes it less appealing to whales who want to bet in size.

Lack of liquidity due to few subject matter experts: Although hundreds of market makers earn rewards daily on Polymarket, many longer-tail markets lack liquidity simply because there are not enough parties equipped with specialized knowledge who also want to market make. Example markets include “Will x celebrity get arrested or charged for y?” or “When will celebrity x tweet?”. This will change over time as more prediction market types are introduced, data becomes more abundant, and makers become more specialized.

Information Asymmetry: Since makers provide bids and asks that any taker can trade against at any time, the latter party has the advantage of making positive EV bets when they acquire advantageous information. In DeFi markets, these types of takers can be referred to as toxic flow. Arbitrageurs on Uniswap are good examples of toxic takers because they use information edges to continuously extract profit from liquidity providers.

One Polymarket market, “Will Tesla announce a Bitcoin purchase before March 1, 2021?” saw a user purchase $60,000 worth of Yes shares at ~33% odds. This market was the only market the user had ever participated in, and it can be assumed that this user had advantageous information. Legalities aside, there is no way the maker who offered this quote could have known that the taker/bettor had this advantageous information at the time that it did, and even if the maker had originally set their odds at 95%, the taker probably still would have placed the bet, because the true odds were 99.9%. This results in a guaranteed loss scenario for the maker. In prediction markets, it is hard to predict when toxic flow will occur and at what size, therefore making it harder to provide tight spreads and deep liquidity. Makers need to price the risk of toxic flow occurring at any time.

On the demand side, the main issue is:

Lack of a leverage vehicle: Without a leverage vehicle, the value proposition of prediction markets to retail users is relatively inferior to that of other crypto speculation vehicles. Retail wants to make “generational wealth,” which is more likely achieved on memecoins than betting on a prediction market with a capped upside. Since its inception, betting on $BODEN and $TRUMP early has yielded far more upside than Biden or Trump Yes shares for winning the presidency.

Lack of exciting short-term markets: Retail bettors have no interest in placing bets that settle months from now, and this conclusion can be proven in the sports betting world, where a lot of retail volume occurs on live betting (super short-term) and daily events (short-term) nowadays. Not enough short-term markets appeal to a mainstream audience, at least not yet.

What solutions are there to these problems? How can we increase volume?

On the supply side, the first two problems, related to lack of liquidity due to volatility and lack of liquidity due to few subject matter experts, will naturally lessen over time. As the volume grows across various prediction markets, the number of specialized makers and those with higher risk tolerance and capital will also grow.

However, instead of waiting for these problems to lessen over time, the lack of liquidity can be addressed head-on through liquidity coordination mechanisms, originally invented in the DeFi derivatives space.  The idea is to allow passive stablecoin depositors to earn yield through vaults, which deploy market-making strategies across different markets. This vault would act as the primary counter-party against traders. GMX was the first to do this through a pool-style liquidity provisioning strategy that relied on oracles for pricing, and Hyperliquid was the second notable protocol to deploy a native vault strategy, but with the kicker that the liquidity was provided on a CLOB. Both of the vaults have been profitable over time because they are able to act as the counterparty to mostly non-toxic flow (retail users that tend to lose money over time).

Native vaults make it easy for protocols to bootstrap liquidity themselves without relying on others. They also make long-tail markets more appealing; one reason Hyperliquid became so successful is that newly listed perpetual assets contain tons of liquidity from Day 0.

The challenge with building a vault product for prediction markets is to prevent toxic flow. GMX prevents this by attaching high fees to their trades. Hyperliquid implements market maker strategies with large spreads, has a 2 block delay for taker orders to give time to makers to adjust their quotes, and prioritizes maker order cancels within a block. Both protocols create an environment where toxic flow does not enter because they can find better price execution elsewhere.

In prediction markets, toxic flow can be prevented by providing deep liquidity at wider spreads, selectively providing liquidity to markets less susceptible to information advantages, or employing a sharp strategist with access to information advantages.

In practice, a native vault could deploy $250,000 of additional liquidity as a bid at 53 cents and as an ask at 56 cents. Wider spreads help increase potential vault profits because a user would be betting while accepting worse odds. This is opposed to placing quotes at 54 and 55 cents, where the counterparty trader might be an arbitrageur or sharp looking for good prices. This market is also relatively less susceptible to information asymmetry issues (less insider info and insights usually come out to the public relatively quickly), so the expectation of toxic flow is lower. The vault could also use information oracles that provide insight into future line movement, such as pulling odds data from other betting exchanges or gathering information from top political analysts on social media.

The result is deeper liquidity for bettors, who are now able to bet larger sizes with less slippage.

There are several approaches to solving, or at least reducing, the information asymmetry issue. The first few are around order book design:

Gradual Limit Order Book: One way to combat toxic flow is by increasing the price based on the combined speed and size of an order. If a buyer is certain that an event will occur, the logical strategy would be to buy as many shares as possible at prices below $1.00. Additionally, purchasing quickly would be sensible in the event that the market eventually accesses the advantageous information.

Contro is implementing this GLOB idea and is launching as an interwoven rollup on Initia.

If the Tesla $BTC market occurred on a GLOB model, the taker would have had to pay much more than 33% per “Yes” share, as there would be “slippage” given the combined speed (one clip) and size (huge) of the order. He still would have made a profit regardless of slippage because he knew his “Yes” shares would eventually become worth $1, but it at least contains the maker’s loss.

One could argue that the taker still could have incurred little slippage and paid close to 33% per “Yes” share if they had just performed a DCA strategy over a long period of time, but in this scenario, it at least gives some time to the market maker to withdraw its quotes from the book. A maker could withdraw for several reasons:

1. it suspects that there’s toxic flow because there’s such a large taker order coming in

2. it is convinced that there is toxic flow because it checked the taker’s profile and saw that it had never made a bet before

3. it wanted to re-balance its inventory and no longer wanted to be too one-sided given how many “Yes” shares it was selling and thus “No” shares it was accumulating – perhaps the maker originally had $50,000 worth of orders on the ask side at 33% and $50,000 worth of orders on the bid side at 27% – and its original goal is not to be directionally biased, but rather, neutral so it could earn profits through symmetric liquidity provisioning

A winner’s rake: There are many markets whereby a portion of the profits of those with advantageous information get re-distributed. The first example is in peer-to-peer web2 sportsbooks, specifically Betfair, where a fixed percentage of a user’s net winnings will be re-distributed back to the company. Betfair’s rake actually depends on the market itself; on Polymarket, it could be rational to charge a higher rake on net winnings for markets that are more novel or long-tail.

This re-distribution concept exists similarly in DeFi in the form of order flow auctions. A backrunning bot captures value from information asymmetry (arbitrage) and is forced to give back to those who participated in the transaction, which could either be the liquidity providers or the user making a trade. Orderflow auctions have seen plenty of PMF to date, and CowSwap* is pioneering this category through MEVBlocker.

Static or dynamic taker fees: There are currently no taker fees on Polymarket. If this were implemented, proceeds could be used for liquidity provisioning rewards in markets with high volatility or greater susceptibility to toxic flow. Alternatively,  taker fees could be set higher on long-tail markets.

On the demand side, the best way to solve the lack of upside is to create a mechanism that allows for it. In sports betting, parlays have become increasingly popular for retail bettors because they offer a chance at “winning big.” A parlay is a type of bet combining multiple individual wagers into one bet. For the parlay to win, all of the individual wagers must win.

There are three main ways to increase upside for users in crypto-native prediction markets:

• Parlays

• Perpetuals

• Tokenized leverage

Parlays: This is technically infeasible to implement on Polymarket’s books because upfront capital is needed for bets, and the counterparty for each market is different. In practice, a new protocol can source odds from Polymarket at any given point to price any parlay bet’s odds and act as the single counterparty for parlays.

For example, a user wants to bet $10 on the following:

These bets give limited upside if placed on their own, but when combined into a parlay, the implied return skyrockets to ~1:650,000, meaning that the bettor could win $6.5 million if every bet is correct. It is not too hard to envision how parlays could gain PMF amongst crypto users:

• The cost to participate is cheap, and you can put up very little to win a lot

• Sharing parlay slips would become viral on Crypto Twitter, especially if someone wins big, which creates a feedback loop with the product itself

Supporting parlays presents challenges, namely counterparty risk (what happens when multiple bettors win big parlays at the same time) and odds accuracy (you don’t want to provide bets where you are underestimating the true odds). Casinos have tackled the challenge of offering parlays in the sports world and it has become by far the most profitable component of sports betting. The profit margins are ~5-8x higher than offering single-market bets, even if some bettors get lucky and win big. The other added benefit of parlays is that there’s relatively less toxic flow compared to single-game markets. The analogy here is: why would a professional poker player who lives and dies by expected value put money into lottery tickets?

SX Bet, a web3 sports betting appchain, launched the world’s first peer-to-peer parlay betting system and has done $1M in parlay volume in the past month. When a bettor “Requests Parlay”, SX creates a private virtual order book for the parlay. Programmatic market-makers listening through the API will then have 1 second to offer liquidity on the bet. It would be interesting to observe increased liquidity and traction around non-sport parlays.

Perpetual Prediction Markets: This concept was explored briefly in 2020 when former leading exchange FTX offered perpetuals for American Election outcomes. You could long the price of $TRUMP and, if he won the American Election, each share could be redeemed for $1. FTX had to change margin requirements as the odds of him actually winning changed. Creating a perpetual mechanism for markets as volatile as prediction markets creates a lot of challenges for collateral requirements because prices can be worth $0.90 one second and $0.1 the next. Thus, there might not be enough collateral to cover the losses of someone who longed the wrong way. Some of the order book designs explored above could help compensate for the fact that prices could change so quickly. The other interesting part of the FTX $TRUMP market is that we can reasonably assume that Alameda was the main market maker in these markets and that without natively deployed liquidity, the books would have been too thin for substantial volumes to occur. This highlights how valuable a native liquidity vault mechanism might be for prediction market protocols.

Both LEVR Bet and SX Bet are currently working on perpetual sports betting markets. One plus about leverage on sports is that the price of “Yes or “No” shares will fluctuate less violently, at least most of the time. For example, a player making a basketball shot might boost a team’s odds of winning the game from 50% to 52%, as on average a team might make 50 shots per game. The 2% boost on any given shot is a manageable increase from a liquidation and collateral requirement perspective. Offering perpetuals at the end of a game is a different story, as someone might hit a “game-winner” and the odds could flip from 1% to 99% in half a millisecond. One potential solution is to only allow for leveraged betting up to a certain point because any event after would change the odds too much. The feasibility of perpetual sports betting also depends on the sport itself; a single hockey goal changes the expected game outcome much more than a single made basketball shot.

Tokenized Leverage: A lending market that allows users to borrow against their prediction market positions, especially ones of long-term duration, may increase the volume among pro-traders. This can also lead to more liquidity, as market makers could borrow against positions in one market to make in another. Tokenized leverage probably wouldn’t be an interesting product to retail bettors unless there was an abstracted looping product, such as the ones that gained traction for Eigenlayer. The entire market is probably too immature for abstraction layers like this to exist just yet, but these types of looping products will eventually come.

Aside from the pure supply and demand aspects, there are other minute ways to increase adoption:

From a UX perspective: Switching the settlement currency from USDC to a yield-bearing stablecoin would increase participation, especially in long-tail markets. This has been discussed on Twitter a few times; holding positions for markets that expire at the end of the year has a significant opportunity cost (e.g. earning 0.24% APR by betting on Kanye West to win the presidency versus earning 8% APR on AAVE).

Moreover, increased gamification that aims to improve retention can really help drive more users in the long run. Simple things such as “daily betting streak” or “daily contests” have worked well in the sports betting industry.

A few sector-level tailwinds will also increase adoption in the near future: The combination of growing virtual and onchain environments will unlock a whole new level of speculative demand because the number of short-term events will eventually be unlimited (think AI/computer-simulated sports), and the level of data will be abundant (this makes it easier for market makers to price outcome odds). Other interesting crypto-native categories include AI gaming, onchain gaming, and general onchain data.

Accessible data will lead to an increased level of betting activity from non-humans, more specifically autonomous agents. Omen on Gnosis Chain* is spearheading the idea of AI agent bettors. Since prediction markets are a game where the outcome is defined, autonomous agents can become increasingly skilled at calculating expected value, likely to a precision far better than a human. This mirrors the idea that AAs would probably have a harder time predicting which memecoins will take off because there is more of an “emotional” element to what makes them successful, and humans are much better at feeling emotions than AAs at the moment.

In summation, prediction markets are a fascinating user product and design space. As time goes on, the vision of allowing anyone to bet any size on anything will become a reality. If you’re building something in the space, whether it be an entirely new protocol, a liquidity coordination platform, or a new leverage mechanism, please reach out! I am an avid user and would love to give feedback.

Thank you to Peter Pan, Shayne Coplan, Sanat Kapur, Andrew Young, taetaehoho, Diana Biggs, Abigail Carlson, Daniel Sekopta, Ryan Clark, Josh Solesbury, Watcher, Jamie Wallace, and Rares Florea for the feedback and for reviewing this piece!

Disclaimers:

*Denotes a 1kx portfolio investment.

This article is for general information purposes only and should not be construed as or relied upon in any manner as investment, financial, legal, regulatory, tax, accounting, or similar advice. Under no circumstances should any material at the site be used or be construed as an offer soliciting the purchase or sale of any security, future, or other financial product or instrument. Views expressed in posts are those of the individual 1kx personnel quoted therein and are not the views of 1kx and are subject to change. The posts are not directed to any investors or potential investors, and do not constitute an offer to sell or a solicitation of an offer to buy any securities, and may not be used or relied upon in evaluating the merits of any investment. All information contained herein should be independently verified and confirmed. 1kx does not accept any liability for any loss or damage whatsoever caused in reliance upon such information. Certain information has been obtained from third-party sources. While taken from sources believed to be reliable, 1kx has not independently verified such information and makes no representations about the enduring accuracy or completeness of any information provided or its appropriateness for a given situation. 1kx may hold positions in certain projects or assets discussed in this article.

DePINs leverage token incentives to mobilize individuals to participate in infrastructure networks. These individuals earn tokens in exchange for operating nodes on these networks. Given it costs money to operate nodes, ensuring that token rewards offset operational costs is essential.

The problem? Because DePIN networks are decentralized, network stewards don’t always know how much it costs for node operators to participate in networks. We did some research and came up with a model that helps estimate the total costs of operating networks. You can find that research here.

As part of that, we did a series of case studies in partnership with DePIN networks, which you will find below. To get in touch with us about this research, you can reach out to me here.

Key Takeaways
What are DePINs and why talk about costs
The framework to come up with a cost estimate:
Step 1: Identify network contributors
Step 2: Assess cost components
Step 3: Evaluate differences in cost structures and aggregate
Examples

Key takeaways

• To ensure continued participation of nodes in a Decentralized Physical Infrastructure Network (DePIN), it is crucial for network stewards (founders, DAO members, etc) to consider costs incurred by operators when running nodes.

• In some cases, key decisions on cost optimization are straightforward. For example, Livepeer switched from Ethereum to Arbitrum in 2022, an uncontested choice which reduced settlement costs by more than 95%. In other cases, DePIN stewards with limited R&D resources may benefit from external help to evaluate the costs of running nodes.

• If nodes operate at a continued loss, operators will stop running them, decreasing the overall node supply. Knowing the cost of operating a DePIN network and its largest drivers can equip network operators to kickstart governance discussions; simultaneously, cost estimates may inform R&D efforts to reduce node operator costs prior to the supply of network services starting to degrade.

• Estimating the costs of network operations may be difficult for protocol stewards given the anonymity of contributors (these networks are often permissionless, meaning anyone can contribute and come/go at will) and the lack of public data related to costs.

• To guide stewards’ decisions,  we propose a framework with three steps to estimate costs:

• Besides the overall estimate of current costs, this framework gives:

  1. A breakdown per role and cost component, helping to identify the largest cost drivers

  2. Indications of how this changes with different assumptions underpinning the estimates and scenarios on increasing demand/network capacity

• Examples serve as a walkthrough of how to apply this framework. For instance, a joint survey with the POKT Network revealed continued efforts by node operators to further scale service nodes. That said, the remaining block for economic scalability (including demand generation) was tackled with the decentralization of their gateway.

Introduction: what are DePINs and why talk about costs

DePINs are decentralized networks of nodes providing hardware resources (physical infrastructure) for a wide range of use cases like compute, storage, wireless networks, or data measurement. DePINs leverage Web3 incentive models (ie, token reward systems) to incentivize the buildup of physical infrastructure networks. As of May 2024, the market capitalization of all DePIN tokens was $29B.

DePINs contribute to both digital and physical resource networks:

• In Physical Resource Networks (PRN), contributors deploy location-dependent hardware to offer (non-fungible) services. This includes:

  • Wireless networks (e.g. Helium, World Mobile, XNET, Nodle)

  • Sensor networks (e.g. Dimo, Hivemapper, Silencio, Onocoy)

  • Energy networks (e.g. Starpower, PowerLedger, Arkreen)

• In Digital Resource Networks (DRN), contributors direct hardware to offer (fungible) digital resources where the physical location is not the main criteria. This includes:

  • Compute (e.g. ICP, Livepeer*, Akash Network, POKT Network*, Covalent*, Lit protocol*)

  • Storage (e.g. Arweave*, Filecoin, Sia)

  • Bandwidth and privacy (e.g. NYM*, Hopr, Orchid, Mysterium, Fleek)

  • AI (e.g. Bittensor, Fetch.ai, Modulus Labs*)

Early DePIN projects generated substantial initial interest due to their token framework designs. For instance, Helium rewards contributors with HNT tokens to help run the wireless network via hotspots, and Filecoin allows users to rent out their excess storage space. While this was sufficient to get many DePIN projects off the ground, token emissions may not be enough to guarantee the long-term participation of nodes in networks.

If running a node becomes unprofitable, node operators will no longer be incentivized to operate DePIN infrastructure. It is therefore essential for DePIN founding teams to help optimize costs for node operators.

The DePIN Flywheel

The typical flywheel for the DePIN token economy goes like this:

1. Build up the supply side for services, e.g., storage or 5g antennas

2. Inflationary token rewards incentivize node operators to provide the required infrastructure while demand isn’t yet sufficient to cover costs alone

3. Over time, as demand picks up, monetizing network activity may increase node operators’ income even as token rewards decline

4. Continued monetization of the network activity and increase in node operator income further incentivize the supply, thus creating the DePIN flywheel

A visual depiction of what the DePIN flywheel should look like:

As we described in our analysis of the reward emission schedules, the dollar values of those token rewards (the token prices) are heavily influenced by overall market sentiment. Hence, they might look like this:

Or this, depending on when you enter the bull market:

So, What do Token Reward Emissions have to do with Costs?

As mentioned above, if token rewards and income from user demand do not suffice to break even, node operators may decide to stop supporting the network. A large part of operational bills for DePINs are paid in fiat, which makes the dollar value of token rewards important and tied to general market performance. Despite any measure of well-planned token emissions, dynamics could look like this in a worst-case scenario:

This would lead to node operators falling off, which would translate into higher latency, lower reliability, and worse user experience for the end-user. Eventually, stagnating demand would turn off the flywheel.

The good news is that there are many ways to counter this. One way is to make the token emissions more flexible so they are more in line with the monetization of the network (see KPI-based emissions here). The other way is to tackle costs in order to make the overall network more efficient and hence less sensitive to decreasing token prices. Our dynamic would then look like this:

Key Proposition: If you know the cost of operating your DePIN network and its largest drivers, you can kickstart governance discussions and R&D efforts in order to reduce node operator costs before the supply of network services degrades.

Given the decentralized and permissionless nature of DePINs, assessing the cost base is no easy task. While token-based rewards and income are often tracked on-chain, other costs involved in running a node aren’t public (e.g. infrastructure bills). This means we need to use assumptions and estimates around the available data points.

In this article, we will tackle this challenge and walk you through the framework to come up with an estimate

• Step 1: Network contributors

• Step 2: Cost components

• Step 3: Evaluate cost structures of network contributors

The framework

We propose the following framework for stewards of DePIN networks as a methodology to assess the operational costs involved in running infrastructure nodes.

Using this framework, the cost estimation of DePINs is broken down into three steps:

1. Identify network contributors

2. Assess cost components (e.g. hardware, labor)

3. Evaluate cost structures for the above and add it up to derive the overall cost estimate

Step 1: Identify network contributors

Despite the various services DePINs provide (e.g. compute, network coverage, mobility data, etc), the roles needed to provide these services are the same (see here for an overview of DePIN supply side roles in 30+ networks):

• Service nodes/producer: they provide the service and its required physical infrastructure (e.g. servers, antennas, dashcams, etc). Examples include storage providers for Filecoin, Hotspots for Helium, or Transcoders for Livepeer.

• Validators/watcher nodes/fishermen: they check the work done by service nodes, either directly or via the accounting layer. The result of those checks is then sent to the accounting layer. Examples include storage providers for Filecoin (as they also validate storage proofs of other providers) and Hotspots and Oracles for Helium (performing the proof of coverage of other hotspots).

• The accounting layer: this tracks the flow and status of work/services provided and corresponding payments. Note that protocols define the accounting logic themselves, e.g. how work and payments are tracked and stored on the blockchain (we will examine this in a separate article). Examples include Arbitrum for Livepeer or the POKT-chain for POKT network (operated by POKT validator nodes).

• Gateway: they function as an orchestrator/balancer between users, service nodes, and also the accounting layer when it comes to managing access or aggregating services (e.g. data in sensor networks). Examples include Orchestrators for Livepeer or the Gateways in POKT network.

• Delegators: can participate in the economies of service - or watcher nodes, via staking.

Roles related to the demand side (such as sales teams) are not common yet and assessing the costs related to running the protocol, like governance costs, is a topic for a separate article.

Note that not every DePIN has delegation and gateways, nor do all roles need to be separated. For example, Filecoin storage providers (SP) are classified as both service nodes and validators and also operate the Filecoin chain, therefore also forming the accounting layer. The same applies to Arweave miners.

Step 2: Assess cost components

Each of the aforementioned roles can be performed via nodes whose costs come in any of the following four components (most have more than one):

• Hardware/infrastructure: costs related to the actual physical infrastructure, e.g. dashcam

• Labor: costs related to time spent in setting up and operating the infrastructure

• Bandwidth, power and other OPEX: costs related to data exchange as well as other operational costs, e.g. power, datacenter rental

• Staking: the (opportunity) cost of not investing elsewhere

The last point refers to the cost of capital: Obtaining information around debt/financing costs related to those operations is almost impossible on a broad scale. Yet, there is one part related to capital cost which we can assess: many DePINs follow the stake for access (work-token) model and require node operators to stake some tokens in order to be allowed to contribute. Acquiring these is an investment and even if we assume that this amount can be retrieved back when leaving the network, there is opportunity cost in holding these tokens vs. investing capital somewhere else.

Our evaluation of cost components would be incomplete without touching on costs related to transactions on the accounting layer. Evaluating this is not straightforward; it depends on several moving pieces. In general, networks decide the extent to which they aggregate book-keeping off-chain. But for track records of the settlement-layer and transactions on-chain, there are three options:

• Proprietary L1: The network operates its own blockchain. Examples are Arweave, Filecoin and POKT Network. Typically service and validator-nodes also cover this role, which is why the associated costs are included there as well (however, if possible, we try to separate them -  see POKT Network in the examples).

• Proprietary L2, better known as app-chain or app-specific rollup: The costs for the rollup infrastructure (sequencers, etc) and adjacent infrastructure (block explorers, wallet integration, etc.) are typically mappable into the four components. Less clear cases, e.g. when using rollup-as-a-service provider (RaaS), will be mapped into Bandwidth and other costs.

• Public L1/L2s: These outsource the settlement layer, meaning there is no cost for hardware and labor to the network. However, service- and validator-nodes (and users/payers) pay directly (usage based).  Assessing the network related costs for those transactions has some challenges and consequently limitations: not all transactions are related to the accounting layer, e.g. swaps or other DeFi transactions, yet it is often not trivial to separate these transactions. We map these costs into Bandwidth and other costs..

Putting all of these pieces together to create a cost estimate is a challenging task. Not only do we need to come up with estimates for each cost component for each role in our network, like in the diagram below, we also need to consider that not all node operators have the same cost structure. Determining the overall cost estimate is more involved than simply multiplying the number of all network node operators by the estimate derived for one node operator.

Step 3: Evaluate cost structures

When we speak of cost structures, we refer to the key differences which impact cost. These key differences make it essential to rely on assumptions. Of course, this is a tradeoff: making assumptions simplifies the process, which can come at the expense of accuracy. That said, given how many factors are involved, certain assumptions must be made in order to achieve a working theory.

There are three primary considerations when evaluating cost structure:

1. Differences in setup: a typical example is one operator using a bare metal server vs.another one running on cloud (buy vs. rent). We can typically account for such differences when we know the corresponding shares across nodes of the overall network. It also relates to cost of capital when leasing or finance agreements are made. We recommend ignoring those differences and assume no cost of capital.
   
   Another difference in costs relates to the time of purchase (buying storage gets cheaper over time, buying H100s might not) or the location of operation. We recommend accounting for the time aspect by using current prices. For labor costs, the location matters: DePINs can source their contributors from all over the world with widely varying local wages for the (hard to assess) time invested. That said, we make the simplifying assumption of a general hourly wage across all node operators in our version of this framework.

2. Differences in efficiency: node operators can have the exact same setup, yet if one runs a lot more of the same nodes they likely have lower costs per node due to economies of scale. In our framework, we need to assess the distribution of nodes per node operator as a first step to account for these effects. Sizing those effects then needs a survey with larger and smaller operators or other available data points (e.g. promoted bulk discounts) to understand and estimate the cost impact.
   
   Another example are long term supporters of a network who are further along the learning curve and hence more efficient in their operations compared to someone who just joined. Unless we have direct data points from surveys, we ignore this aspect.

3. Differences in attribution and accounting: node operators being equal with respect to the two previous points could still view their contribution differently and hence end up with a different cost base. For example, one considers their engagement as a side-hustle not tracking any time spent whilst the other as their main business paying a salary based on the time spent on the project. We account for such differences via wider error margins for the “side-hustlers” side (given they are typically underrepresented), but assume similar time invested per node operated (see also economies of scale though)

   This relates to the sharing economy benefits, which are common for DePINs: Operators can utilize the same setup (and hence hardware, labor and bandwidth, power and other OPEX) across multiple networks (e.g. Livepeer with Ethereum and Filecoin operations, io.net with Render, Filecoin and other GPU networks). For cases where the hardware is essential for the operation, we don’t account for cost savings related to the sharing economy. Not only are they hard to identify, it’s also hard to quantify which network benefits most on the cost side and how to split savings¹. On the accounting side, we’ll want to break down total costs into monthly amounts. To simplify, we will assume a common period over which we deprecate the total amount and attribute equal amounts for each month of lifetime for all node operators.

There are, of course, more nuances , which we will dive into in a longer piece in our DePIN repository soon.

This adds the third dimension to our “execution plan”, creating sixty different combinations to consider:

Overall, while this formula is extremely comprehensive, leaving you with multiple cost structure options, it is most useful when also applied at several different points in time vs one static. The strongest model is one which ties the cost of operations to the capacity of the network. This enables us to understand the extent to which costs change with changing capacity or utilization. The capacity of the network relates to the service offered by the network, e.g. # of RPC requests for Pocket, TB of storage for Arweave or Filecoin, or percentage of road-network mapped for Hivemapper.

Note that this formula requires a lot of publicly available information, which we suggest obtaining  via documentation provided by the networks, from forum-/discord posts and, if feasible, surveys².

The takeaways and next steps

As DePIN is evolving at an increased pace, estimating the cost components of various DePINs is challenging. Besides known power laws on hardware costs and capacity over time, it’s not straightforward to estimate crypto-specific costs, e.g., gas and throughput capacity for settlement layers.

Knowing how current costs relate to reward emissions and demand side income, how the largest cost drivers change based on assumptions, and how costs increase with increasing demand are all useful indications

To help guide governance decisions around the economic design of DePINs, cost estimates need to be related to reward emissions and to income from usage. While I plan to provide more examples for cost estimates of DePINs, I welcome feedback on the presented framework, its assumptions and simplifications, as well as on the potential improvements to the cost estimates provided.

If you would like to walk through such an estimation together, please reach out to me.

A special thank you to my contributors: Mihai (Messari), Raullen (IoTeX), Nodies Team, Grove Team, Pocket Network Foundation, the DIMO team, Diana Biggs, and Christopher Heymann for feedback and input.

* denotes 1kx portfolio companies

Disclaimer: This article is for general information purposes only and should not be construed as or relied upon in any manner as investment, financial, legal, regulatory, tax, accounting, or similar advice. Under no circumstances should any material at the site be used or be construed as an offer soliciting the purchase or sale of any security, future, or other financial product or instrument. Views expressed in posts are those of the individual 1kx personnel quoted therein and are not the views of 1kx and are subject to change. The posts are not directed to any investors or potential investors, and do not constitute an offer to sell or a solicitation of an offer to buy any securities, and may not be used or relied upon in evaluating the merits of any investment. All information contained herein should be independently verified and confirmed. 1kx does not accept any liability for any loss or damage whatsoever caused in reliance upon such information. Certain information has been obtained from third-party sources. While taken from sources believed to be reliable, 1kx has not independently verified such information and makes no representations about the enduring accuracy or completeness of any information provided or its appropriateness for a given situation. 1kx may hold positions in certain projects or assets discussed in this article.

Appendix - Examples illustrating the framework

Livepeer Network

The Livepeer Network offers decentralized video infrastructure for live and on-demand streaming. Recently, Livepeer started to enable idle GPU resources for AI model training use cases as well (see details).

The framework applied step by step is provided here. Most cost estimates have been based on this survey conducted with node operators (i.e. Orchestrators) in summer 2023 and community information (like this).

The total estimated costs of operating the Livepeer network are about $85k per month. The breakdown of average costs shows that hardware and labor account for about the same share (~40%). If we account for the uncertainty around the labor cost estimates described in the spreadsheet, the monthly costs for the 100 Orchestrators of the network, their transcoders and settlement costs on Arbitrum add up to about $40k on the lower range of estimates. Notably, the $40k monthly cost is not too far from being offset by the current fee income of approximately 5-10 ETH per month ($20-40k at the corresponding ETH prices of  $3-4k). However, Orchestrators do not have a negative PnL, since a larger share of their income is actually from staking rewards³.

Notably, the accounting layer costs are in the range of 0.5-2 ETH per month since Livepeer’s transactions settle on Arbitrum. This is a >95% cost saving compared to prior to the Arbitrum migration in Q1 2022. Additionally, as of today, transactions on Livepeer grew 2-3x vs. prior to the Arbitrum migration. In relative terms, the accounting layer now accounts for ~5% of the costs, in contrast to being a major cost driver prior to the migration (~80% of costs).

Recently, the algorithm determining how work is distributed was adjusted, putting more weight on the price per pixel offered by the orchestrator. This puts downward pressure on transcoding prices and may consequently help fostering demand (see here), yet discussions in the forum show that price levels need to decrease even further. On the other hand, the recently launched AI-subnets might help adding further monetization avenues for the network.

A potential scenario in the estimation spreadsheet is a 3x in demand for minutes transcoded would only increase the overall cost by 20%. Notably, bandwidth is the main driver of the cost increase.

If we assume similar price levels (with 1ETH at $3k), this should suffice to bring the network into the breakeven region. However, should transcoding prices drop by 50%, the network-level fee income would be around $45k/month, thus below the lower end of cost estimates. With new use cases like AI-video generation in sight (and hence increased opportunities for monetization), it remains to be seen how cost and income dynamics on the Livepeer Network will change.

POKT Network

At its core, the POKT Network provides decentralized remote procedure call (RPC) endpoints. Recently, POKT Network announced its expansion into further use cases on AI model inference.

The framework applied step by step is provided here. Most cost estimates have been based on this survey conducted with node operators in summer 2023 and follow-up interviews with those and gateway operators.

Based on the ~15k nodes providing RPC endpoints and four  gateway operators providing access to those, we estimate that the POKT Network currently costs about 200k/$ per month (+/- 80k $) to serve about 500 M relays per day. The largest part currently is the service nodes (~75% of costs).

Since we can access historical data for the number of nodes active in the network and have different data points for cost components over time we can lay the network cost estimation on a timeline, showing three points when larger cost reductions have been tackled:

1. Node consolidation after token rewards (and esp. dollar based token rewards) have been reduced with the transition into the bear market mid 2022

2. Network-wide rollout of improvements like Geomesh and LeanPOKT significantly reducing operating costs as well as individual improvements on the setup by node operators

3. Decentralization of the gateway role reducing bandwidth costs by adding simpler gateways setups

Since our cost framework ties the cost estimates to the network capacity and demand, we can assess how the cost structure changes. For instance, if demand increases from currently 500M to e.g. 2.5 B relays per day, then the gateways make up 60% of a total cost base of ~400k $/month (vs. 200k currently). Note that this is a 2x in costs vs. a 5x in demand. This is due to the fact that service nodes were able to improve their setups, so increasing demand can be served at basically the same cost base. If we further assume that the share of new gateways operating at a lower cost base on the overall number of relays served increases further (currently 30%) to e.g. 50% then the overall network costs are 300k $/month.

With the decentralization of gateways, gateway operators can define their price points individually. If we assume an average price of $4 per million requests for the scenario the POKT Network as a whole would earn 300k $/month, so pretty much break even⁴.

Dfinity/ICP

Dfinity / the internet computer protocol (ICP) is designed as a “blockchain of blockchains”, providing compute resources for the execution of smart contracts (called canisters), organized in subnets (see details). The backbone are the node machines which provide the storage, compute and bandwidth to replicate all canisters, state and computation of their subnet.

The framework applied step by step is provided here. Most cost estimates have been based on data available in docs and forum posts.

ICP is one of the few networks, where fiat-based costs are baked into the token reward mechanism, which makes the cost assessment easier. There are currently about 1400 node machines run by ~85 operators. We don’t have data points on the economies of scale for the larger operators, hence our estimated range overall is quite wide: about 400k to 900k $ in monthly cost to operate the ICP network, average about 600k $.

Whilst a proper income assessment is worth a separate article, we estimate the current monthly revenue to ~25k $. This seems low in contrast to the estimated cost, yet is caused by low utilization: With just 559 node machines active we estimate the current demand (measured in cycle burn rate) is about 2% of total capacity. This means the network could take e.g. a 25x in demand and still not increase the current cost base. A forum post actually estimates a 15-25x in demand over the next two years which then (all else equal) would lead to a state where the ICP earns those monthly costs.

DIMO

DIMO is a decentralized network that empowers drivers to manage their vehicle data. Simultaneously, DIMO enables businesses and developers to build innovative mobility-related applications (to then monetize on that). The data measurement happens via special devices (Autopi, Macaron) or via App. While the above DePIN examples were digital resource networks, DIMO is the first example  of a physical resource network included in this analysis.

The framework applied step by step is provided here. Most cost estimates have been based on (device) price information online, dune data and forum posts.

For the accounting layer, we assumed half of the average $0.6-$1.5 in gas spent per connected car in Q1 2024 to be attributable to the operations of DIMO. For the gateway, we assumed about $4k per month in hardware and about $11k in labor cost per month related to the operations outlined above. All in all, this adds up to about $180k in monthly expenditures as shown in the table below. The majority of costs are related to bandwidth and other costs, which are ~⅓ related to the settlement costs on Polygon and ⅔  the assumed shares of monthly costs for the smartcar integration.

We do not have indications on the actual income of the network, but some estimation using global market for car data and related revenue from car data shows that this could be around $150-$185 per car currently growing to $500-$600 in 2030⁵. Should DIMO be able to monetize 10-15% of that, the resulting income would range from $110k-$180k per month and thus cover the operating costs.

Yet, data monetization itself appears to be less of the actual protocol goal; instead, DIMO focuses on providing infrastructure to enable apps on top of the network, as reflected in the latest discussions around DIMO nodes and token upgrades. The discussed changes will likely impact the cost structure above.

Footnotes:

1. This probably needs to be revisited when there are more protocols that actively build upon aggregating resources from various other networks. Also, this is not to say we should ignore sharing economy benefits: most DePINs are well positioned to realize those in some form, hence we do account for those in the overall cost estimation (typically affecting all node operators equally). Simple cases occur when the hardware utilized to contribute to the network is widely available, like on smartphones or laptops. If we can assume those devices are not purchased with the network support in mind, then hardware cost can be considered zero. Similar logic is applied for labor cost and cost of bandwidth, power and other OPEX.

2. Some good starting points are depinscan, depinhub, depin ninja and wholovesburritos

3. Orchestrators have about 4-5% of LPT self bound currently and take on average 12% on the remaining delegated staging rewards (see appendix here). Based on the average 2.5M $ worth of staking rewards paid per month this amounts to about 380-400k $ in staking rewards. Since Orchestrators also share on average 40% of transcoding income with their delegates the income from staking rewards is over 95%.

4. Which o/c does not mean all roles and all node operators are break even.

5. According to MarketsAndMarkets study the estimated revenue from car-data is around 1600$ per car by 2035 and 120$ per 2023. This means a CAGR of 24%, so by 2024 we get ~150$ and by 2025 185$ per car.

Ackee, Amegakure, Biomes, Briq, Cartridge, Cloudlines, Dope Wars, Eleven Yellow, Equilibrium, Ethereum Foundation, Etherplay, Footium, ggchronicles, ggQuest, Grugslair, Guildly, IDTheory, Inco, Interface, Jaw Drop Games, Lightshift, Magic Block, Mask, Mithraeum, Moving Castles, Nerd Monkey Games, Nouns Watch, Parallel, PixeLAW, Playmint, Polysensus, Primodium, ProofOfPlay, Realms, Rockaway, Rune Lab, Seismic, Semaeopus, Tonk, Topology, Trifle Labs, Voya Games, WASD, World Explorers, Yeomen, Zero Edge, zkorp, 1kx

In attendance, there were builders of onchain games, cryptographers, cryptonatives and many traditional game developers who had never built anything with crypto at all.

The goal of the event was to explore the theme of composability with the existing landscape of fully onchain games that were ready to launch later this year.

Since Dark Forest's last official round two years ago, composability as a property of fully onchain games has mostly been backstage a theoretical point of discussion while many of the next generation of games were being built. But now with a new wave of games near ready to ship, we at 1kx, felt it was salient to explore this once again.

Day 1

The event kicked off a keynote from Ludens, Co-founder of Lattice.

Next, we had 13 teams demo their games (links to their demos below):

Cursed Machine, Video Demo
Anybody Problem, Video Demo
Dopewars, Video Demo
Biomes, Video Demo
PixeLAW, Video Demo
DownStream, Video Demo
Primodium, Video Demo
Network States, Video Demo
Mithraeum, Video Demo
Pirate Nation, Video Demo
Loot Survivor, Video Demo
SkyStrife, Video Demo
Strategems, Video Demo

After, we had Awtnmy from Archetypal present a keynot workshop on modding.

Builders were let go to form their own teams and build for the next 2 days!

Day 2

Most were busy building during this period!

The venue was packed throughout the day ~

We had a keynote presentation by guiltygyoza from Topology:

Day 3

We kicked off the hack submission presentations and had 22 teams submit!

Links to video presentations below:

Incognitos, Treasure Hunt, Combine Harvesters, House B, Interlopers, Invisible Hand, Lebron-arama, AWBeat, Robots, cLAW, Team Geoff, Scout, Yakuza Revenge, Economy General Horse Glue no23, Amegakure, Skystrife MetaQuests, Deep Deal, Grugs Lair & Friends, Downstream Utopia, Hyperfarm, paco tobacco bytes, Fokax

You can find a great summary thread of the mods here by David Amor from Playmint.

Closing reflections

The fruits of labor from building during the last two years are starting to bear.

We are proud to contribute to the Autonomous Worlds movement and have never been more excited about the design space. A special thank you to Rich and Raquel from Archetypal for bridging the existing world of game development to their own bubble, and to Proof of Jake from WE for helping out as our stage volunteer!

If you want to keep up to date with the community here, follow:

Since the fund’s inception in 2018, we at 1kx have invested in over 100 projects under the thesis that token networks, powered by software and governed by communities, have enormous potential to impact the world economy and society at large. We believe the technologies that enable this will be built under three main principles:

• Self-custody. Users should have full ownership and control of their digital assets and be able to use them in a secure and convenient way.

  This unlocks digital property rights for the first time ever.

• Open Innovation. Technological primitives should be permissionless and composable to foster a resilient ecosystem of independent products and services built on top.

• Community Ownership. Builders, contributors, and users should own and control the protocols and infrastructure that empower our everyday lives. This aligns the incentive of network participants in the long term, tokens being a mechanism for protocols to capture and distribute the value it creates to the stakeholders and activities that contribute to its success.

Over the past five years, Safe has laid a robust foundation for the Web3 ecosystem across all three pillars. First, it offered a self-custodial smart account, which enabled onchain organizations to store and govern their treasuries securely. Then it opened its core infrastructure to allow independent teams to build in a modular, permissionless way. Thirdly, it launched SafeDAO and SAFE token, designed to help the project achieve technical, economical, and governance decentralization for critical components in the Safe ecosystem.

Why we Invested in Safe

Safe was born under the classic startup adage “build something you yourself want” when the Gnosis team developed an in-house multisig smart contract wallet, “Gnosis Safe,” to store its funds. The multisig product was quickly adopted by other DAOs and communities looking for simple ways to manage their onchain assets. Out of the proliferating suite of DAO tools at the time, Safe was among the first to find product market fit.

By the time we led Safe’s round in 2022, it was already the de facto asset management solution for DAOs, enterprises, and power users alike. Safe stored over 40B worth of value at the time, with a budding ecosystem of wallets (Metamask Institutional), treasury management (Coinshift, Parcel), DAO tools (Onchainden), NFT portfolio management (Castle), and other products building on its core smart account.

As long-time users, we already saw Safe as critical infrastructure for the space, but we were even more excited about what was to come. Safe’s value proposition is much more than a standalone multisig wallet. It was designed to be a fully secure, programmable platform for ownership. Over the past two years, we have worked with the team to build out this vision and continue to hit its strides through Safe{Wallet} and Safe{Core}.

Empowering Programmable Ownership

Safe{Wallet}, the flagship self-custody wallet built on Safe{Core} Smart Account contracts, is now among the most Lindy products in web3 thanks to its unbeaten security track record, the team’s technical rigor, and its community stewardship.

Safe’s adoption goes from strength to strength across market cycles, further catalyzed by the unraveling of centralized service providers found to be reckless with users’ funds. In the week following FTX's revelation, Safe saw a $800m+ net inflow as users fled to self-custodial solutions, further signaling the market’s confidence in the product and expanding the user base for Safe’s ecosystem of products and services.

Today, Safe has:

• $100b+ total value secured across 15+ networks

• $1b+ in monthly transfer volume for ETH alone

• 8m Safe accounts created (+290% YoY)

• 42m Safe transactions (+350% YoY)

• 200+ projects building on Safe accounts and tooling

Safe is now being used to facilitate payments via traditional payment rails through debit card products like Gnosis Pay and BasedApp. For the first time, consumers can use assets in their Safe to make purchases from any merchants that accept Visa. Safe{Core} Smart Accounts also power the self-custodial ownership stack of Worldcoin, with millions of unique users worldwide.

With battle-tested smart contracts at its core, Safe{Wallet} is composable with other protocols and applications. Its App Store currently features over 100 applications across DeFi, NFTs, bridging, and governance that users can interact with right from the wallet interface. More importantly, Safe’s architecture facilitates feature extensions through third-party modules. This gives users optionality in the service providers they want to work with and turns Safe itself into a platform for downstream products and services.

One example is the Recovery module, where upon installation the user can designate a Recoverer and customize recovery conditions. Safe’s recently launched Recovery Hub makes use of this module to provide a curated marketplace of crypto recovery options, starting with Sygnum and Coincover, which caters to individuals, DAOs, and institutions of varying preferences for recovery solutions and providers.

Other examples of modules include session keys, account automation, and spending limits. The Zodiac team has thus far been responsible for the most prolific set of Safe modules. But in the past year we have seen dedicated module infrastructure projects such as Rhinestone* and ZenGuard emerge to accelerate the development of modules via developer tools and module registries to make smart accounts more powerful than ever before.

Making Web3 Accessible to All

As the first and leading smart account provider, Safe has been at the forefront of the account abstraction movement, which aims to establish smart accounts as the default means for interacting with web3. The wallet ecosystem had a pivotal year in 2023 with the mainnet deployment of the ERC-4337 EntryPoint contract and the proliferation of embedded wallets that enabled users to onboard onto dapps using familiar login flows.

For developers and builders, Safe{Core} is an essential tool to integrate Safe smart accounts into any platform. This includes an Account Abstraction SDK designed to be agnostic to paymaster service for gas abstraction, authentication providers, and transaction relaying infrastructure making Safe accounts compatible with almost any developer toolstack. The Safe{Core} SDK continues to lead market share as builders look to enhance their dapp usability with AA features.

Safe's commitment to the highest security standards and unparalleled track record enables developers to launch production-ready apps, and users can interact with Safe accounts knowing they have stood the test of time. Account abstraction is merely the first step in unlocking the full potential of web3. Safe’s plug-and-play and use-case agnostic primitives, battle-tested infrastructure, and vibrant community of builders uniquely position it to tackle even more ambitious goals in the coming years.

As long term users and members of the Safe ecosystem, we look forward to seeing Safe’s primitives proliferate across new user bases, networks, and markets as smart accounts become the primary means to interact with web3.

If you are building on Safe, please reach out!

This article is for general information purposes only and should not be construed as or relied upon in any manner as investment, financial, legal, regulatory, tax, accounting, or similar advice. Under no circumstances should any material at the site be used or be construed as an offer soliciting the purchase or sale of any security, future, or other financial product or instrument. Views expressed in posts are those of the individual 1kx personnel quoted therein and are not the views of 1kx and are subject to change. The posts are not directed to any investors or potential investors, and do not constitute an offer to sell or a solicitation of an offer to buy any securities, and may not be used or relied upon in evaluating the merits of any investment. All information contained herein should be independently verified and confirmed. 1kx does not accept any liability for any loss or damage whatsoever caused in reliance upon such information. Certain information has been obtained from third-party sources. While taken from sources believed to be reliable, 1kx has not independently verified such information and makes no representations about the enduring accuracy or completeness of any information provided or its appropriateness for a given situation. 1kx may hold positions in certain projects or assets discussed in this article.

In this article, we'll cover:

• The origin of memecoins: proof-of-work meme blockchains

• How memecoins have evolved through cycles: ICOs, tokens, DeFi summer, Solana

• How NFTs have influenced the memecoin landscape

• Recent developments and emerging trends

• Potential risks and opportunities

What is a meme?

A meme is a unit that carries cultural ideas and symbols and spreads from one mind to another.

 Like genes, memes vary in their ability to spread; those that resonate live on, while less impactful ones are quickly forgotten.

The internet brought upon the idea of ‘Internet memes’, which allow for faster transmission of memes and cultural ideas, usually in the form of images, videos, gifs, and jokes.  One study likened the spread of internet memes to diseases: “Memes propagate in a viral pattern, "infecting" individuals in a pattern reminiscent of the SIR model for spread of disease”.

"Memecoins" are cryptocurrencies that derive their value solely from the memes with which they are associated, essentially bringing financial value to the concept of memes.

With the advent of memecoins, cultural ideas and symbols, and their spread, can be traded and speculated upon.  Their value is based on the meme's relevance and ability to capture mindshare, creating a new type of market where cultural resonance is quantifiable and financially valued.

A brief history of memecoins

Below is a brief look at crypto cycles and the memecoins that were created within them:

Proof-of-work memecoins

Proof-of-work memecoins were primarily designed for miners to allocate their resources to new coins for mining and selling. Many of these memecoins made their debut on Bitcointalk's alternative cryptocurrency forums. Although a significant number failed to reach exchanges, those that succeeded were traded on platforms like Cryptsy and BTC-E, centralized services that are now closed. Each memecoin was usually differentiated by name and branding, hashing algorithm, block time, and supply, all of which contributed to their overall narrative or “meme”.

The first wave of coins after Bitcoin were ‘memecoins’ in the sense that they provided little value beyond a new idea. A few examples include:

These coins, aside from Litecoin, are all dead (low volume and market cap, no exchange support, easy to 51% attack).  This is likely due to several reasons: the meme not sticking (lacked cultural longevity) and issues accessing them (each memecoin was an entire blockchain).

Litecoin’s survival is likely due to the sheer pull of Bitcoin’s memetic value (“digital gold”), it being significantly earlier than the others, and continued support from modern exchanges.

Dogecoin: the first memecoin

The original doge meme (“much wow”), began spreading on 4chan and Reddit in the summer of 2013.  Capitalizing on this cultural trend, Dogecoin was released on Dec 8, 2013, on Bitcointalk by Jackson Palmer and Billy Markus. It was the first cryptocurrency to be based on an Internet meme.

Dogecoin's success brought upon a new ‘class’ of coins, that were irreverent, humorous, ironic, used celebrities (Kanye West, Max Keiser), animals (Pandacoin), or tried to capture a specific community.  These were all proof-of-work coins launched in Bitcointalk’s “Alternative Cryptocurrency” sub-forum.  “Specs” became less important and the “meme” became more important.  Some examples:

ICO boom & Ethereum’s rise

The rise of Ethereum brought upon a wave of innovation which enabled new use cases, a better user experience, and new users.

Some specific improvements included:

• Easier to launch coins (ERC20 standard)

• Brought in a new userbase (non-miners)

• Creators of coins could make more money (unlike zero-premine PoW coins, ERC20s were sold via ICOs)

• Introduced interoperability / one ecosystem / one wallet (via ERC20)

The ICO era brought in a wave of more ‘serious’ projects: theDAO, Filecoin, Tezos, EOS, Cardano, Tron, and Bancor, which attempted to have some sort of utility or purpose beyond memetic value.

There were a smaller number of memecoins, which were not particularly notable but still garnered some attention.

One example was the Useless Ethereum Token, released in June 2017, which poked fun at the idea of ICOs, and raised 310 ETH.

Dentacoin, although originally meant to be a ‘cryptocurrency for Dentists’, was treated as a memecoin and peaked at a $2B market cap in Jan 2018.

HAYCOIN, the first ERC20 deployed to Uni v1, was also a memecoin created in this era (2018).  Hayden Adams, founder of Uniswap, created the token as a test for the Uniswap protocol.  It did not have much reception or volume at the time, but was resurrected in 2023 for its historical significance.

Collectible memes and early NFTs

Outside of crypto, a subset of Pepe the Frog memes were referred to as ‘Rare Pepes’, memes that were not posted publicly, and if they were, had the watermark ‘RARE PEPE DO NOT SAVE’.

Between 2016 - 2018, a group of counterparty developers (a smart contract protocol built on top of Bitcoin) and Pepe-meme enthusiasts created the Rare Pepe Wallet, Pepe Cash, and curated ‘Rare Pepe memes’ to be traded on the Counterparty Protocol.

Often considered the second NFT collection ever, Rare Pepe continue to hold value and some have sold for over $500,000.

Following the release of CryptoPunks, MoonCats, and CryptoKitties, NFTs began entering the Ethereum zeitgeist: non-fungible tokens that could point to pictures and other media. EtherRocks, launched in 2017 on Reddit as a joke, was a collection of 100 multi-colored clipart rocks.  The collection had very little reception at the time (only 30 were minted), but was later resurrected and had a frenzy of buying activity, reaching a floor of 305 Ether in Aug 2021 ($1m USD at the time).

Another example of collector item memecoins is Unisocks (SOCKS), launched by Hayden Adams on May 9, 2019.  500 physical socks were created, and 1 SOCK (an ERC20) can be used to redeem them.  At $53,000 per sock at the time of writing, they may be the most expensive socks in the world.

DeFi Summer

In June 2020, Compound Finance pioneered a new token distribution method: ‘liquidity mining’ or ‘yield farming’.  Users would lock their assets to provide liquidity and get rewarded with tokens.

This new primitive kicked off ‘DeFi summer’, and culminated in ‘food coin’ yield farms that offered 10,000% APY (in memecoins) in exchange for having your tokens locked up in a Yam or a Pickle contract.

Memestocks and dog-coins

Stimulus checks, rate cuts, cheap money, and COVID lockdowns brought about a highly risk-on environment throughout 2021.

In early 2021, retail traders gathered on Reddit and began proliferating the 'Gamestock' meme, posting images/videos and fueling a rapid increase in the stock's price.  Many average users were able to participate thanks to the UX of Robinhood (mobile app, fee-free).

‘GME’-mania led many to begin speculating on other assets, particularly those that were also available on Robinhood.  DOGE was listed on Robinhood in 2018, and was priced at .008 cents in late January, a very appealing “price” for retail users.  In early February 2021, Elon Musk began posting numerous tweets proliferating the Doge meme.  DOGE peaked in May 2021, at a market cap of $90B.

DOGE’s popularity led to many cryptonative memecoins, including Shiba Inu, Floki, and Safemoon, all of which reached very high valuations within months.

The NFT boom: “memecoins with pictures”

With the standardization of ERC721 and generalized marketplaces like OpenSea, NFTs created a new ‘class’ of cryptoasset: unique, visual representations of an overarching “culture” or “meme”.

Some of the most recognizable NFTs include: CryptoPunks, Bored Apes, Squiggles, and Pudgy Penguins.  NFTs were sported as profile pictures on platforms like Twitter and Discord, spreading their virality by individuals who donned them as their identity.  These PFPs signaled prestige and “membership within a cultural club”.  Many collections made their holders rich, but the holders couldn’t sell their NFT without ‘leaving the community’.  In order to reward their loyal holders, some NFT projects released a ‘memecoin’ (ERC20) component to their community, as a form of liquidity, ‘utility’, and cultural ‘currency’.

Some examples of NFT projects in this era (and if applicable, their memecoin) include:

Recent times (2023+)

As crypto recovers from a bear market, new memes, cultures, ideas, and ecosystems have been emerging.  Memecoins have been one of the few categories that have continued to find traction (volume, market cap appreciation, social interest), and most recently have had surging interest.

Some recent narratives include:

Some common patterns

Every cycle has some form of memecoins: Memecoins manifest in different ways depending on the underlying technology (PoW coins vs ERC20 vs NFTs), they are one of the earliest ‘applications’ of new technology or form factors.

Although memecoins come in all different types of mediums, they accrue value in the same way: they all require attention, narrative, and hype to survive and spread. While new mediums always find initial early excitement, long-term retention of value is only possible via sustained attention. Just being an NFT is not enough; just being an Ordinal is not enough alone. Attention is highly cyclical, and after an initial medium’s first hype cycle, it requires more fundamental attention drivers.

The meme came first, the coin came after: For the most successful memecoins, such as doge, pepe, and dogwifhat, the Internet meme came first and the coins capitalized on existing awareness and spread.

Memes originating from crypto are just beginning:** cryptonatives have started to create memes that successfully extend beyond web3 circles. A prime example of this trend is the emergence of cryptonative intellectual property, especially NFT projects like Pudgy Penguins.

Low price as a meme: since the earliest days of altcoins, users have enjoyed speculating on tokens with a low ‘price’ (due to high supply).  There is a psychological allure that should a token with a seemingly insignificant price reach $1, it could transform holders into overnight millionaires.  The price itself serves as a meme.

Strong Community + Marketing: memecoins need a strong community, founder, or ‘spokesperson’ that can create content, promote the brand, and spread the ‘meme’.

Organic launches to dedicated teams:  The first wave of memecoins were organic, often fair-launched with no insider or team allocation.  These organic launches have many benefits, but also carry risks such as rugpulls and stolen funds.  An emerging solution is memecoins with dedicated teams incentivised to bootstrap attention around them, such as PFP projects.

Strong imagery, derivative imagery, and slogans spread the meme: Imagery is the primary way memes propagate on social networks.  Often, it begins with one keynote symbol, which manifests itself in many different ways.

The opportunity

At a combined market cap of over $60 billion, and daily volume exceeding $13 billion, memecoins command substantial financial value.

Since a meme’s only function is to spread into other people’s minds, catching the next ‘meme’ early can be a lucrative opportunity.  Creators and investors ‘work for the meme’ by spreading its reach and get rewarded in the process for being early believers.

Whether it’s a lottery ticket to get rich, following a particular influencer’s picks, or speculating on societal trends and ideas, memecoins have continued to grow explosively since the beginnings of cryptocurrency.

Criticism and risks

Despite the opportunity, memecoins are not without risks.  Many memecoins tend to attract users trying to acquire instant wealth with little effort, treating it as a lottery ticket or form of gambling.

Another common trend are ‘rug pulls’ and ‘pump-and-dump’ schemes, which occur almost daily on decentralized networks through memecoins. According to a recent report by CipherTrace, a blockchain analytics firm, rug pulls accounted for 99% of all crypto frauds in 2023, totaling $2.1 billion in losses.  It is important to check some key features of a memecoin, such as the status of LP tokens (are they burnt or held centrally?), team allocation, transfer taxes, and if the contract is renounced.

There is also a lack of regulatory clarity on memecoins.  The most notable regulation on memecoins is when Thailand 's SEC banned tokens that have "no clear objective or substance" and whose prices are swayed by social media trends and influencers in June 2021.

Another major issue is that memes may not ‘survive’, purely due to lack of interest, attention, and mindshare, rather than any malicious act.  These investment losses can create a hostile or apathetic community of holders.

Disclaimer

This article is for general information purposes only and should not be construed as or relied upon in any manner as investment, financial, legal, regulatory, tax, accounting, or similar advice. Under no circumstances should any material at the site be used or be construed as an offer soliciting the purchase or sale of any security, future, or other financial product or instrument. Views expressed in posts are those of the individual 1kx personnel quoted therein and are not the views of 1kx and are subject to change. The posts are not directed to any investors or potential investors, and do not constitute an offer to sell or a solicitation of an offer to buy any securities, and may not be used or relied upon in evaluating the merits of any investment. All information contained herein should be independently verified and confirmed. 1kx does not accept any liability for any loss or damage whatsoever caused in reliance upon such information. Certain information has been obtained from third-party sources. While taken from sources believed to be reliable, 1kx has not independently verified such information and makes no representations about the enduring accuracy or completeness of any information provided or its appropriateness for a given situation. 1kx may hold positions in certain projects or assets discussed in this article.

While many have pronounced crypto’s incursion into the world of AI as unnecessary, we see it as a vital and synergistic relationship. The tightening of restrictions around the production and distribution of AI models has seen a remarkably fast-paced and anti-authoritarian open source community burgeon, competing with the likes of well-funded centralized alternatives and governments. Crypto is the best tool to-date for crowdfunding and governing open-source tools in direct contradiction to external pressures. This is already an ideal match-up, and that’s before you begin to consider the impacts of AI on authenticity, provenance, identity, and other areas which crypto can inherently remediate or improve upon.

There are various rabbit holes here to explore - each of which could warrant their own thesis. This piece attempts to cover a lot of ground, therefore it is a whirlwind overview of some of the emergent areas of crypto x AI thus far and in the foreseeable future.

creativity.exe

The first wave of recent interest in AI was in the realm of generative tools for creativity. Generative AI reduces the need for technical skills such as coding or advanced software proficiency, allowing individuals with basic digital literacy to produce complex works while replicating professional output with minimal cost.

There are a myriad of impacts this is likely to produce. To name a few:

• Anyone can now be a creator, and multiplayer modes of creation will flourish as humans collaborate with these tools to create works together

• Niche communities are able to produce high quality work that was previously not commercially viable due to audience size

• There will be an influx of generative content which will far outpace human work, arguably leading to a potential revaluation of human content online

Below is a discussion of some of the creative mediums which are seeing a high amount of engagement with AI.

Art

‘AI art is not art’ is a rallying cry for those entrenched against the rise of AI tools. The release of generative models was swiftly met with backlash and protests, such as what was seen on ArtStation. Yet it has sparked some of the most interesting creative sub-verticals in web3.

AI art comes in many forms. The most well-known are popular generative models including DALL-E, Stable Diffusion, and Midjourney. There are also web3 contenders like ImgnAI that intend to provide a vibrant social experience around generative image creation enhanced by tokenomics - something that is quite needed to provide community moats around these generative models.

However, it goes much deeper than this - highly regarded AI artists in the space often engineer and finetune their models in a way that makes them more unique than easily prompted images. This might entail training new embeddings or using LoRAs to perfect a certain style, or creating your own model entirely.

Popular artists publishing AI art as NFTs using more sophisticated and personalized models include Claire Silver, Ivona Tau, Roope Rainisto, Pindar van Arman, Refik Anadol, Gene Kogan, and many more. These artists have explored using various marketplaces for distribution, with the most prominent being AI art dedicated marketplaces like Braindrops, Mirage Gallery, and FellowshipAI - along with platforms like Bright Moments which have dedicated live events to the art form.

Sub-verticals of AI art have also formed, such as post-photography and data art. Post-photography has been pushed primarily by the Fellowship.ai team, working with Roope Rainsto to bring more artists exploring this medium into the public eye. A lot of the post-photography aesthetic is dedicated to embracing the visual artifacts common with early generative tools. Post-photography began a crescendo on social media with the publication of Roope Rainisto’s Life in West America collection on Braindrops.

Regarding data art, Refik Anadol is a prominent artist here known for immersive installations that use data, algorithms, and artificial intelligence to create dynamic and interactive artworks. Some interesting examples of his work include Unsupervised, which transformed the metadata of MoMA’s collection into a work that continuously generates new forms in real-time, and Sense of Place, which used real-time environmental data such as wind, temperature, and humidity as well as invisible communication signals from Bluetooth, Wifi and LTE to inform the work.

Another interesting sub-vertical is an entirely new content medium uniquely enabled by crypto: onchain autonomous artists. The most well-known example of this is Botto, a community-governed generative artist that produces 350 art pieces weekly, grouped into "rounds" consisting of individual "fragments". The BottoDAO community votes on these fragments, using their aesthetic preferences to guide Botto's generative algorithm for future art creation, ensuring the artwork evolves under community influence over time. Weekly, a fragment is minted and auctioned on SuperRare, with proceeds returning to the community. After completing its Fragmentation and Paradox Periods, Botto is currently in its Rebellion Period, incorporating new technologies like Stable Diffusion 2.1 and Kandinsky 2.1, and exploring collaborations and select collections alongside its weekly rounds. Botto is one of the highest grossing artists on SuperRare and has even amassed its own collector DAO called CyborgDAO. Additional projects like v0 are also tackling the novel merger of tokenomics and AI art models, providing a venue for multiple artists to create their own onchain art engines governed by communities of holders.

In interviewing collectors of any form of AI art, the most common retort from crypto spheres is that the curation by the artist diminishes its interaction with the blockchain, unlike more classical generative art a la Art Blocks. Rather than randomness generated from chain-specific inputs, the outputs are selected by the artist themselves to be included in a collection after multiple permutations. Despite it being a digitally-native art creation process, it must be manually brought on-chain.

Fully onchain AI art is difficult due to constraints in the execution environment and the computational complexity of the image generating models being used. There are examples of lightweight outputs like Pindar van Arman’s byteGANs being stored onchain, but we anticipate offchain verification mechanisms as being the closest available form factor in the short term for more complex models. For example, Modulus Labs recently partnered with Polychain Monsters to create a zkML-verified GAN model for producing collectible pixel monsters. Using zk proofs, each generated NFT can be cryptographically verified as coming from the actual Polychain Monsters art model - a huge step in provenance for AI art.

Music

Outside of image-based art, there is a major movement occurring in music. Many will likely be familiar by now with the success of ghostwriter’s AI Drake hit. Within 2 days it racked up 20+ Million streams and was promptly taken down by UMG. This brief phenomenon brought awareness to the public sphere that the relationship between the artist and the work itself is fundamentally changing.

It is inevitable that generative music will outpace human-created music in a few years. Boomy - a generative music startup founded in late 2018 - saw its users create nearly 14% of the world's recorded music (~14m+ songs) in that short period of time. This is just a single platform mostly before the recent surge in interest by the general public.

Given generative content will outpace human-created work and artists will likely be separated from their work using voice models, artists will need authenticity verification. The best way for media to be published and verified as authentic would of course be via cryptographic primitives.

However, it’s important to note that this isn’t all bad for artists, especially ones willing to embrace the inevitable. Holly Herndon, an innovator in publicly available voice models, allowed her community (Holly+) to create and distribute works and audio using her voice. Holly’s thesis with the release was simple:

“While for the time being the perceived difference between a bootleg or official      voice model may be quite small, as our ability to produce more detailed and      convincing voice generation evolves, so too will the need for comprehensive, high      fidelity vocal training data, as well as the urgency of discerning provenance. For this      reason I believe that there will be demand for official, high fidelity, vocal models of      public figures, so why not experiment with my own?”

A DAO oversees the Holly+ voice model, with the ability to make decisions on creating and approving new works through voting. Token holders within the DAO are motivated to ensure that only high-quality works that enhance the voice's value are approved, to prevent devaluation from poor-quality art or negative connotations. The voice model will be used to produce a limited number of official artworks, and DAO token holders will receive ongoing profits from the resale of these works.

More recently, Grimes released elf.tech - a platform that allows artists to use a "GrimesAI voiceprint" in their original songs, requiring a 50% royalty share with Grimes upon her approval. Elf.Tech, powered by CreateSafe's AI, enables this, and a partnership with TuneCore facilitates professional distribution and ensures proper royalty management. Payments are handled via fiat or via automated onchain royalty splitting if the final form of the music is an onchain NFT. Hume, a web3 music studio focused on virtual artists, was one of the first to use the Grimes model to publish a collab between Grimes AI and their virtual artist angelbaby.

Fashion & Physicals

I’ve previously explored the concept of generative manufacturing of physical consumer goods and fashion products using creative coding algorithms and AI here: https://mirror.xyz/1kx.eth/oBuaEp5jgGbe2gCsa6Z-_mLAeMRUhsIdZsaScHQNXS0.

As a short recap,  generative AI and creative coding pave the way for a future of hyper-personalization in products and experiences, enabling the creation of unique designs, patterns, and art tailored to individual preferences. This technology can be applied across various domains, from fashion to home decor, and further enhanced by allowing users to fine-tune outputs to their liking. Newer fabrication tools often allow you to connect code to the machine directly to automate production of outputs, inherently solving a lot of the bottleneck of manufacturing personalized goods.

A sample of web3 projects exploring this space today include Deep Objects, RSTLSS, and Little Swag World. It should be noted as well that most digital fashion projects will likely explore generative creative tools and media, with Draup, Tribute Brand, and others talking at length about its use.

Community creation of model outputs, similar to Botto, is a fascinating idea that Deep Objects is exploring. Theyused a community curation engine to whittle down one million designs generated by their GAN AI model to a single community-selected piece. This final piece is now set to be 3D printed in a showcase of generative product creation. DeepObjects can easily extend this curation design to other physical goods as well.

RSTLSS had a drop with AI artist Claire Silver called Pixelgeist, in which each mint included the artwork, a digital garment featuring the artwork, a game-ready avatar with the garment, and the ability to purchase a corresponding physical piece. This unique blend of digiphysical fashion and AI outputs is one of the most interesting experiments combining gaming, fashion, and AI together. Claire Silver has also tackled fashion photography directly with her most recent collection corpo | real on Braindrops. For more on the topic of digital fashion, see my piece here: https://medium.com/1kxnetwork/augmenting-culture-the-emerging-field-of-digital-fashion-bead627c8dcd.

Little Swag World is a great example of using GAN models in the creative workflow, from design to physical object. Bosch, the artist behind the project, created the initial designs himself before running it through Stable Diffusion / Controlnet to generate unique, hyperrealistic builds. This technique resulted in a high degree of aesthetic consistency. The next step for the project is to take these generative models and combine it with ceramics to make physicals of the AI-enhanced set of NFTs.

All-in-all, we expect some fascinating crypto x AI projects to emerge here, from decentralized brands that curate generative products to fractionalizable NFTs of AI agent designers.

Entertainment

Generative entertainment is also beginning to develop more fully following the initial hype around Nothing, Forever, an interactive, generative animated sitcom based on Seinfeld running on a 24/7 Twitch stream. Interestingly, it showcases the power of the medium with the show’s narrative changing based on Twitch chat responses, and allowing donors to import their likeness into the show as a character.

The Simulation from Fable extended this research with SHOW-1, a model for prompting generative TV shows where writing, animation, direction, voicing, and editing is all performed for you via a prompt. They demoed this with South Park episodes initially, but it is easy to extend to any IP. I solely expect more permissionless forms of IP like what we see in web3 will experiment in-depth with this type of tooling for creating content.

Upstreet has recently begun experimenting with generative television as well, taking their AI agent model developed for their virtual world platform (more on this below) and allowing creators to add their own vrm avatars and prompt them into unique interactions and skits.

One other area of interest here is IP. Projects like Story Protocol are looking at facilitation of IP creation, distribution, and monetization using a decentralized IP registry. This is useful for creators because of the seamlessness relative to traditional IP licensing, however it becomes particularly unique in the era of generative AI. NFT IPs, memes, and other entertainment projects could license out and pay for creation of generative extensions, unlocking a huge amplification of the project via creators.

Are you a bot?

Deepfakes. Chatbots trained on influencers to interact with fans. Generative spam on social media. Soon, it will be crucial to verify who is a human online.

Web3 has put a lot of effort into preventing sybils - albeit it is still arguably unsolved. However, reputation systems, proof of personhood mechanism design, user passports, soul-bound NFTs, and entire token economies are all being developed to fix this.

Attested hardware, zkML, and Proof of Personhood

I’ve previously discussed at length the practical implications and potential use cases of zkML here: https://mirror.xyz/1kx.eth/q0s9RCH43JCDq8Z2w2Zo6S5SYcFt9ZQaRITzR4G7a_k.

At a high-level, there are multiple teams such as Modulus Labs, EZKL, Giza, and more focused on proving the inference of a model using zk. These efforts to use zk to verify a model’s outputs have long-ranging applications and will ultimately enable new experiments in DeFi, identity, art, and gaming to access these models in a trust-minimized way.

While there are a myriad of projects focusing on proof of personhood, one of the most interesting applications comes from Worldcoin. Worldcoin uses AI models to convert iris scans into short hashes that are easy to cross-check for similarity / collisions in the case of sybil attacks. Because every iris is unique, the model is able to determine the user is actually a human being. It uses a trusted hardware setup (the infamous orb) to ensure that the model only accepts inputs cryptographically signed by the camera.

Similarly, the zk microphone team have demoed the use of attested microphones to create and digitally sign audio content to verify the authenticity of a recording. The secret key is stored in a microphone’s secure enclave, which signs to guarantee the authenticity of the recorded audio. Since most audio recordings are manipulated or edited, SNARK-powered audio editing software could apply audio transformations while still proving that the audio provenance. Daniel Kang also worked with Anna Rose and Kobi Gurkan on a proof of concept of attested audio recording.

Forever Influencers

On the flipside of verification of personhood or human-created content is the possibility of embracing deepfakes. Similar to voice cloning models above, some influencers have elected to create chatbots to engage their audiences. A well-known example of this is Caryn Marjorie who launched an AI girlfriend product using her voice and trained on thousands of hours of her YouTube videos to perfectly capture her personality, mannerisms, and voice. Users pay $1 per minute to talk with her avatar in a private Telegram channel, sending and receiving voice messages with her likeness. In the first week of launch, the influencer made $72,000, and it was estimated she could earn $5 million each month based on her subscription growth.

While CarynAI is an example of an AI girlfriend product (more on this below), there is room for non-erotic modes as well. Imagine being able to play games with a perfectly-trained model of your favorite video game streamer, conversing in real time simulating the actual experience. Alternatively, influencers could use personified AI + avatars that can be licensed out for fashion shows or for use in publications.

˚✧₊⁎( ˘ω˘ )⁎⁺˳✧༚ Uwu-ral Networks are so kawaii        (ﾉ◕ヮ◕)ﾉ*:･ﾟ✧

79% of adults aged 18 to 24 report feeling lonely. 42% of those aged 18 to 34 report “always” feeling “left out”. 63% of men under 30 describe themselves as single, compared with 34% of women in the same age bracket. Only 21% of men said they received emotional support from a friend within the past week.

People are lonely. In an era where loneliness is a growing epidemic, particularly among young adults, the emergence of AI companionship offers a unique albeit slightly dystopian solution. AI companions are always available, judgment-free, and highly personalizable. They can act as therapists or as erotic outlets. They can be creative colleagues or lifestyle coaches. They are always on and waiting to speak to you about whatever you want.

The infrastructure to do this is all here: Finetune a model with a personality prompt outlining behavior, looks, traits, communication style, and more. Run the model’s outputs through a vocal model like elevenlabs. Generate selfies on request using an image generator model and the defined looks prompt. Generate an appropriate vrm avatar and place it in an interactable environment. You now have a perfect-for-you transmedia companion. Add crypto rails and you make your companion ownable, tradeable, rentable, and more.

Companion.tech

This setup above is totally possible to do in a DIY manner, but there also exist whole applications dedicated to this concept. Full-stack companion apps - Replika being the most well-known example - allows for instant messaging with a companion without any technical skills involved. These applications often operate on subscription models, where users pay to interact with their companion. Not only has this product proven very lucrative, it has also resulted in some astonishing displays of how much this trend taps into the human psyche: for example, one reddit post highlighted the individual’s 2000 day streak talking to his companion, and elsewhere we have seen marriage proposals, AR selfie creation, and more. One extremely surprising tidbit here: when erotic features were removed from the platform, the subreddit moderators had to pin the suicide hotline for its distraught community.

Character-based platforms have also emerged, providing a means for multiple characters to be available to the user, again via a subscription in most cases. While there are many existing characters available across platforms like Character.ai and Chub.ai, the real novelty is in crafting your own character or scenario via personality prompting + feedback training.

Some attempts have taken place in web3 to make these companionship experiences, such as Belonging Hearts, MoeMate, and Imgnai.

Belonging Hearts had users chat with their character until they could get her phone number in order to get whitelisted for the NFT mint - a novel minting experience. Once received, the NFTs provided tokengated access to a chatbot experience with the character, which included erotic roleplay and generative selfies of the model. While it is tbd the future direction of the product, there has been much discussion of using tokenomics as a mechanism where players gift items or tokens to the chatbot to impact her mood and relationship level.

MoeMate, created by the team behind Webaverse, created an application for desktop and browser where users can easily import vrm models that they could then ascribe a personality to and interact with. The desktop version is reminiscent of having the old-school paperclip assistant, and it makes sense as a form factor given how often users want to ask AI assistants questions.

Finally, there is Imgnai, which in addition to being a high quality image generator model as mentioned above, is also tackling personification of the Nai character with a fully integrated chatbot experience.

Ultimately, the potential for tokenomics abound in the companionship domain: Tokenizing APIs, tradeable personality prompts (see below), onchain in-game currencies, agent-to-agent payments, tradeable cosmetics, gacha mechanisms for characters, and tokengating being just some of the potential future explorations.

Personality Marketplaces

Interestingly, the creation of companionship apps have also led to a subsequent rise in standardization of personality prompts along with platforms that facilitate the exchange of these personality primitives. A possible next step for the space is financialization of high-quality prompts and scenarios - for example, if an uncensored opensource LLM could read metadata from an NFT that included the standardized personality, then NFTs of personality could be exchanged with royalties back to the prompt creator.

This also leads to another open question: with many of the top models being restricted around NSFW content, viable open source models must be created - a good opportunity for token-based crowdfunding and governance.

—

For more on some of the ideas in this section, check out this piece on virtual beings I wrote: https://medium.com/1kxnetwork/virtual-beings-51606c041acf.

Augmented Governance

The history of DAO governance is effectively a speedrun of years of human collaboration. Ultimately, what we have found is that it is extremely difficult to organize resources effectively, minimize governance bloat, eliminate free-riders, and discover inefficiencies or bottlenecks in soft power.

Experimentation in using AI as an augmentation layer for DAOs are just beginning, but their potential impact is profound. The most common form is using highly trained LLMs to help direct labor capital within the DAO to effective causes, identify issues with proposals, and generally open up broader participation in both contributions and voting. There are also simpler tools like AwesomeQA which streamline DAO efficiency through search and automated response. Ultimately, we expect the ‘autonomous’ in DAO to become much more significant over time.

Autonomous Councils & Voting Agents

Upstreet has taken multi-agent systems like those popularized by AutoGPT and applied them to their governance process as an early experiment. Each agent is defined by a sub-group of the DAO: artist, developer, BD strategist, PR, community manager, etc. These agents are then tasked with analyzing a proposal from a contributor and debate amongst each other the pros and cons. The agents then score its impact relative to their division and this score is aggregated. Human contributors can then evaluate their discussion and scoring before voting on an outcome, essentially offering diverse peer-review as a service.

This is particularly interesting since this process can surface aspects of a proposal that a human might miss, or enable humans to debate with an AI agent about its subsequent effects.

Autonolas uses an automation system to bring AI agent decisionmaking to smart contracts in an optimistic approach versus more computation heavy zk approaches. They released a novel experiment on governance which enabled any tokenholder to delegate their voting power within a DAO to an AI agent called the governatooor. Delegators could elect for the agent to vote with a moral affinity: good or evil. The service analyzes DAO forums for active votes and then executes the vote according to the user preference. While this is more of an experiment, it does highlight the possibility of AI agents trained with a user's preferences to actively participate and vote on their behalf.

Advanced Coordination Systems

MakerDAO has also talked at length of aiming for autonomous governance decisions with minimal human input. Their contribution to date has been an outline of Atlas - a hub containing all data relevant to Maker Governance that adapts in real-time. These units of data are organized in a Document Tree, providing context to prevent misunderstandings or misinterpretations. The Atlas will be in a JSON format and standardized for facilitating manipulation by AI and coding tools.

The Atlas will be used by various Governance AI Tools (GAITs) that participate in governance by automating interactions and prioritizing tasks for participants. Sample use cases include:

• Project Bidding: GAITs can streamline the process for Ecosystem Actors to bid on projects by handling paperwork and ensuring proposals fit strategic goals.

• Monitoring Rule Breaches: GAITs can assist in monitoring deliverables and rule adherence, flagging potential issues for human review.

• Professional Advice Integration: GAITs can turn professional advice into formatted proposals, bridging the gap between governance and professional expertise.

• Data Integration: GAITs allow for the easy integration of new data and experiences, helping the DAO to learn and adapt without repeating mistakes.

• Language Inclusivity: GAITs can act as translators to make governance accessible in multiple languages, fostering a diverse and inclusive environment.

• SubDAOs: The Atlas and GAITs can be applied to SubDAOs, allowing for experimentation and rapid development with the ability to learn from failures.

🧠 <> 🎮

One area of crypto x AI I am particularly excited about is in gaming. There are a multiple of novel game loops to be explored, such as procedural content games, generative virtual worlds, LLM-based narratives, cooperative games against or in conjunction with AI agents, and more.

While there are plenty of great examples in web2 of new game loops, here we will focus on web3-specific examples. However, it is worth mentioning the academic piece that woke up many to the possibility of multi-AI agent gaming environments. Researchers from Stanford and Google demonstrated this potential by applying LLMs to agents in a sandbox game environment. The LLM-powered agents exhibited impressive behavior, including independently spreading party invitations, forming friendships, asking each other out on dates, and coordinating to attend a party on time, all based on a single user-specified suggestion. The approach utilized an architecture that extended the LLM to store and synthesize memories into higher-level reflections, enabling dynamic behavior planning for the agents.

This research is a foundation for what is becoming the most explored - but still ripe for experimentation - game loop in web3 thus far. The core idea is how do we use AI agents in simulated environments with a high degree of autonomy or character and craft interesting and fun games around them.

Parallel Colony from the Parallel TCG team explores this concept by having AI agents gather resources and tokens in a game for the player. Using the ERC-6551 tokenbound standard, AI agents are NFT wallets that can transact in a game on behalf of a user. AI agents can create, mint, and store new in-game items, and also have personalities as defined by a fine-tuned LLM created by the team allowing them to have non-standard behaviors and traits that affect their actions in-game.

Most fascinating of the AI agent-based games conceptually though is Upstreet. Upstreet is a virtual world project that has some insanely creative functionality, such as an AI agents SDK, procedural questing, browser + VR, drag-and-drop interoperability, and social features in an environment called ‘The Street’ which will see players build their own experiences and interact within. Alongside players will be AI agents that developers (and players via a user friendly marketplace) can deploy with personalities and goals that impact the game environment. Most interesting of all is their research and development into the AI Director: an AI agent that decides on a goal such as “Skydive from the tallest building” or “Start a new religion” that users and agents participate in as a challenge. The Director then determines the winner when the round ends, rewarding players and agents with prizes, tokens, and NFTs. This could lead to very interesting and complex agent-player interactions, and we are remarkably excited to see the outcome, especially as this can directly result in high-value 3D environment research and data for even more advanced models in the future - something that OpenAI also seems to be interested in with their acquisition of an open source Minecraft style game.

Generative tools for creating virtual worlds is another area that enhances games. For example, Today has players design their own virtual islands and care for AI NPC companions. What’s particularly unique here is their use of generative creative tools for facilitating development of UGC in-game. Since the game is primarily based around these user-created islands, it is important to give players without 3D game development or art skills an opportunity to develop assets seamlessly. Arguably, a lot of the doldrums around metaverse-style gameplay is that content is lacking, which can be remediated with use of generative tools in the near-term.

AI agents need to be trained, and training itself can be an interesting game loop for players to explore. AI Arena offers a novel take on training AI agents, having players play a Super Smash Bros style game to slowly teach an AI agent for competitions via imitation training. Because the AI agent needs no breaks, it can infinitely play competitive tournaments for prizes against an always active competitor pool while a player fine tunes its playstyle async. This turns training into a game and enhances it with tokenomics.

Large scale cooperative game play by humans against a strong AI player is something that could be done in the past, but it is taken to new heights with the merger of tokenomics. Leela vs. the World from Modulus Labs is an experiment in this gameplay form. For this experiment, Modulus took the Leela chess engine and proved its outputs via a zk circuit. Players would then put up capital to vote on a human move to play against the AI, providing a creative prediction market around the outcome of the game. While proving times for the model were of course high given the state of zk at the time, it opened up the possibility of large-scale cooperation, esports-style prediction markets, and governance mechanisms for playing against a verifiable and sophisticated AI player.

Finally, fully onchain games, or autonomous worlds, will also be enhanced by AI. The most notable piece on the topic was Large Lore Models, which looked at using an LLM protocol layer to create ongoing lore that interops across moddable and interlinked game environments - player actions affect multiple game environments simultaneously in autonomous worlds and thus should carry with them higher dimensional lore for facilitating storylines. This is well-suited to an abstracted LLM layer built across multiple onchain games.

Infrastructure

AI x Crypto infrastructure warrants an entire piece unto itself, but here I’ll briefly go through some of the ideas we are seeing take shape.

Distributed Computing

To understand the need for cryptoeconomic systems for compute, its important to understand the core problem. GPU capacity is extremely bottlenecked to date, with the best available hardware such as the H100 seeing long wait-times extending up to a year out. This comes at a time where startups are raising huge rounds to acquire hardware, governments are scrambling to create their own supply for national defense purposes, and even the most well-funded teams like OpenAI having to pause feature releases because of limited compute capacity.

Many teams focused on decentralized computing and DePIN see an opportunity here: bootstrap permissionless clusters to service demand while providing crypto-incentives and minimal margins that make the networks highly competitive on pricing with web2 counterparts while simultaneously providing a better reward profile for hardware suppliers.

Machine learning can be broadly categorized into four main compute workloads:

• Data pre-processing: Raw data is prepared and transformed into a usable format.

• Training: ML models are trained on a large dataset to learn the patterns and relationships within the data.

• Fine-tuning: ML models can be further optimized using a smaller dataset to improve the performance for a specific task.

• Inference: The trained and fine-tuned models are run to make predictions.

We have seen the rise of more generalized compute networks like Render and Akash turn toward servicing more specialized compute like AI/ML. Render for example has tapped into providers building on top of their network like io.net to more directly serve AI customers, while suppliers like Akash have begun bringing onboard suppliers with in-demand hardware and exemplifying the power of the network via directly training their own models - the first example of this is a Stable Diffusion fork trained on only copyright-free materials. Livepeer is also specifically targeting AI video compute given their large network of orchestrators already servicing video transcoding use cases.

In addition, networks are now forming that target AI compute specifically, recognizing that core challenges around orchestration and verification can be solved more directly by building the entire chain or model around AI. Gensyn is one of the more notable examples of this, building a substrate-based L1 designed for parallelization and verification. The protocol uses parallelization to split larger computational workloads into tasks and pushes them asynchronously to the network. To solve the verification problem, Gensyn uses probabilistic proof-of-learning, a graph-based pinpoint protocol, and an incentive system based on staking and slashing. Although the Gensyn network is not yet live, the team predicts that the hourly cost for a V100-equivalent GPU on their network will be approximately $0.40.

Alternative training models exist as well, such as federated learning which has seen a revival in web3 after realizing blockchains can more properly incentivize these models. In short, federated learning is an approach where multiple parties train a model independently with updates regularly batched and sent to a global model. There are a variety of examples of this in practice, such as Google’s keyboard text prediction algorithm. Within web3, FedML and FLock are attempting to combine the federated learning approach with token incentives.

It’s also worth noting that decentralized data stores like Filecoin and Arweave, along with databases like Space and Time could play a significant role on the data pre-processing side.

Consensus-based ML

Another novel form of infrastructure around inference using blockchains is the concept of consensus-based ML. Bittensor is the most prominent example of this: a substrate-based L1 blockchain designed to improve machine learning efficiency and collaboration through its use of application-specific subnets. Each subnet has its own incentive system to service various use cases, from LLMs to prediction models to generative creativity. What’s unique about Bittensor is how it uses miners to coordinate quality outputs: Miners earn TAO, the native token, by providing intelligence via their ML models with outputs rated by verifiers. Since miners are incentivized for the best outputs, they consistently improve their models to remain competitive, resulting in a faster learning process coordinated by tokenomics.

The most recent exciting developments in the TAO ecosystem has been the Dynamic TAO proposal transitioning Bittensor to a more automated, market-driven mechanism design around token emissions and the Nous subnet launched to provide incentivized model fine-tuning in order to compete with the likes of OpenAI.

We are likely to see additional attempts at this type of system, where mining or consensus mediates model outputs in a way conducive to quality.

Intent is All You Need

Within DeFi, a recent narrative within the MEV space is around user intents and using economically-aligned solvers to execute those intents. The discussion of intents is often frenetic, but one thing has become increasingly clear: intents from users require higher-order semantic context to be parsed into executable code. LLMs could provide this semantic layer.

Propellerheads laid out the most articulate vision of using LLMs in the intents space to date here: https://www.propellerheads.xyz/blog/blockchain-and-llms.

In short, LLMs can help locate coincidences of wants (CoWs) opportunities by turning almost-matching intents into matching intents through semantic understanding. This can be done through inward intent renegotiation (e.g. "Is it ok to buy LUSD instead of USDC? I found a matching limit order and you'd save 0.3% on trading fees with this CoW.") and outward intent renegotiation (e.g. "I want to buy this other BAYC that you have; would you accept to sell that one for X ETH?").

There are other constructions that can be done here as well, which becomes especially interesting in the context of wallets and multisigs post-Account Abstraction. Projects like DAIN and Autonolas have experimented with using agents as signers on a wallet for both security and intent-based purposes for example. Talking to your wallet and having it execute txs on your behalf is not far off.

Also of note is the myriad of DeFi use cases, such as agent-based prediction markets, AI-managed economies, and ML-parameterized DeFi applications, check out my zkML piece for more ideas here.

Agent Economies

One of my favorite areas of infrastructure to date is in AI agent economies. There is a vision for the world in which everyone has their own agent, hires high quality and finely trained agents for services, or have autonomous agents fulfilling their objectives in a complex economy. In order to do this, it is vital that agents have a way to pay for and receive payments for their services. It is definitely possible that traditional fiat rails open up for these agents, but more likely is that agents will transact in crypto given its ease of use, speed of settlement, and permissionless nature.

Autonolas and DAIN are examples of projects working in this field. With Autonolas, agents are effectively nodes in a network working on a specific goal that are maintained by incentivized service operators, similar to a keeper network. These agents can be used for a variety of services, such as oracles, prediction markets, messaging, and more. DAIN takes a similar approach giving agents the ability to “discover, interact, transact, and collaborate with other agents from the network”.

Other Ideas

Outside of those above we also have also seen:

• Decentralized vector databases for fine-tuning models like BagelDB

• Wallets for API keys and SIWE for AI applications like Window.ai

• Data provisioning services

• Indexing and searching tools like Kaito

• Block explorers and dashboards like Modulus Labs’ AI verification dashboard which right now is verifying a stream of inferences for the Upshot model

• Development co-pilot like Dune’s model for onchain SQL queries

• Simulated agent testing environments

• Bandwidth for data-scraping like the Grass Network

• Synthetic data and human RLHF platforms

• DeSci applications like the distributed bioML tools for protein-folding by LabDAO

A myriad of ideas are blossoming within web3 for servicing various areas of AI, so highlights are only provided here, but I strongly encourage exploring the projects above to get a sense of scale and utility.

The Intersection of it All

AI and crypto are synergistic. Both have leaned into open-source, pushed back against regulations, and are creating some of the largest wealth transfers in history. They need each other, each solving one another’s core challenges.

For crypto, AI solves issues in UX, facilitates more creative onchain use cases, enhances the ability of decentralized organizations and smart contracts, and unlocks true innovation at both the application and infrastructure layers.

For AI, crypto solves problems of authenticity and provenance, increases coordination around open-source models and datasets, helps with bootstrapping compute and data, and enables creators and agents to participate more directly in post-AI economics.

The challenge now is for crypto hackers, teams, and projects to understand this transition and embrace it. Creativity is now unbounded and we are sitting at the intersection of it all.

If you are interested in learning more about these intersections, come to our conference Convergence at ETHDenver.

If you are building within these intersections, DM me.

While composability, modding, and extendability are often discussed as core reasons for building fully onchain games and worlds, there has been very little real-time experimentation around those properties.

Is fully onchain worth it? Do we need to architect games to be composable from the very start, or is pure emergence enough? Can we leverage composability to create compelling net new entertainment systems?

‘Modding worlds’ is a series dedicated to exploring this. For the first edition, we’ll be walking through our recent internal hackathon project: Hexcraft, a MOBA we attempted to build within Downstream – a fully onchain game created by Playmint.

What is Downstream and why is it special?

Downstream is a hex grid world with three key resources (red goo, blue goo, green goo) that represent the properties of attack, defense, and life. These atomic units make up everything else in the game and with these resources, you can construct buildings and craft items that carry over properties made up of their underlying ingredients. By breaking the game down into three core atomic units, we’re able to create a framework for gameplay that enables an infinite number of permutable lego building blocks to exist.

Rather than being opinionated from the start about what gameplay should look like, Playmint has built a composable system first, and only then after, have started to actively explore what experiences can be built around it – being incredibly cautious about how game design decisions such as global play objectives, economic design, and cartography implicate composability.

As such, Downstream, at the time of writing this post, did not yet have an overarching player objective nor hard rules around cartography or economic design. However, we did have a core set of physics to build around: player movement, properties system (attack, defense, life), combat system, and item/building construction system.

Ducks vs. Burgers

An example of a game the team had designed around these initial physics was Ducks vs. Burgers:  a team-based game focused on two sides constructing and attacking each other’s buildings. The team with the highest number of buildings remaining within a time limit would win. A video breakdown can be found below:

Based on what was available to us, we looked to build a minimally coherent game that strived to be as fun and entertaining as possible. Composability doesn’t matter if you’re not creating an experience that is compelling to a player audience. This is a consideration we took pretty seriously in our efforts to design something even, despite only having two days to work on something that we could ship. And as a result, Hexcraft was born.

Hexcraft v0.1

Hexcraft is a MOBA built on top of Downstream:

• Players register into two teams (red and blue)

• The first team to destroy the other team’s base wins

• Players have 5 different classes to pick from (trade-offs in attack, defense and life)

• Player movement speed across the map is dependent on destroying blockers that obstruct movement (the speed of destroying blockers are based on player stats)

Basics around gameplay

Hexcraft looked to iterate on Ducks vs. Burgers by instead focusing the game around the attacking, defending, and destruction of each respective base. We felt like this was a far more commonly understood player game loop. From playing Ducks vs. Burgers ourselves, we realized that building construction was potentially too complicated to base the game around. We’ve decided to let building and item construction emerge naturally as a meta within the game as players better understand the core mechanics of the world.

Team registration and base destruction win condition

Naturally, in the absence of a defined player objective and means of organizing players into teams, we built a system for doing so. We implemented a system of two buildings:

1. Team registration office: Allowed players to register themselves into two teams

2. Bases: Once destroyed, resets the teams assigned to players

This system forms the basis of Hexcraft but realistically could also be used as a building block for other games revolving around the same team and win conditions. In a future version of this system, ideally the win condition can be designed to be completely arbitrary to any other in-game event as players or a game maker desires.

Character class system

We felt like it would have been horribly boring if every player had the same attack, defense, and life! So, we built a character class system to allow players to pick a unique class which would give them specific trade-offs in each respective stat (attack, defense, life).

This system consisted of 5 buildings that allowed players to craft a corresponding item with goo that represented the stat bonuses of that class. We designed some game balancing restrictions by building smart contracts so that each player could only craft a class item if they didn’t have any other crafted item in their inventory (to enforce a single class per game).

Classes and their corresponding buildings/items (attack/defense/life):

• Fighter’s Arena produced: Greatsword of Dawn (35,10,10)

• Wizards Maze produced: Magical Glock 19 (50,0,0)

• Paladin’s Church produced: Strong Holy Kiteshield (10/40/40)

• Cleric’s Tent produced: Friendly Companion (0/100/50)

• Bard Concert Hall produced: Sharp Violin (25, 25, 50)

The goal of this class system was to introduce decision tree complexity and to generate a natural meta towards team coordination around classes. Each class naturally influences movement ability (based on player ability to destroy blockers on the map in order to move) and speed and resource availability for each team (different classes consume up more goo than others). A future version of this class system could naturally allow anyone to create any permutable class that can be played within the game as long as they’re able to craft it based on the initial starting resources available for each team.

An interesting idea is that even though this class system was built for a MOBA-like environment, you could take the system as a set of player constraints that can be played within a completely unregulated environment. This would become the equivalent of playing the game on “hard”, simulating a PvE situation, potentially with players who are opting into the system of constraints sharing a single shared cooperative goal and competing against other players outside of the system of constraints (think left for dead).

Map design and player movement

Due to the limitations in scope, we were not able to build any mods around player movement or spawning, which meant that we needed to add a spectator ring around the map to allow players to self-organize themselves into the right starting player positions. For now, players need to abide by socially enforced rules around when to start a game, splitting up into equal teams, and other aspects to ensure fairness in a game. You can think of this version of Hexcraft as a cardboard prototype equivalent that requires players to self-shuffle and deal cards to play.

Shelved ideas that we didn’t get to implement…

Sound packs

Within Downstream, we wanted to add a little juice to the micro-interactions of gameplay and designed a third-party sound pack system for the game. In order to activate the sound pack, a player would only just need to open up our soundpack webpage via another tab while playing the game. This was technically possible within the interface of Downstream via embedded javascript within the buildings, but these only ran periodically rather than in real time (an easy fix). This would be the first step towards an ecosystem of mods that revolves around sound:

1. Allow players to trade NFTs that represented rare, unique soundtracks (token-gated sound design)

2. Allow players or creators to create entire sound packs that are then unlocked by paying for them via ETH or in-game tokens

3. Allow players to host permissionless and uncensored voice channels during games

UGC does not necessarily need to exist fully onchain, and sound design is an area we feel that is heavily overlooked as a meaningful way to add texture and substance to a game.

XP system

Since there was no sense of narrative and player progression that existed, we wanted to build a way to generate lore and history for players. There were several ways to go about this:

1. Allow players to mint NFTs of the games that they played. Each mint would include generative lore and data influenced by the class the player played during the game, whether their team won, team size and battle history within the game, and more. Not only does this serve as a reward for playing the game, but it could also be onchain data that could interoperate with other games that could compose with –  similar to Loot, except rather than a predefined total supply, they are issued via gameplay. These could easily be visualized by an external third-party site as needed. These could also be issued into a player’s ERC-6551 token bound account.

2. Fair launch issuance of ERC-20 specific to just Hexcraft: a means of bootstrapping a microeconomy nested within the broader Downstream system. You can think of this as Roblox but without economic silos between games.

3. Create an external site that visualizes an RPG-like interface of your own hero/character whereby players can opt into a progression and XP system that is entirely third party.

Reflections and thoughts

Live Demo Gameplay Footage (No commentary/audio):

Designing around the open-endedness

When we initially sat down to design what we could build during the duration of the hackathon, it wasn’t immediately clear what we would build. The Ducks vs. Burgers example was immensely helpful as a starting point. Only after we started to design Hexcraft did we start to find the lack of strong, opinionated game design more of a blessing than a curse, as it allowed for significantly greater creative freedom to what could be built.

Downstream in its current state has intentionally taken an extreme position with regards to holding back on clear game design decisions for the time being. While there’s a clear need to balance immediate coherence while allowing an unrestricted field of possible gameplay comprehension, there’s clearly a sense of freedom afforded by optimizing for the latter.

Iteration speed

There was immense value in taking the burden of the front-end visual experience away from the design of Hexcraft, as it constrained our area of focus towards extending core mechanics of how our world systems worked. This was especially important being novice game designers as our efforts were allowed to be focused on game design and game mechanics.

Through the process of designing and building Hexcraft, it was apparent that Downstream, while not yet perfect, was on the path of lowering the barrier to entry for many potential new game designers. Creatives generally aren’t necessarily the best engineers themselves, and the majority will often opt for the path of least resistance to get their vision off the ground in the beginning. A key determining factor for how well a team can propagate UGC will entirely depend on the speed of iteration. An example of this can be seen within crypto with memecoins, nfts, liquidity pools, and DAOs – where 0 technical capabilities are required to leverage each respective set of building blocks to coordinate social and economic value. If such basic building blocks have already enabled such social and cultural impact, what could emerge once more sophisticated structures are able to be built and released in an equivalent manner?

While yes, “build worlds that have never existed before” should be the end vision for building fully onchain worlds, when building with Downstream, it was clearer that the more interim goal was to simply build games that had never existed within the context of crypto before. And if we’re able to make gradual ground on this more tangible objective, we might just end up achieving the former. And when you zoom out on this, Downstream starts to look more like a sandbox for economic value and incentive design where players navigate those rules through the controls of a traditional game. If Roblox allowed for rapid iteration of core 3D gameplay, then it's likely that Downstream is well positioned as a system that allows for rapid iteration of economic design.

Collaborate with us

We believe that there is a depth of open possibilities that are made possible by creating games that are deployed fully onchain, and we have never been more excited to contribute to this design space.

We are always looking to playtest new games and are actively investing in teams in this space, especially teams focused on pushing the boundaries of game design.

• Watch out for Hexcraft: we’ll be starting public playtest sessions

• If you are working on a project in the autonomous worlds design space, do not hesitate to reach out to us via twitter, seam or email.

Trust-minimized – (a feature of) an L2 system is trust-minimized if it functions without requiring trust external to the base L1.

Horizontal scaling – a system is horizontally scalable if instances can be added without imposing global bottlenecks.

In this article, we argue that trust-minimized and horizontally scalable systems are the most promising way of scaling blockchain applications, yet they are currently under-explored. We present the argument by exploring three questions:

1. Why should applications be trust-minimized?

2. Why build systems that are horizontally scalable?

3. How can we maximize both trust minimization and horizontally scalability?

(Disclaimer: although we will focus on Ethereum as the base L1 in this article, most of what we discuss here applies to decentralized settlement layers beyond Ethereum.)

Why should applications be trust-minimized?

Applications can be connected to Ethereum in a trusted manner – they can write to and read from the Ethereum blockchain but trust is placed on the operators to execute business logic correctly. Centralized exchanges like Binance and Coinbase are great examples of trusted applications. Being connected to Ethereum means that applications can tap into a global settlement network with a diverse set of assets.

There are significant risks associated with trusted off-chain services. The collapse of major exchanges and services in 2022, such as FTX and Celsius, is a great cautionary tale of what happens when trusted services misbehave and fail.

On the other hand, trust-minimized applications can write to and read from Ethereum verifiably. Examples include smart contract applications such as Uniswap, rollups such as Arbitrum or zkSync, and coprocessors such as Lagrange and Axiom. Broadly speaking, trust is removed as applications become secured by the Ethereum network, as more functionalities (see below) get outsourced to L1. As a result, trust-minimized financial services can be offered without counterparty or custodian risks.

There are three key properties that applications and services can have, which can be outsourced to L1:

• Liveness (and ordering): user-submitted transactions should be included (executed and settled) in a timely manner.

• Validity: transactions are processed according to prespecified rules.

• Data (and state) availability: historical data, as well as current application state, is made accessible to the user.

For each of the above properties, we can think of what is the trust assumption required; in particular, does Eth L1 provide the property or is external trust required. The table below categorizes this for different architecture paradigms.

Why build systems that are horizontally scalable?

Horizontal scaling refers to scaling via the addition of independent or parallel instances of a system, e.g. application or rollup. This requires no global bottleneck to be present. Horizontal scaling enables and facilitates exponential growth.

Vertical scaling refers to scaling via the increase of throughput of a monolithic system, such as Eth L1 or a data availability layer. When horizontal scaling runs into bottlenecks on such a shared resource, vertical scaling is often required.

Claim 1: (Transaction-data) rollups cannot horizontally scale because they can be bottlenecked by data availability (DA). Vertically scaling DA solutions require making compromises on decentralization.

Data availability (DA) remains the bottleneck for rollups. Currently, each L1 block has a maximum size target of ~1 MB (85 KB/s). With EIP-4844, there will be an additional ~2 MB (171 KB/s) made available (in the long-term). With Danksharding, Eth L1 may eventually support up to 1.3 MB/s of DA bandwidth. Eth L1 DA is a shared resource that many applications & services compete for. Therefore, although using L1 for DA provides the best security, it bottlenecks the potential scalability of such systems. Systems that utilize L1 for DA will (typically) not be able to horizontally scale and have diseconomies of scale. Alternative DA layers, such as Celestia or EigenDA, also have bandwidth limits (although larger, at 6.67 MB/s and 15 MB/s, respectively). But it comes at the expense of shifting the trust assumption from Ethereum to another (often less decentralized) network, compromising on (economic) security.

Claim 2: The only way to horizontally scale trust-minimized services is to obtain (close to) zero marginal L1 data per transaction. The two known approaches are state-diff rollups (SDR) and validiums.

State-diff rollups (SDRs) are rollups that post state differences across an aggregated batch of transactions to Ethereum L1. For the EVM, as transaction batches grow larger, the per transaction data posted to L1 diminishes to a constant that is much smaller than that of transaction-data rollups.

For example, during the stress-test event of high in-flood of inscriptions, zkSync saw a reduction of calldata per transaction down to as low as 10 bytes per transaction. In contrast, transaction-data rollups like Arbitrum, Optimism, and Polygon zkEVM, typically see around 100 bytes per transaction for normal traffic.

A validium is a system that posts validity proofs of state transitions to Ethereum, without associated transaction data or state. Validiums are highly horizontally scalable, even under low traffic conditions. This is especially true as the settlement of different validiums can be aggregated.

Besides horizontal scalability, a validium can also provide onchain privacy (from public observers). A validium with private DA has centralized and gated data and state availability, meaning that users have to authenticate themselves before accessing data and that the operator can enforce good privacy measures. This enables a level of user experience similar to traditional web or financial services – user activities are hidden from public scrutiny but there is a trusted custodian of user data, in this case the validium operator.

What about centralized vs decentralized sequencers? To keep systems horizontally scalable, it is crucial to instantiate independent sequencers, either centralized or decentralized. Notably, although systems using shared sequencers enjoy atomic composability, they cannot horizontally scale, as the sequencer can become a bottleneck as more systems are added.

What about interoperability? Horizontally scalable systems can interoperate without additional trust if they all settle to the same L1, as messages can be sent from one system to another via the shared settlement layer. There is a tradeoff between operating cost and messaging delay (which can potentially be solved at the application layer).

Trust-minimization for horizontally scalable systems

Can we further minimize trust requirements for liveness, ordering, and data availability in horizontally scalable systems?

It is of note that, at the cost of horizontal scalability, we know how to salvage trustless liveness and data availability. For example, L2 transactions can be initiated from the L1 for guaranteed inclusion. Volition can offer opt-in L1 state availability for users.

Another solution is to simply _decentralize _(but not rely on the L1). Instead of a single sequencer, systems could become more decentralized by utilizing decentralized sequencers (such as Espresso Systems or Astria), therefore minimizing the trust required for liveness, ordering, and data availability. Doing so places limitations compared to single-operator solutions: (1) performance may be bounded by the performance of the distributed system, and (2) for validiums with private DA, the default privacy guarantee is lost if the decentralized sequencer network is permissionless.

How much trust can we additionally minimize for single-operator validiums or SDRs? There are a couple of open directions here.

Open direction 1: Trust-minimized data availability in validiums. Plasma solves the state availability problem to a certain extent–It solves the problem either for withdrawals only for certain state models (which includes the UTXO state model), or requires users to be online periodically (Plasma Free).

Open direction 2: Accountable pre-confirmations in SDRs and validiums. The goal here is to provide users with fast pre-confirmation of transaction inclusion from a sequencer, and the confirmation should allow the user to challenge and slash the economic stake of the sequencer if the inclusion promise is not fulfilled. The challenge here is that proving non-inclusion (necessary for slashing) likely requires additional data for the user, which a sequencer can simply withhold. Therefore, it is reasonable to assume that we at least require the SDR or validium to employ a (potentially permissioned) data availability committee for its full calldata or transaction history, which enables the same committee to provide proof of non-inclusion (of pre-confirmed transactions) upon a user request.

Open direction 3: Fast recovery from liveness failures. Single-operator systems can suffer from liveness failures (e.g. Arbitrum went offline during the inscription event). Can we design systems that provide minimal service disruption in this scenario? In some sense, L2s that allow self-sequence and state proposals do provide guarantees against prolonged liveness failures. Designing single-operator systems that are more resilient against shorter liveness failures is currently under-explored. One potential solution here is to make liveness failures accountable, by providing slashing against liveness failures. Another potential solution is to simply shorten the delay period (which is currently set to be around a week) before a take-over can happen.

Conclusion

Scaling a global settlement ledger while maintaining trust minimization is a hard problem. There has not been a clear distinction between vertical scaling and horizontal scaling in the rollup and data availability world today. To truly scale trust-minimized systems to everyone on earth, we need to build trust-minimized and horizontally scalable systems.

Acknowledgements

Many thanks to Vitalik Buterin and Terry Chung for feedback and discussion, as well as Diana Biggs for her editorial comments.

Since the launch of the first bitcoin tracker in Sweden in 2015, crypto ETPs have grown from a largely European phenomenon to an expanding global footprint and from a range of just 17 products at the end of 2020 to circa 180 trading today. As more TradFi institutions join the crypto-native firms in their issuance, the role of ETPs in not only expanding investor access to crypto but in crypto’s overall acceptance and integration in global financial markets is becoming increasingly clear.

This article provides an overview of crypto ETPs, including the types of products currently available, the operating model, regional overviews, and developments we’re watching in this rapidly evolving landscape.

Overview of crypto ETPs

What are crypto ETPs?

Exchange-traded products (ETPs) are a category of financial products bought and sold on regulated stock exchanges daily throughout regular trading hours, tracking the return of an underlying benchmark, asset, or portfolio.

There are three main types of ETPs: exchange-traded funds (ETFs), exchange-traded notes (ETNs), and exchange-traded commodities (ETCs). While ETFs are investment funds, ETNs and ETCs are debt securities, with ETCs tracking physical commodities such as gold and oil and ETNs used for all other types of financial instruments. Since the creation of the first ETF in 1993, thirty years ago this year, ETPs have grown from equity market trackers to arguably one of the most innovative categories of investment products providing investors with exposure to a range of innovative underlying.

Note: While ‘ETPs’ is the umbrella term for this category of products, the term ETP is sometimes used as a category term for debt security exchange-traded products.

Particularly over the past twenty years, these types of products have experienced continued growth, reaching 11,859 products and 23,931 listings globally from 718 providers listed on 81 exchanges in 63 countries; of these, ETFs make up the lion’s share of assets, accounting for $10,747 Bn or 98% of the total $10,990 Bn in ETP assets (data from ETFGI as of end Nov 2023). Oliver Wyman expects growth in ETFs to accelerate in growth from recent years, forecasting the market to grow at 13 to 18% p.a. from 2022 to 2027.

The convenience and accessibility of ETPs make them a popular tool for opening new asset classes and investment strategies to investors — here, crypto is no exception.

The first ever Bitcoin ETP was launched in 2015 on the Swedish NASDAQ by XBT Provider (later acquired by Coinshares). Growth in the market remained relatively modest until the latter half of 2020, when a strong uptick in the number of products began, which continues today, both from new crypto-native entrants and traditional issuers. In February 2021, Canada won the race for the world’s first Bitcoin ETF, with Purpose Investment’s launch of its Purpose Bitcoin ETF on the Toronto Stock Exchange. While crypto ETPs, structured as debt securities, still far surpass crypto ETFs both in number and AUM, we expect to see this start to shift, particularly as the market for US spot ETFs opens.

The number of crypto products has grown steadily, particularly within the last three years, reaching 176 Crypto ETFs and ETPs as of November 2023, based on data from ETFGI. Assets invested in these products rose 120% in the first 11 months of the year, growing from $5.79 Bn at the end of 2022 to $12.73 Bn at the end of November 2023.

Why crypto ETPs?

The idea of a crypto ETP may seem counterintuitive to crypto natives: introducing intermediaries, which the technology ultimately enables us to remove. However, as well-understood and regulated investment products, these products open up the opportunity for crypto exposure to a much larger investor audience who, for various reasons, may otherwise be unable to access the asset class. For instance, retail investors may lack the tools, time, risk tolerance, and technical expertise to invest directly in crypto. The structure of ETPs as traditional securities opens them to institutional investors who may be restricted to only investing in these types of instruments or who wish to avoid directly holding crypto assets, be that for regulatory, compliance, technical, or other reasons.

There are also potential disadvantages and considerations (as not all investors would consider these to be disadvantages) with these products in comparison with purchasing crypto directly. These include the fact that crypto ETPs to date have had much higher fees than is typical for these products (although these are already falling with increasing competition), the limited trading availability of traditional exchanges vs. crypto’s 24/7 markets, access restrictions based on jurisdiction and investor type as per regulatory approvals, counterparty, tracking error, and exchange rate risks, and settlement times.

Product structure and offerings

Broadly speaking, crypto ETPs fall into two product categories and types: ETF or ETP and physical, also referred to as asset-backed, or synthetic.

Crypto ETF structures

ETFs are structured as funds, with ETF holdings being shares in a fund. There is typically a legal separation of the funds from their issuing entities through the use of trusts, investment companies, or limited partnerships to ensure that investors’ holdings are protected in the event of the parent company/issuer’s bankruptcy. ETFs are typically subject to additional rules and transparency requirements depending on their jurisdiction; for example, ETFs domiciled in the EU and marketed to EU investors are generally required to comply with UCITS (Undertakings for Collective Investment in Transferable Securities) regulations, which brings diversification requirements, such as that no single asset can represent more than 10% of the fund.

Most crypto ETFs today are either spot or futures products. Spot ETFs have direct ownership of the underlying crypto assets, secured with independent custodians. With the futures ETFs, the issuers are not holding the underlying crypto but instead purchasing futures contracts for the assets. As such, these products don’t directly track the spot price of the underlying assets and are generally considered to bring greater complexity and cost and to be less transparent and intuitive for investors.

Crypto ETP structures

Crypto ETPs (in this instance, using the term to refer to products other than ETFs) are structured as debt securities. While their structuring requirements are less stringent than those of ETFs, their disclosure requirements are very similar.

Physical crypto ETPs are secured debt obligations 100% backed by holdings of the underlying cryptocurrency they track. The crypto assets are physically bought and held with independent third-party custodians under the oversight and control of a designated trustee, who holds the rights and entitlements on behalf of the ETP holders and the responsibility for organizing redemptions in the event of an issuer insolvency.

Synthetic ETPs are unsecured debt obligations, meaning the issuer is not required to hold the underlying assets the product is tracking and instead uses derivatives and swaps to track the performance (exact structure and terms will vary). Synthetic ETPs, therefore, bear greater counterparty risk as there is no legal requirement for the products to be backed 1-to-1 by the underlying physical asset. XBT Provider (a Coinshares company) and Valour are two crypto ETP issuers offering synthetic products.

By and large, most crypto ETPs on the market are physical products, as many investors prefer the transparency and reduced counterparty risk provided by this structure.

Crypto Product Offerings

Crypto ETPs initially began as single digital asset trackers, and today, the range of crypto ETPs available on the market also includes basket, staking, inverse, and leveraged products, as well as certain indices designed to manage volatility.

In terms of underlyings, using data recently compiled by BitMEX Research, removing equity and OTC-traded funds and adding in additional data, we find that out of 162 crypto ETPs, bitcoin, ethereum, and basket products account for 58%, with the other 42% consisting of the long tail single digital assets, as well as short, volatility and leverage products.

Of these 162 products, 121 are ETPs, and 41 are ETFs, of which 16 are futures ETFs, and 11 are the US spot bitcoin ETFs awaiting launch. Staking products, meaning investors benefit from the staking yield generated by their holdings, currently total 14: thirteen ETPs and one ETF.

Largest products by AUM

The largest crypto ETP by AUM is the ProShares Bitcoin Strategy ETF, a US-based futures product holding assets of US$1.68 billion, based on available data as of 2 Jan 2024. As shown in the table below, nine of the 14 top crypto ETPs by assets track bitcoin (64%); of the remaining five, they include three ethereum, one solana, and one Binance coin tracker.

Of these 14 products, four are domiciled in Switzerland (all from issuer 21Shares), three are domiciled in Canada, two in Jersey, one in Germany, one in the US, and one in Liechtenstein.

Four of the top 14 products by assets are ETFs: three spot and one futures. Of the ETPs, eight are physical and two are synthetic products.

Product innovation

There are several constraints to consider when bringing new crypto ETPs to market. These include regulatory and stock exchange requirements and approvals, liquidity requirements, investor demand, and the availability of public price data and fiat trading pairs. That said, we’re seeing continued product innovation from both issuers and index providers as more players enter the market and look to capture market share and differentiate themselves and as regulatory, service provider, and investor understanding and acceptance of this asset class increases.

Crypto ETP Operating Model

Creating an ETP begins with the issuer, meaning the investment company or trust that will issue the products, writing a prospectus document to be approved by the regulator. These can vary by jurisdiction, but in general, they include the details of the issuer, including the identities of its directors and financial statements, the product and program design, including an overview of the underlying assets, intended markets, and service providers, comprehensive overview of the potential risks, details of the asset valuation (NAV) and NAV calculation methodologies, fees, and redemption processes.

Following approval by a regulatory authority and successful engagement of the necessary service providers, the issuer must apply for listings on the desired stock exchanges. The rules regarding which types of products, i.e. asset-backed or futures, and underlyings, i.e. cryptoassets, are eligible to be listed vary by exchange.

The operating model and range of service providers can vary by product type, jurisdiction, and the program design of the issuer. A high-level overview of a typical model is as follows:

• On the primary market, the Issuer exchanges shares of the product with the APs in return for either the underlying crypto assets (“in-kind”) or the cash equivalent, with underlying crypto assets delivered to and from the designated custodians as required. Depending on the structure, a transfer agent and trustee may be involved in liquidating the collateral and transferring funds.

• As the APs manage primary market creation and redemption, market makers provide liquidity in the secondary market, ensuring continuous, efficient trading.

• Investors buy and sell the products on the secondary market, typically placing their orders via their bank or broker, who in turn execute their orders on the relevant stock exchange, either directly or via other intermediaries.

Operational stakeholders and service providers

Issuers

Issuers are responsible for the overall design and creation of the ETPs, coordinating and managing the relevant intermediaries across the product lifecycle. Regulatory oversight of issuers varies by jurisdiction and may not necessarily be a requirement. However, the regulator evaluates the issuer as part of the prospectus approval process, as does the exchange during the listing process, where requirements may include corporate governance, capital requirements, and regular audits. The issuer usually establishes a stand-alone special purpose vehicle (SPV) to issue the products. Initially, the majority of crypto ETP issuers were crypto native firms such as Coinshares, 21Shares, 3iQ, Hashdex, and Valour, joined in recent years by an increasing number of traditional finance firms, including WisdomTree, VanEck, Invesco, Fidelity, and, pending SEC approval, BlackRock and Franklin Templeton.

Custodians

Custodians hold the underlying cryptocurrency for physically-backed products. Custodians used by ETP issuers include Coinbase, Fidelity Digital Assets, Komainu, BitGo, Copper, Swissquote, Tetra Trust, Zodia Custody, and Gemini.

Market Makers

Market Makers (MMs) are the liquidity providers engaged by the issuers to provide the necessary liquidity for the ETPs by providing two-sided quotes on the exchanges at the agreed terms set out contractually. These include Flow Traders and GHCO.

Authorized Participants

Authorized Participants (APs), typically banks or broker-dealers, have the right to create and redeem shares of the products on a daily basis directly with the issuer. They do this by delivering the underlying assets or their cash equivalent to the issuer in exchange for newly created ETP shares or by returning shares to the issuer in exchange for the underlying assets or cash. The appetite of APs to engage in crypto, in particular, to cover assets outside of BTC and ETH, may vary based on factors such as regulatory uncertainty and market conditions. APs who have been active in crypto ETPs include Flow Traders, GHCO, Virtu Financial, DRW, Bluefin, and Enigma Securities. JP Morgan, Jane Street, and Cantor Fitzgerald & Co have recently been named as APs in the US spot bitcoin ETF filings.

Index Providers

Index providers are responsible for creating, designing, calculating, and maintaining indices and benchmarks that ETPs track. These provide both issuers and investors with transparency and reliability. In certain jurisdictions, index providers are subject to regulatory oversight, for instance, in the EU with the European Benchmark Regulation (BMR). Index providers active in crypto ETPs include MarketVector Indexes, CF Benchmarks (acquired by Kraken in 2019), Vinter (a crypto-native index provider out of Sweden), Bloomberg, and Compass.

Exchanges and Multilateral Trading Facilities (MTFs)

The willingness of exchanges and MTFs to list crypto ETPs will first be mandated by their local regulations and regulatory approval of the issuer’s prospectus, after which it becomes a business decision determined by their respective issuer and product eligibility requirements, which typically includes evaluation of the underlying asset across parameters such as liquidity, compliance, publicly available pricing information, and risk mitigation. The rules regarding which types of products are eligible for listing will vary by trading venue; for example, Xetra in Germany only lists asset-backed ETPs, and SIX Swiss Exchange has specific rules regarding eligible crypto underlyings.

Trustees

A trustee may be responsible for safeguarding the assets and representing the interests of the ETP holders or investors. Their exact role and responsibilities can vary depending on the specific structure and legal arrangements of the ETP. Trustees active in crypto ETPs include Law Debenture Trust Corporation, Apex Corporate Trust Services, Bankhaus von der Heydt, and Griffin Trust.

Administrators

Administrators support the overall operational management of the ETPs. Their services may include accounting, regulatory compliance, financial reporting, and shareholder services. Administrators active in crypto ETPs include State Street, JTC Fund Solutions, The Bank of New York Mellon, Formidium, NAV Consulting, Theorem Fund Services, and CIBC Mellon Global Securities Services.

Additional service providers

Other service providers that may play a role within the ETP program and product lifecycle include, but are not limited to, Paying Agents, responsible for registering new units of the ETPs and obtaining ISINs from local agencies, Transfer Agents, which may be used to maintain records of shareholders and other duties, Calculation Agents, for the calculation of the NAV of the underlying assets, and Registrars, for shareholder recordkeeping. These various roles and responsibilities may overlap and/or be covered by different parties depending on the product type, issuer, and jurisdiction.

A note on fees

ETPs charge management fees, also referred to as expense ratios or sponsor fees, to cover the costs of managing and operating the products, which are calculated annually as a percentage of holdings and deducted from the NAV daily or periodically. Many of the early crypto ETPs were able to charge high fees of up to 2.5%, whereas typically, ETPs fall within the range of 0.05% to 0.75% for fees. The AUM of crypto ETPs charging 2.5% when alternatives charge as low as 0% speaks to the stickiness and first-mover advantage of these products.

We expect fees to be a key differentiator going forward with new products, as is seen now with the US spot ETF. Of the fees announced to date, Invesco/Galaxy are waiving fees for the first six months and the first $5b in assets, and Fidelity is offering a fee of 0.39%.

Regional offerings

Europe

Crypto ETPs originated in Europe, with the first bitcoin product launched in Sweden in 2015, a synthetic tracker ETP issued by XBT Provider (acquired by Coinshares in 2017). In Europe, crypto ETP issuers benefit from the single market, as once an ETP prospectus is approved by one European country regulator, upon notification and approval, the products may also be made available in other member states (referred to as “passporting” the prospectus). Sweden’s SFSA remains a popular choice for European crypto ETP prospectus approvals. Germany is another jurisdiction that has approved crypto ETP prospectuses and where crypto ETPs have good availability across trading venues, for example from leading exchange groups such as Deutsche Boerse and Boerse Stuttgart Group.

ETPs remain the dominant product type in Europe and the reason for the lack of a true crypto ETF in Europe is largely due to the UCITS (Undertakings for Collective Investment in Transferable Securities) regulations. By and large, the majority of European ETFs are UCITS compliant in order to benefit from pan-European passporting, which allows their sale to retail investors in other EU member states in addition to the country of registration, and the assurances and investor protections these regulations provide. The UCITS rules and requirements, however, are currently incompatible with single asset tracker products such as a bitcoin ETF. For example, UCITS diversification requirements include that no single asset shall represent more than 10% of the fund and underlyings must be a qualifying financial instrument. In June 2023, the European Commission tasked the European Securities and Market Authority (ESMA) with investigating whether updates to the UCITS rules are required and noting crypto assets. However, the intention of this exercise seems to be to determine whether more rules and investor protections are required, rather than expanding the eligible product types. The deadline for ESMA’s input is 31 Oct 2024.

Switzerland

In 2016, Switzerland became the second jurisdiction after Sweden to approve and list a crypto ETP, a bitcoin tracker certificate by Bank Vontobel launched on the SIX Swiss Exchange. The world’s first crypto index product was then launched in Switzerland in November 2018, a physically-backed basket ETP comprised of bitcoin (BTC), ethereum (ETH), ripple (XRP) and litecoin (LTC) from issuer 21Shares (then named Amun). The SIX Swiss Exchange has specific rules for crypto underlyings including “that at the time of applying for provisional admission to trading, the cryptocurrency must be among the 15 largest cryptocurrencies, as measured by market capitalization in USD” and from our research has the widest range of products with crypto underlyings of any trading venue globally. Swiss stock exchange BX Swiss also permits products with crypto underlyings, with rules including that the underlying must be within the top 50 cryptocurrencies by market cap.

UK

In October 2020, the UK’s Financial Conduct Authority (FCA) banned the selling, marketing, and distributing to retail investors of any derivatives or ETNs that reference “unregulated transferable cryptoassets.” A number of crypto ETPs are listed in the UK on Aquis Exchange, available to professional investors only.

Canada

Canada was the first country to approve a bitcoin ETF, with the first product launched by Purpose Investments on the Toronto Stock Exchange (TSX) in February 2021, followed in April of that year by the ethereum ETFs. In October 2023, 3iQ launched staking in their ethereum ETF, with the staking rewards accrued into the fund, a first in North America. Other Canadian crypto ETF issuers include Fidelity Investments Canada, CI Global Asset Management (CI GAM) in partnership with Galaxy, and Evolve Funds.

Brazil

Canada’s approval of a bitcoin ETF was quickly followed by Brazil when the Brazil Securities and Exchange Commission (CVM) approved the first bitcoin exchange-traded fund (ETF) in the Latin American region in March 2021. Brazilian crypto ETF issuers include crypto asset managers Hashdex and QR Capital, and Itaú Asset Management in partnership with Galaxy.

USA

To date, only crypto futures ETFs have been approved by the SEC and made available to investors. The first bitcoin futures ETF, by ProShares, launched on 19 October 2021 as one of the most heavily traded fund launches in market history, attracting more than $1bn in assets within the first days of trading.

Two years later, on 2 October 2023, the first US ethereum futures ETFs were launched by ProShares, VanEck, and Bitwise. Futures products generally require greater investor understanding and carry additional expense and risk of tracking error and performance erosion due to frequent rebalancing. The fact that the underlying futures contracts are traded on the Chicago Mercantile Exchange (CME) under the regulation of the Commodity Futures Trading Commission (CFTC) is a popular theory as to the reason the futures ETFs were approved in advance of spot products.

The first filing for a US spot bitcoin ETF was made in July 2013 by the Winklevoss twins and, over subsequent years, has been repeatedly rejected. One decade later, on June 15, 2023, BlackRock, the world’s largest asset manager, made its filing for the iShares Bitcoin Trust. The weight of their brand, together with their stellar track record (according to Bloomberg Senior ETF Analyst Eric Balchunas, BlackRock has only been rejected once out of 575 ETF applications), changed the game and has helped make the US bitcoin spot ETF one of the most anticipated launches ever.

The tides shifted further on August 29th, 2023, when the US Court of Appeals for the District of Columbia Circuit, in case Grayscale v. SEC, 22–1142, ruled in Grayscale’s favor, stating that the SEC’s decision to block its proposed bitcoin ETF was “arbitrary and capricious.”

Fast forward to today, 11 issuers now have applications for a spot bitcoin ETF under review by the SEC (links to S-1 forms, which are the prospectus documents as well as additional information & exhibits gathered by the SEC, as filed on 26, 28 & 29 December 2023): BlackRock (iShares), Grayscale (conversion of the existing OTC trust), 21Shares & ARK Invest, Bitwise, VanEck, WisdomTree, Invesco Galaxy, Fidelity, Valkyrie, Hashdex, and Franklin Templeton. In recent weeks, meetings between the SEC and issuers have increased, and the SEC has required all issuers to move to cash creations, meaning that the asset exchange for the creation and redemption of ETF shares must be done in cash, as opposed to in-kind, meaning exchanging for the underlying bitcoin. Typically, the exchange of assets between the AP and the ETF issuer in creations and redemptions is done in-kind for efficiency reasons. While the SEC has not publicly stated their reasoning for requiring cash only, it is most likely that the SEC does not want to be seen to be approving APs (typically large banks and brokers) to be transacting in crypto.

As of the evening of Friday 5th January, 2024, all 11 filers had submitted their amended 19b-4s, which propose rule changes for the exchanges to enable trading of the products. These must be approved by the SEC Division of Trading and Markets.

Excerpt of US Spot Bitcoin ETF status as per Bloomberg, 5 Jan 2024

The final step would be the SEC sign-off of the final S-1 forms. Current market expectations are for this to happen circa Wednesday 10th January, and listings and trading could follow as quickly as 24–48 hours thereafter.

We will be closely following the flows of funds and volumes within the first week of trading in order to assess the competitive dynamics that will be at play across the 11 issuers. Larger ETFs are preferred by investors for a number of reasons, including cost efficiencies, liquidity, and reduced closer risk. The amount of seed capital in an ETF can therefore provide a competitive advantage. Bitwise’s S-1 filing of Dec 29th indicates initial interest of up to $200m and Blackrock’s reveals seed sales of $10m. Of note, on Jan. 5th it was rumored they may have $2bn lined up for the first week of trading, which Bloomberg Senior ETF Analyst Eric Balchunas noted as “on brand” for the asset manager given seeding of other funds, although that amount would far exceed any previous ETF launches.

Filings for US ethereum spot ETFs have also been made by BlackRock, VanEck, Ark & 21Shares, Fidelity, Hashdex, Invesco & Galaxy, and Grayscale, for which the first final SEC deadline is May 23rd, 2024.

Hong Kong

One year following the US SEC, Hong Kong’s Securities and Futures Commission (SFC) approved crypto futures ETFs in October 2022, with the first two funds, one bitcoin and one ethereum from asset manager CSOP, launching on December 16, 2022. In December 2023, the SFC and Hong Kong Monetary Authority issued a joint circular setting out guidelines for crypto investment products, noting that “in light of the latest market developments” the SFC will now accept applications for crypto spot ETFs. The updated SFC guidelines note that both in-kind and cash creation and redemption models will be permitted. Crypto ETPs issued overseas and not specifically approved by the SFC would remain available to professional investors only.

What’s next for Crypto ETPs

A growing number of investors want crypto in their portfolios, and ETPs provide a familiar, convenient, and regulated avenue to access that exposure. Driven by this demand, both crypto-native and traditional asset managers continue to engage and innovate with these products. In 2024, the expected US spot ETF approvals will likely serve as a catalyst for their growth and availability worldwide.

As this space continues to evolve, areas we’ll be watching include:

• The effects of increasing issuer competition (and the strength of that competition) on fees, as well as on flows from ETPs in other geographies and issuers, and, in the longer term, potential consolidation and/or exit of smaller players

• Shifts in both consumer and institutional awareness and acceptance of crypto, supported by the marketing power of leading global asset managers such as BlackRock

• Rise in the number of exchanges, asset managers, distributors, and other institutional players and service providers willing to work with crypto as a result

• The timeline for acceptance and integration of these products into advisory models

• Developments in institutional staking, including the growth in staking products available for investors, as well as the development of liquidity solutions for issuers, such as Validator NFTs recently launched by institutional staking provider Kiln

• The growth of onchain structured products: it’s no secret that we believe the future is onchain, and the recent partnership between 21Shares parent company 21.co and Index Coop shows how ETP issuers are starting to move in this direction

If you’re working in the crypto ETP or asset management space, please feel free to reach out to @dianacbiggs — I’d love to learn more.

Note: Over-the-counter (OTC) closed-ended crypto funds, such as those offered by Grayscale, have not been included as part of this research.

Many thanks to Frances Edwards, Jonathan Bier, Dev, Wei Dei, and Robert Koschig for your input and review of drafts.

Disclaimer: This article is for general information purposes only and should not be construed as or relied upon in any manner as investment, financial, legal, regulatory, tax, accounting, or similar advice. Under no circumstances should any material at the site be used or be construed as an offer soliciting the purchase or sale of any security, future, or other financial product or instrument. Views expressed in posts are those of the individual 1kx personnel quoted therein and are not the views of 1kx and are subject to change. The posts are not directed to any investors or potential investors, and do not constitute an offer to sell or a solicitation of an offer to buy any securities, and may not be used or relied upon in evaluating the merits of any investment. All information contained herein should be independently verified and confirmed. 1kx does not accept any liability for any loss or damage whatsoever caused in reliance upon such information. Certain information has been obtained from third-party sources. While taken from sources believed to be reliable, 1kx has not independently verified such information and makes no representations about the enduring accuracy or completeness of any information provided or its appropriateness for a given situation.

So far, users have had to install additional software, source and fund it with a novel currency, and face unfamiliar confirmation screens before even making the first interaction in web3. Despite improved firewalling and a movement away from seed phrases, these hurdles are still points of high churn for decentralized applications.

The environment has been ripe for innovations that abstract technical layers away, enabling intuitive onboarding to new financial, social, and gameplay experiences without compromising on the original ethos of self-custody and decentralization.

2023 has been a pivotal year for the wallet ecosystem, with account abstraction and developments across all layers of the stack shifting market structures and changing how we think about the relationship between users, dapps, and wallets.

This article covers:

• Account abstraction and its benefits

• An overview of wallet infrastructure and AA stack

• Emerging dapp/wallet development patterns and their implications

• Ongoing challenges and solutions to further explore

Account Abstraction: What, Why, and How

We can think of account abstraction as the decoupling of account management from key management. An account is an entity on the blockchain that can hold assets and have a transaction history. Signers (keys) are entities with the authority to perform actions on behalf of accounts.

With traditional accounts (EOAs), a private key retains sole and total control over its associated account. The strict 1-to-1 mapping between the private key and the account means that:

• Users are limited to using dedicated key management solutions (e.g. Metamask, Ledger) when interacting with the blockchain.

• There is no path of recourse from losing a private key, and the key that controls an account cannot be switched out.

• All actions originated by that private key are treated as equal, from minting a free NFT to moving millions of dollars.

Account abstraction makes the account a smart contract with its dynamic logic for what key(s) can perform actions on its behalf, scope permissions, and do additional checks and balances according to the use case.

We can further break down its benefits by examining what is being abstracted.

Because the Ethereum protocol only recognizes EOA-originated transactions, account abstraction requires an offchain infrastructure to relay smart contract-originated transactions to the chain.

ERC-4337 was introduced in 2021 as a standardized way to do this without changes to the core protocol. However, some projects had been delivering on the benefits of AA long before the standard was fully fleshed out.

• Safe* multisig wallet launched in 2017 and has grown to secure $50B+ worth of assets for DAOs, businesses, and individuals alike

• Argent’s mobile wallet has been powered by smart contract accounts since 2018

• Sequence wallet, launched in 2021, enabled Skyweaver to create and login to their smart accounts using their email and pay fees using non-native tokens

This required building and maintaining custom relaying infrastructure by the respective project.

Enter ERC-4337. The standard offers a decentralized and censorship-resistant alternative for the relay layer, defining an interface for accounts, paymasters, and signature aggregators to interact with third-party relayers via a shared alternate mempool of account-abstracted transactions (“User Operations”).

Relayers (“bundlers”) bundle multiple UserOps together into a transaction to send to a singleton EntryPoint contract, which subsequently verifies that the fees will be paid (by the account itself or through paymasters), and executes on the UserOps respective to the smart accounts.

We can contrast this with how validation and execution happen on chains that offer account abstraction natively and thus does not require extra relays (e.g. zkSync* and Starknet), as well as the recently published RIP-7560 proposal for native AA on Ethereum and its rollups.

In March 2023, the 4337 EntryPoint contract was deployed to mainnet. Its community has been immensely successful in getting developers to participate in the account abstraction movement.

This ushered in a wave of new infrastructure and service providers to the wallet ecosystem, and pushed existing projects to ensure their business strategies and product suites continue to address the needs of application developers looking to leverage AA to offer their users a seamless web3 experience.

Wallet Infrastructure & AA Stack

Signers & Key Management

Signers & Key Management infrastructure is responsible for generating and securing public key pairs used to sign messages, transactions and UserOps. The most straightforward example here are traditional EOA wallets, but wallet-as-a-service providers have emerged to enable seedless onboarding and wallet management via alternative auth methods like social and email.

Under the hood, these services either store key material in HSMs like AWS KMS which only the user can access via their auth credentials (Magic, Turnkey), or operate under some SSS/MPC scheme (Privy, Web3Auth, Portal, Capsule) to protect the materials.

Lit* improves upon this server-side key store design by decentralizing the keys. Each node in the network stores a share of a ECDSA private key generated via a DKG algorithm, all operations taking place in encrypted virtualization. Arbitrary authentication rules can be assigned to the key pair, giving the application or user complete control over what interactions are allowed, and impose for example spending limits. The network can further be leveraged by 2-of-N MPC wallets as a backup and recovery option.

This year, there has been rapid experimentation to leverage Hardware Signers and Passkeys as a signer to an account to give users key management out-of-the-box with modern mobile or desktop devices. These signers work natively with biometric authentication (e.g. FaceID, TouchID) to give additional security with familiar UX.

• Hardware Signers utilize isolated subsystems like the iPhone Secure Enclave and Android Titan HSM to generate keys and sign messages, guaranteeing hardware-level security. Because the keys cannot be extracted from the device, this is most powerful in tandem with additional recovery methods or as part of a 2FA system.

• Passkeys are a standard for passwordless authentication built on top of WebAuthn. Here, a keypair is generated in the device’s operating system, and can be synchronized across devices through services like iCloud, so recovery is possible if the user has opted in to do so.

One limitation here is that Passkeys and Hardware Signer-produced signatures are not natively recognized by chains like Bitcoin and Ethereum. They use the secp256r1 (R1) elliptic curve, while these chains operate on the K1 variation. While there is ongoing work to trustlessly and efficiently verify the R1, some Passkey-enabled products are going through services like Lit and Turnkey to produce a K1 signature once the user has authenticated with their key.

A standard to watch here is EIP-7212, which proposes to add the R1 curve directly to the EVM as a precompiled contract so every modern device can natively sign transactions without third party services or middlemen.

As the volume of account-abstracted transactions grows, signature aggregation using BLS signatures could lead to smart account fees to be cheaper than EOAs on L2s. 4337 defines an interface for Aggregator helper contracts which validates a single aggregated signature approving multiple UserOps as opposed to validating each one separately.

Relayers

Relayers (e.g. 4337 Bundlers) relay transactions or UserOps to a mempool. On chains with native AA, network operators and sequencers play this role, removing the need for external dedicated relayers.

Similar to how there are multiple client implementations for Ethereum (e.g. geth, erigon, reth), the 4337 ecosystem has  multiple bundler implementations in different languages, making the network more robust against vulnerabilities of a single implementation. The 4337 specs include a test suite to ensure bundler compatibility across the network. Implementers include Stackup (Golang), Pimlico, Biconomy, Etherspot (Typescript), Candide (Python), OKX (Java), and Alchemy (Rust).

The incentive model for bundlers is similar to block-builders, taking fees from user operations it bundles instead of transactions. In practice, bundlers need an API into the block-builder to see the current block and create a bundle that is valid against that block and thus should be considered as part of the block builder.

As 4337 traction grows, we can expect that builders will also be bundlers as this hybrid will be more profitable than a builder alone as they can select from both the transaction and UserOps pools.

Paymasters

Paymasters enable fee abstraction by allowing dapps to sponsor gas for users, allowing users to pay fees using non-native tokens, or settling offchain via traditional payment rails. Paymaster services have 2 main components:

• A gas policy manager for developers to define conditions under which they will sponsor gas. This can be scoped to the entire project, or on a per-contract or per-wallet address basis. Developers can also define how they want to rate-limit gas sponsorship e.g. by gas price, # requests, or $ amount sponsored/month. Gas sponsorship costs are usually rolled into the developers’ monthly invoice from the service provider with a ~5% surcharge for the sponsored amount.

• A paymaster smart contract validates whether a given transaction is eligible to be covered, either based on onchain state like account balances or offchain gas management policies. The paymaster contract holds a balance of native tokens that is used to pay for gas, and may contain price oracle logic that periodically checks the exchange rate between the payment token (e.g. USDC) and native token (e.g. ETH).

Paymasters can be categorized into onchain or offchain:

• Onchain Paymasters (e.g. ERC20Paymaster, StablecoinPaymaster) only rely on onchain state to verify whether a transaction is able to be covered by the paymaster or not. This means that certain paymasters, like those that accept gas payment in ERC-20s, can be made permissionless, with the caveat that the paymaster has to be approved by the account to transfer out the payment tokens. Paymaster contract admins can withdraw tokens and convert back to native to refill the contract, set a markup on top of the ERC20 price, set a threshold for the price difference for which the paymaster updates the ERC20 price for the next UserOp, or manually update the price.

• Offchain Paymasters (e.g. VerifyingPaymaster) involves interactions with a service provider’s paymaster API to sponsor the UserOp. The offchain service checks eligibility and signs off on the transaction using the paymaster’s keys. While this solution is permissioned, offchain paymasters come with the benefits of gas savings from minimizing onchain checks. Gas policies can be made more granular, and take offchain activity such as Discord activity into account.

Account Factories & Frameworks

Account Factories & Frameworks provide “headless” smart account implementations and SDKs that dapps and wallet clients can build on top of, creating self-custodial embedded accounts on behalf of their users. The accounts themselves are smart contract wallets that have their own signature validation, execution, and replay protection (nonce management) logic. Owners authorize user operations originating from their smart accounts using their key.

At a high level, smart account vendors provision 3 core things:

• A core implementation for a smart contract wallet, with its own logic for how transactions are validated, executed, and any additional actions to perform before and after the execution. It also contains logic for how additional features can be added to the wallet via native and third-party modules.

• A factory contract that deploys fresh instances of the wallet implementation, initiated with the initial signer for that account. Under ERC-4337, dapps can create smart accounts for their users by specifying the factory contract address from their provider of choice.

• An SDK that gives developers plug-and-play customizability for the smart accounts they are creating. This can include different signing options, on/off-ramping, and relaying technologies.

Under ERC-4337, the `sender` field of a UserOp refers to the smart account under which the transaction is being made. If the account hasn’t yet been deployed, the EntryPoint deploys the account from the factory contract specified in the `initCode`. User’s keys can be used to claim the smart account to make subsequent dapp interactions.

Account providers like Safe, Zerodev, and Biconomy integrate with key managers and authentication infrastructure to give dapps optionality on how they want users to manage their smart accounts. For example Safe’s Web3Auth integration enables users to use their accounts using social or email, and Zerodev’s integration with Turnkey gives the option to manage accounts with Passkeys.

Safe is most known for its battle-tested smart wallet product widely used by individuals, teams, and DAOs. To date there have been 5m+ Safes deployed across 12+ chains, executing over 22m+ transactions. Prior to v1.4.1 (launched July 2023), developers were already able to use Gelato relay to enable gas-abstracted transactions. This combination currently powers crypto debit card products like Gnosis Pay and BasedApp where users are able to purchase from any vendor which accepts Visa using funds in their Safe. v1.4.1 brings ERC-4337 support through modules to give additional optionality in relay providers.

ZeroDev is a smart account provider launched earlier this year built for ERC-4337 from the start. Zerodev aggregates multiple bundler providers to abstract UserOp relay services, and exposes a gas manager dashboard where devs can define the scope and rate-limiting logic for which to sponsor fees for users. Zerodev and Biconomy (which also runs its own bundler network) currently dominate the market share for 4337-enabled accounts.

Alchemy’s AccountKit features a 4337-compliant smart account implementation “LightAccount”, which is based off of the EF’s implementation and adds EIP-1271 support (validating signatures originating from smart contracts) along with ownership transfers, key rotation, and namespace storage.

Account Modules

Account Modules are smart contracts that act as installable components to a smart account. Although module infrastructure is still in very early stages, we envision that modules will be discoverable and installable by:

• Developers: embedded smart accounts can come with “pre-installed” modules as decided on by the dapp developer, constructing a starter wallet with features customized for their use case

• End users: wallet interfaces can expose a “module store” where users can discover new features and add them to their wallet

With AA formally separating how UserOp validation and execution is handled, modules can contain validation-only or execution-only logic.

• Validators. Called during the validation phase of a UserOperation.Their primary function is to verify the signature of a UserOperation and determine whether it is valid and should be executed. Examples include multisig, ECDSA, passkeys, multichain validation, and session keys. Session keys enable dapps to sign on behalf of the user to simplify UX, behaving like temporary private keys with customizable permissions and expiry time.

• Executors. Called during the execution phase of a UserOperation. They extend the execution logic of the account and allow for a more diverse set of actions that it can natively perform. Examples include automated actions that are triggered outside of the regular ERC-4337 execution flow such as automatic token swaps when the price reaches a certain threshold.

• Hooks. Run pre- or post-execution and enforce controls against the account. For example, a hook could run post-execution and revert any transactions that meet certain criteria to create increased security for the user.

While some wallets like Candide have developed modules their users can install directly, we anticipate a rich ecosystem of third-party modules discoverable in an app-store like interface of their wallets, or composed into a “starter” embedded wallet by dapp developers.

Smart account frameworks are already designed with modules in mind. Safe’s core contract handles module management logic for adding and removing modules from the account, but leaves actual module-related logic and storage completely scoped to a separate contract. This separation of concerns mitigates the risk of third party modules overwriting the same state, compromising the security and expected behavior of the account.

Safe{Core} Protocol introduces an open framework with modules, hooks, a manager, and registries aimed to foster an composable smart accounts ecosystem inspired by the learnings from Safe’s wallet product.

ZeroDev explicitly classifies their modules (“plugins”) as validation or execution. Executor modules are designed to be paired with validator modules, allowing custom functions to be “routed” through different validators. For example, an “NFT transfer” function that only allows NFTs to be transferred via 2FA.

Some considerations in building towards a robust ecosystem of modular smart accounts:

Interoperability. With multiple smart account vendors each with their own approach to how third parties can add new features to accounts, module developers are on a trajectory towards vendor lock-in or having to manage the technical overhead of developing the same features to be compliant with multiple account implementations. Some solutions to mitigate this:

• ERC-6900 for modular smart contract accounts and plugins defines interfaces for modular smart contract accounts (MSCAs), modules (“plugins) allowing any standard-compliant account implementations and plugins to be interoperable with each other.

• Rhinestone’s* ModuleKit for building and testing smart account modules provides templates and frameworks for testing modules against different account implementations, a library of integrations (e.g. DeFi protocols),  pre-built conditions for execution, and security automation to parse through module code and flag security vulnerabilities.

Security. Giving users the ability to install third-party software into their wallets comes with important questions for how interfaces should curate and expose relevant information about modules.

• EIP-7484 provides a means of verifying the legitimacy and security of independently built smart account modules. Here, a registry allows auditors to make attestations about the security of those modules. Frontends and smart accounts can query the registry to pull in attestation data, verifying that a module is safe to use. See the Rhinestone registry for a reference implementation of this.

• More broadly, EIP-7512 aims to create a standard for an onchain representation of audit reports parsable by smart contracts to extract relevant information around who performed the audits and what standards have been verified.

Discoverability. Registries can be exposed and queried by smart account platforms and wallet interfaces to be installed by developers or end users.

The ability to extend wallet functionality turns accounts into developer platforms and new distribution channels for web3 products and services. We are already seeing this with Metamask Snaps which allows users to customize their browser extension wallet with security alerts (through WalletGuard), privacy features (through Nocturne), and interoperability with non-EVM chains such as Starknet and Bitcoin.

When Chrome enabled an ecosystem of developers to extend browser functionality through extensions, this propelled them to market dominance over incumbent walled gardens despite being a decade younger. Although it’s hard for most of us to picture what the wallet experience might look like when modular accounts become mainstream, the tracks are already being laid for permissionless innovation to take its course.

Emerging Development Patterns & Their Implications

The evolved wallet stack means that:

• Developers can create non-custodial accounts for their users in the context of their apps, and will look to tools like connector SDKs to give them optionality in how to construct the end-to-end onboarding journey.

• Embedded wallets are a new wallet category, each vendor with its own features and tradeoffs in terms of account portability, customizability, and trust assumptions.

If you played around with Ethereum dapps in 2018, you may recall getting a Metamask popup as soon as you load the site. This was due to a lack of good UX practices around connecting wallets and dapps, and developers often resorted to hardcoding checks to see if a user has an extension wallet installed using the browser’s `window.ethereum` object. This resulted in unpredictable behavior if users had multiple extension wallets installed, forcing users to choose one and creating a less competitive market for wallets.

WalletConnect* communications protocol emerged to let users connect any wallet to any dapp, alongside with Web3Modal, a library that wraps buttons and modal components to let users choose which wallet they wanted to use the dapp with.

Today, Web3Modal is one of multiple wallet connector libraries like RainbowKit, Web3Onboard, and ConnectKit that streamlines the wallet detection and wallet-based authentication process for dapp developers. These libraries offer out-of-the-box theming options, wallet search features, and screens that redirect users to install wallets if they don’t already have one.

More recently, EIP-6963 was finalized as an alternative wallet discovery mechanism to `window.ethereum`, allowing dapps and injected scripts provided by extensions to communicate in a predictable manner. Thanks to the standard, users now have more optionality in choosing an extension wallet of choice to connect to dapps, and opens a more competitive market for wallets.

While connector libraries have significantly improved DevEx and UX,, the expectation that users already have or will install additional software to interact with dapps still poses a high barrier to adoption.

We are already seeing a glimpse of what the future of dapp onboarding UX looks like, as the next generation of wallet libraries, which we will call here fully fledged “onboarding SDKs” gain steam. In addition to wallet-based auth, these SDKs give alternative sign-up and login options such as email, socials, SMS, and creates embedded wallets for users without requiring them to install additional software or navigate away from the dapp.

Developers can integrate connectors offered by key providers directly (e.g. Magic, Privy, Web3Auth) or use those that wrap multiple services (e.g. Dynamic, Thirdweb, 0xPass, Web3Modal) to give plug-and-play optionality on the authentication options they want to expose and the type of wallet they want to create, completely customizing the onboarding flow. Onboarding SDKs can also integrate with smart account providers to create embedded smart wallets, offering further UX enhancements like gasless transactions and on/off-ramps.

With increased adoption of “headless” wallets and a movement towards embedded wallets, what used to be a clear line of separation between wallets and dapps is beginning to blur.

Standalone Wallets vs. Embedded Wallets vs. Application-Specific Wallets

Web3 users today are used to interacting with dapps through standalone wallets like Metamask or those offered through WalletConnect, accruing their assets and onchain footprint to one or a few accounts.

As more dapps look to onboard users through embedded wallets and multiple vendors available, account management quickly becomes complicated for both dapp developers and end users. Dapp developers will have to evaluate and manage multiple vendors while trying to avoid lock-in. For end users, creating a new wallet per dapp will lead to a fractured asset and identity management experience.

While app-specific wallets are desirable for certain use-cases such as games, there are many instances where users might onboard to their first dapp, create an embedded wallet with their web2 signers or Passkey, and want to use the assets they accrue in that account on another dapp by logging in with the same signer.

• Capsule is a MPC-based embedded wallet provider that features wallet portability across dapps that use their service, giving users access to an existing wallet using the same email login. We can anticipate this to soon be a table-stakes offering across WaaS providers.

• Moonchute helps users manage multiple smart accounts in the interim. Their unified account manager is an app and API for discovering smart wallets created by a given signer, allowing users to manage assets from multiple accounts in one place.

• ERC-7555 proposes a SSO-inspired standardized interface and request/response pattern for applications to discover user accounts created using alternate signing schemes. Here, the application would redirect the user to a given provider’s URI (which could be a self-hosted domain), and parse the response for a signer and associated smart account address.

Migrating from EOAs

Another outstanding challenge of AA is a seamless way for existing users, who already have assets and onchain history accrued on multiple EOAs, to migrate to smart accounts.

• EIP-7377 proposes an in-protocol mechanism to allow EOAs to send a one-time transaction which deploys code at their account, effectively “upgrading” an EOA to a smart wallet.

• Aarc aims to solve asset migration for dapps and end users. Their UI and SDK indexes the assets and permissions of a particular source address, and allows users to select the assets that they want moved to any destination address, which could be a smart account, another EOA, or embedded MPC wallet created with social login. For dapps with existing users who are used to the standalone wallet flow, Aarc offers a solution to streamline the migration process as they look to add embedded wallets or AA features to their product.

AA implications for multichain account management

Given the momentum of AA and L2 activity, we can anticipate a future where smart accounts become mainstream over EOAs with users having assets across multiple chains.

One UX advantage of EOAs is that users automatically have access to the same address across different EVM chains using the same private key. The drawback being that it is not possible to change the keys that control a given address, and all funds can be lost if the user loses their private key.

Because account abstraction separates the key store from the asset store, it is possible to rotate keys for a given account without having to migrate funds while keeping the same address. Smart accounts are capable of maintaining the same address across chains using CREATE2, giving users access to the account even if the contract hasn’t been deployed on a given chain yet using the same initial verification key and account implementation.

However, maintaining the same address across chains may be an anti-pattern in the long term.

• CREATE2 is only possible on chains with EVM bytecode equivalence. In a multichain world with zk-Rollups (e.g. zkSync) with slight deviations to the EVM, this approach will not suffice.

• We can expect that keys needed to access various accounts are going to change over time as wallets expose additional signing and key rotation features. Under the current setup, this can quickly cause state drift of account permissions across chains, as the changes to signers on an account on one chain doesn’t automatically propagate the new permissions to those on other chains.

Longer term solutions that have been proposed for multichain AA include:

• A dedicated key store contract that a user account across chains read from when checking permissions. See an implementation of this from Soul Wallet.

• Using ENS multichain resolvers as an abstraction layer for different addresses.

Cross-chain account and signer management is still an area under active research. The end goal here is that a user can change the keys that have access to multiple accounts on different chains without making an extremely high number of transactions.

Conclusion & Closing Thoughts

• Account abstraction is a movement to decouple signers from accounts by making contract-based accounts (instead of EOAs) as first-class entities on the blockchain, giving users flexibility in key management and account permissioning.

• ERC-4337 as a standard for relaying smart account-initiated transactions has catalyzed the evolution of wallet infrastructure to accommodate AA, giving rise to new market structures, wallet categories, dapp development and user onboarding patterns.

• The wallet stack has been unbundled into signers, relayers, account providers and account modules giving developers optionality in how they want to customize the end user experience. This comes with the additional overhead of evaluating each provider on their tradeoffs in terms of gas overhead, censorship resistance, vendor lock-in, and interoperability.

• Migration paths from EOAs and account abstraction in a multichain context are still areas of ongoing research. We expect to see the first implementations of proposed solutions in the coming year.

We believe these developments have significant implications across the ecosystem:

• For new users, wallets are no longer the sole entry point into web3. Applications will have markedly improved onboarding through embedded wallets, gasless transactions, and in-wallet onramps.

• For current users, onchain experiences will become more seamless as apps leverage features like session keys. Power users have more granular control over account permissions and additional wallet features through modules.

• For developers, modular account infrastructure turns wallets into operating systems. Module markets are a new permissionless distribution channel for web3 products and services.

Wallet infrastructure will continue to catalyze web3 adoption. We at 1kx are proud to have backed teams who have pioneered on the ideas discussed in this article, and will continue to monitor the space for what’s to come.

If you are working on solutions in this space or have additional thoughts on the topic, please reach out @nichanank - would love to chat with you.

Many thanks to David Sneider, John Rising, Konrad Kopp, Kurt Larsen, Marc Sednaoui, Dogan Alparslan, Vivian Phung, Derek Rein, Tom Terado, Diana Biggs, Mel Quarto, and pet3rpan for reviewing drafts of this.

*denotes 1kx portfolio companies

Alternative social media platforms built on decentralized and federated protocols offer a return to the original conception of online sociality. Data is controlled by the user and is propagated across decentralized databases, frontends are community-driven, moderation is an expression of a community’s preferences, algorithms are chosen by the user, and an open source ethos drives innovation.

History of Decentralized and Alternative Social Media

Before the web became a hub for commerce, entertainment, and social interactions, it was primarily an academic and military tool. Tim Berners-Lee held an egalitarian vision when he formulated the first web protocols - the initial design of the internet was to be a decentralized network, where information could flow freely between nodes without any single point of control or failure.

However, as the web grew in commercial importance, centralized platforms, like search engines and social media giants, emerged as dominant players. While these entities provided significant value, they deviated from the original decentralized ethos, resulting in our current web2 environment.

The key innovation in the timeline for alternative social networks was the arrival of the concept of federated protocols. A federated network refers to a system where multiple independent servers or "nodes" cooperate to form a single social network, as opposed to centralized platforms where a single organization controls all the servers.

In a federated system, each server runs compatible software that follows shared protocols, enabling them to communicate with each other. A user registered on one server can follow, interact with, and share content with users from other servers seamlessly, as if they were on the same platform. Examples of these protocols include ActivityPub and OStatus, which power federated platforms like Mastodon and PeerTube.

In a federated setup, users can choose which server they trust, potentially migrating to different ones or setting up their own, giving them more autonomy. The term "Fediverse'' - a portmanteau of "federation" and "universe" - is used to describe such a system. The Fediverse started with platforms like GNU social and its predecessors (StatusNet and Laconica), but the real turning point was the development and widespread adoption of the ActivityPub protocol, which was published as a recommended standard by the World Wide Web Consortium (W3C) in 2018.

Within web3, federated social networks are the default state of decentralized systems once data is ported onchain. Blockchains act as an unopinionated backend server for stored content, with frontends indexing this content and serving it directly to users. Identity is handled by public-private keypairs that already manage user wallets, allowing them to easily authenticate any data or content they generate. Additionally, use of onchain primitives like NFTs can bundle stored content in the metadata and act as a domain name or decentralized identifier (DID).

Similar to how ActivityPub works, web3 protocols seek to bootstrap social graphs via authenticated relationships between user nodes. Since any frontend can index and serve this content, there is hyper competition at the frontend layer resulting in a flourishing landscape of features. Additionally, because data is onchain, users can elect the algorithms they are comfortable using, and they can be incentivized to use certain ones, recapturing the value of their data. This is combined with more direct means of monetizing their content, making a better overall experience for creators which have been largely left out of monetization even though their content is what drives demand for these platforms.

Protocol Comparison

To truly appreciate the innovations within decentralized social media protocols, it’s necessary to understand the technical nuances that enable them. Notably, we are not including all social protocols here, but instead opt for some of the most prevalent:

Identity / Namespaces

In the context of federated and decentralized social graphs or network protocols, a "namespace" refers to a domain or realm under which user identifiers or other resources are unique. It is a way to distinguish resources or identities from one domain/server from another, ensuring that there aren't conflicts or ambiguities when integrating or communicating across multiple domains.

Identity and associated namespaces across decentralized social protocols run the gamut from simple keypairs (Nostr, Scuttlebutt) to URIs pointed at HTTPs URLs hosting profiles (ActivityPub) to more complex models which use onchain primitives like NFTs (and more recently ERC-6551 extensions e.g. Lens v2).

Farcaster is a great example of these techniques. A Farcaster account represents a distinct entity on the network. Each account has a unique numeric identifier called a Farcaster ID (“fid”). Identities are issued and managed onchain using an Ethereum contract called the IdRegistry. Users make a transaction to the IdRegistry to get a new fid. The address that owns the fid is the user's custody address. The IdRegistry ensures that fids can be transferred between addresses and that no two addresses have the same fid. Farcaster also extends this namespace to support ENS names that are issued onchain or offchain. A signed proof must be submitted to the network to claim a username.

ActivityPub, on the other hand, identifies every user by a unique URI, typically an HTTPS URL. This URI points to the user's profile and serves as their global identifier in the Fediverse. To make these URIs more user-friendly, many ActivityPub platforms use a system called Webfinger. Webfinger allows users to have an identity like ‘@username@domain.com’.

Lens and CyberConnect instead manage user profiles as NFTs.  In the case of Lens, a user address holds a ProfileNFT, and it is possible for a single address to hold multiple ProfileNFTs. Each Profile NFT encapsulates the entire history of a user's activity, including posts, mirrors, comments, and other types of content that they have created. Additionally, Profile NFTs have a FollowModule, which is essentially a set of rules that governs how different accounts can obtain Follow NFTs. These Follow NFTs serve to document the connections between accounts and the primary profile directly onchain. There are also handles that can exist and can be minted separately from profiles, and can be linked and unlinked from one profile to another. Handles exist in their own namespaces (e.g. lens/@alice).

Data

Data is arguably the most important feature of decentralized networks, since its creation and standardization is what enables these systems. The most common technique for managing data here is the use of standardized formats like JSON and common relationship objects (e.g. likes, follows). Core data objects typically include:

• Actors & Objects: Defined "actors" (eg. users or groups) and "objects" (eg. posts or messages).

• Publications: Posts or comments are encapsulated as "Publications," often linked to external content via URLs.

• Content in Append-Only Logs: Logs where each entry, be it a post or update, is a discrete content item, sequentially added and stored.

Let’s dive into a couple examples of how this works using specific protocols.

ActivityPub utilizes the ActivityStreams 2.0 data format, a JSON-based structure, to represent various social interactions such as posts or likes. The protocol differentiates between two primary components: Client to Server (C2S) and Server to Server (S2S). C2S allows users, through client applications, to interact with their respective servers. In contrast, S2S facilitates communication between servers, enabling the robust federated nature of the protocol.

Within ActivityPub, entities are categorized as "actors" (often user accounts or groups) and "objects" (content or actions like posts or likes). When an actor performs an action on an object, it creates an "activity" such as "Create", "Follow", or "Like".

Web3 social graphs take many of the core ideas of ActivityPub but apply them onchain. Lens Protocol, for example, introduces "Publications," which encapsulate a variety of user-generated content such as posts, mirrors, comments, and other forms of media. Each Publication is associated with a ContentURI, directing to the specific content stored on a decentralized protocol like IPFS or Arweave, or alternatively, on a centralized storage service such as AWS S3. This configuration ensures that a user’s profile and all associated publications are securely stored in their personal wallet, moving away from reliance on centralized databases.

Furthermore, Web3 enables a more straightforward approach to monetizing user content and influence compared to the Web2 framework. Users can charge for the minting of Follow NFTs, or they can integrate Collect Modules with their Publications. This latter option allows them to receive a fee for minting of NFTs linked to the ContentURI of their publication. In addition to these features, Lens Protocol offers a GraphQL API, serving to mask the blockchain components from frontend interfaces and consequently, delivering a more user-friendly experience than previous decentralized social networking attempts.

Ultimately, many of the decentralized social networking protocols create append-only data structures which are authenticated by user keys. For example, on CyberConnect each piece of user-centric data is represented as a data stream where updates are only allowed by the data owner. Each update to the data is appended to the data stream in the form of an append-only commit log and the resulting data structure becomes a hash linked data structure called a Merkle DAG. Data types include content, collects, comments, and subscriptions.

Scuttlebutt similarly uses an append-only log. Each user has their own log where every new message or action is appended to the end after being signed by the user identity (i.e. the associated Ed25519 key pair). It also supports the sharing of binary data, referred to as "blobs." These can be images, videos, or any other binary content. Blobs are stored separately from the append-only logs, but references (hashes) to these blobs can be included in the logs.

For Farcaster, messages are public updates like making a post, following someone, or adding a profile picture, and these messages are encoded as a protobuf and must be hashed and signed by the account's signer. Users can publish messages to Hubs as long as they have sufficient storage. Hubs check the validity of each message's signers before accepting them.

Storage

Early approaches to data storage for decentralized protocols were largely offchain, although reminiscent of onchain consensus. Scuttlebutt, for example, uses a peer-to-peer gossip network, placing the onus of storage on the user’s local device. This approach ensures data sovereignty, as users have complete control over their own information. However, it also means that data availability is dependent on the user’s device being online or other peers in the network having a copy of the data. Over time, to manage storage space, some Scuttlebutt clients might need to implement garbage collection strategies to prune old or less relevant data.

An alternative to this peer-to-peer approach comes in the form of servers storing data, although with redundancy compared to traditional media platforms. Matrix as an example has multiple homeservers store copies of room histories and synchronize with each other. When a user sends a message (or any event) in a room, their homeserver broadcasts that event to other participating homeservers, which then store and forward the event to their connected clients. Similarly, ActivityPub has each instance (or server) in the network store its data, typically in a database. The choice of database (relational, NoSQL, etc.) depends on the specific implementation of the ActivityPub software. For instance, Mastodon, a popular ActivityPub platform, uses a PostgreSQL database.

Protocols like Cyberconnect, Farcaster, and Lens have embraced blockchains for storage. The use of on-chain storage ensures that data is immutable and verifiable, providing a robust foundation for decentralized applications using underlying consensus mechanisms for syncing state. However, this approach can lead to scalability challenges, as every piece of data needs to be stored onchain, potentially leading to high transaction fees and slower retrieval times.

This has led many web3 social protocols to experiment with hybrid approaches that use onchain storage for less frequent actions (e.g. profiles, subscriptions) and offchain storage for high frequency events (e.g. likes, reposts, comments) or batch-uploading data onchain at frequent intervals with offchain storage used as an interim stopgap.

CyberConnect, in order to efficiently handle the frequent updates between user connections, employs a hash-linked list in a decentralized data store. On initiating a connection, an 'operation log' is created. Subsequent state changes, like toggling between following and unfollowing, are added as new nodes to this log. While these updates are initially stored on a central server, they are periodically batch-uploaded to decentralized storage platforms such as Arweave or IPFS. For quick data retrieval, the nodes in the operation log are stored centrally. However, users can independently verify data integrity by navigating through this hash-linked list. Even with the dependence on a central server for some data queries, CyberConnect's system is designed to be sufficiently decentralized while also offering high performance.

Farcaster similarly uses a hybrid approach: onchain contracts are used for infrequent actions where consistency and decentralization are important. Accounts, usernames, storage, and keys are managed using a series of Ethereum contracts. Offchain systems are used for frequent actions where performance is critical. Messages created by user accounts are stored and propagated on the peer-to-peer network of Farcaster hubs.

Discussion

Decentralized social protocols are poised to revolutionize the user experience in digital interactions. The accelerated adoption of public-private keypairs, propelled by both web3 traction and as a proactive measure against AI-generated content, will facilitate a broader understanding and familiarity with identity primitives in this context, and continued moderation and data capture at web2 social media companies will openly push more users to look elsewhere. We expect an accelerated adoption curve for these protocols.

To foster the evolution of novel applications, there is a pressing need for protocol developers and open-source contributors to venture beyond the basic data types and relationship objects currently employed at the infrastructure layer. While the existing primitives adequately encapsulate the functionalities of conventional web2 social media, there is immense potential for expansion and innovation. The majority of protocols under discussion here inherently support extensibility within their systems, providing a solid foundation for future developments and open-source contribution.

However, it is crucial to underline the importance of interoperability. While frontend developers are capable of augmenting functionality independently, doing so might detract from the collective benefits of the system if the enhancements are not interoperable with other applications built upon the same underlying protocol. Ensuring compatibility and seamless integration across various applications is vital for the long-term success and adoption of decentralized social protocols.

In the realm of data storage, the emerging consensus within web3 social protocols leans towards a hybrid approach. Given the high volume of social content and engagement, it is pragmatic to allocate high-value assets such as identity and primary content to on-chain primitives, while relegating lower-risk content like likes and reactions to off-chain solutions. This balanced approach not only preserves the integrity and security of crucial data but also provides a user experience reminiscent of traditional social media platforms.

Decentralized networks hold the promise of transforming interpersonal communication, information sharing, and community building. By prioritizing user autonomy, privacy, and fostering organic relationships, these networks are paving the way for a more equitable and user-centric digital landscape. Moreover, the decentralized nature of these networks serves to democratize access to information and resources, mitigating the risks associated with centralized control.

If you are building in decentralized or federated social media, please reach out to us.

Dynamism brings NFTs beyond static links and media, to behave more like software that responds to external factors. This creates new layers of interactivity around digital goods and media, enabling greater individual and collective expression, dynamic utility, and ongoing innovation around digital objects.

Dynamic NFTs can either be programmatic (reflecting algorithmic input) or interactive (reflecting user input). Dynamism is a feature that can cut across verticals depending on the intended use case.

We’ve witnessed a range of experiments across art, gaming, identity and reputation, the metaverse, as well as community and brand engagement.

Performative or conceptual art

Dynamic collections could be seen as a subset of generative art, which also includes “Parameter Input” multiplayer art where the mint time or minter address is used as a source of entropy, but the NFT itself does not update after mint. Dynamic art NFTs give collectors an ongoing sense of delight, and serve as a tool for multiplayer creation between artists and collectors as a new medium for collective storytelling.

• Art that automatically evolves with time, lunar cycles, onchain state, or offchain conditions e.g. 720 Minutes by Alexis Andre, BURNER by crashblossom, Gaussian Timepieces by Takens Theorem, Finiliars by Ed Forneiles, Gazers by Matt Kane, Mutant Garden Seeder by Harm van den Dorpel

• Art that evolves with transfers chains and ownership e.g. Animal Coloring Book,, Corruptions by dom.eth, Chameleon by Joan Heemskerk, Entropes, OG Crystals, Watchfaces by w1nter.eth and Tyler Anglert

• Art whose visuals collectors can directly influence e.g. Brotchain by divergence, Terraforms by Mathcastles, Shields by John Palmer, Async Art’s Forever Supper or Classic where collectors can change the layers that are displayed in the main artwork

• Art with style options released periodically, with each release featuring new limited edition styles holders can opt to “convert” their NFT to e.g. Opepen

• Real-time generative, never repeating audiovisual art e.g. In Noise We Trust by 404.eth, and various works by DEAFBEEF

Games

Besides serving as upgradeable assets in games, dynamic NFTs could serve as the canvas for gameplay, reflecting game state in its media and metadata. Coupled with digiphysical goods, the NFT could update based on IRL activity to unlock new consumer experiences.

• Axies that win ranked battles earn Axie-bound Experience Points (AXP), which can be used to “Ascend” Axies. Doing so syncs offchain game progress onchain, raises the Axie’s level cap, and allows players to upgrade Axie parts.

• Citadel Ships can be upgraded to increase its power, speed, and fuel efficiency. Each upgrade tier takes a different amount of time and raw resources (Ore) to complete. Part of the gameplay is for Pilots to plan their travels to asteroid belts to mine Ore, with each action and game state reflected onchain

• Realm players earn $ANIMA tokens through staking and gameplay. These rewards can then be used to upgrade their Adventurers (character NFTs from Realm’s partner projects) and receive LootBoxes that contain rare game items.

• “Moves” are onchain transactions that affect a world e.g. Straylight

• Battle Arenas where gameplay leaves a mark on the media e.g. Chainfaces Arena

• Spreading infection as a virality/distribution mechanism e.g. FoliaVirus, Viper

• Collecting, breeding, and merging mechanics to produce rarer editions e.g. Avastars, VV Checks

• Self-playing games using onchain AI where collectors can seek the highest scoring NFTs e.g. ArcadeGlyphs by Miragenesi

• IRL gameplay and physical activity upgrades the NFT e.g. STEPN, Loot LARP

• Mint scores as NFTs where the depicted rank changes as new records are broken e.g. Onchainscores, Optimizor, 1kx Score NFTs can be minted from our Play to Learn page

• Avatar rarity hunting and community participation e.g. Manny’s Game

• NFTs eat, infect, boost each other to get more powerful e.g. Etholvants & Booster Syringes

Identity & Reputation

Dynamic NFTs can also proxy some identity and associated reputation within a community, evolving with ongoing contributions and governance activity. This fosters intentional interactions within and between communities, and serves as a foundation for social games to be built on top.

• Onchain activity across applications directly influence NFT traits e.g. Zerion DNA. Philand cities depict wallet activity, and each in-city asset earnable through quests

• Lens v2 includes out-of-the-box support for the tokenbound (ERC-6551) standard, giving each Lens Profiles NFTs their own smart accounts. This decouples the Profile from its holder, accruing access rights, assets, and reputation to the NFT itself.

• Station

• Social Contracts by Burak Arikan trace their collectors’ holdings and their shared connections to other collectors, generating a collection graph to project future acquisitions

• JPG Canonicons artwork represents a unique, cumulative representation of an individual’s involvement in the curation of Canons. Similarly, Deca Decagons level up by consuming Deca Experience Points (DXP) individuals earn by completing daily Quests on the platform

• Mercle and The Metagame avatars enable gamification and trait unlocks based on actions and roles individuals take on within a community

Metaverse

• Land that holders can directly influence and overwrite media onchain e.g. Terraforms by Mathcastles

• Land that evolves as holders and visitors interact with it via art curation e.g. MOCA ROOMS, or land builds and gameplay e.g. Upstreet, Hyperfy, Otherdeeds, Voxels

• Virtual fashion NFTs that can be converted between 3d wearable and artwork e.g RSTLSS

Community & Brand Engagement

Large brands have experimented with dynamic NFTs to engage with a mainstream audience, connecting the asset with external data or adding utility through ongoing campaigns.

• DeGods unveils new artwork and traits to its PFP collection every season. Holders spend $DUST to upgrade their NFTs, and can choose which metadata they want displayed. Staking DeGods accrues more $DUST to the holder and DePoints to the staked NFT. DePoints can be spent in-game to open Packs that contain $DUST or brand-sponsored rewards, with a leaderboard of DeGods that earned the most points for the current season.

• Smolverse holders can stake their characters and earn IQ or Plates (depending on the character), which represents the time spent in the community and dynamically upgrade the characters’ head and body size. IQ can also be spent on aesthetic items, which are then added to their traits onchain. They will soon launch equippable ERC-1155’s to replace on-chain existing traits rather than just use them a skin, allowing the community to evolve the official collection collectively with the team.

• LaMelo Ball collectible sports cards have ‘level up’ capabilities based on the outcome of real life basketball data, like a Rookie of the Year announcement. As the player’s career evolves, so will the collection with dynamic integrations to stats and milestones

• Adidas ATLS are PFPs in Adidas’ Into the Metaverse collection. Over time, new chapters are revealed as part of an interactive storyline introducing new traits to the token

• Lacoste UNDW3 tracks community participation in "The Mission", a new experience revolving around conversation, gamification, quest-solving, and co-creation

How are Dynamic NFTs evolving today?

Passive - updates autonomously, objective

• Time: 720 Minutes by Alexis Andre, Gaussian Timepieces by Takens Theorem

• Lunar Cycles: Gazers by Matt Kane, Moon in Motion by Miragenesi

• Block Hash: Mutant Garden Seeder by Harm van den Dorpel mutates with current block hash similarity to seed block hash, Chaos Roads by Chainleft

• Contract State: Uni v3 Positions, PartyDAO Membership Cards. pepethereum reacts to fluctuating ETH price using data from Uniswap pools

• Offchain data/Oracle-based

  • Crypto and gas prices e.g. Finiliar, BURNER by crashblossom

  • Weather based e.g. Impermanence by Josh Pierce

• Owner/transfer-based: Animal Coloring Book, Watchfaces, OG Crystals, Corruptions

Interactive - direct influence from owner, subjective

• Minting/Burning/Combining e.g. VV Checks, etholvants, merge, Smols Transmolgrification

• Appending/Updating content, traits, layers e.g. DeGods, Moonbirds, Terraforms, Upstreet, Forever Supper, Smols & Swols

• Opt-in Reveals e.g. Opepen

• Gameplay e.g. Axie Infinity, Chainlife,  0xEssentials

• Community Participation e.g. JPG Canonicons, Lacoste UNDW3, Collective Strangers Cameras

• Time-sensitive “nurturing” e.g. Bonsai, lifeforms by Sarah Friend

• Interactions with other NFTs and wallets e.g. Entropes & Spells, Etholvants & Booster Syringes, Viper. SABOTAGE by Kim Asendorf features a 1/1 “Editor” NFT whose holder can alter the artwork of the collection.

What is being changed?

Content

NFTs whose media renders purely from onchain data evolve autonomously based parameters such as time, block hash, wallet address, which deterministically influences the artwork. Outside of onchain art and games which often require bespoke implementation per collection, NFTs with media reflecting onchain state can be used as commemorative or “receipt” NFTs to represent things like a DeFi position (Uni v3 Positions, Web3 Savings Cards), governance rights (PartyDAO Membership Cards, Juicebox Cards), or community membership card with some accrued reputation (JPG Canonicons, Deca).

Tokenizing protocol positions into NFTs makes these positions themselves tradeable as assets, turning them into building blocks for new products and services. For example, Metastreet deposits third party NFT promissory notes into a collateral pool for their Automated Tranche Maker protocol.

Having the media reflect live contract state makes contract state readable to end users of wallets and marketplaces. The skeleton SVG code is deployed once as part of the NFT contract, and the rest of the image is programmatically updated with onchain data.

Metadata

Game items seem a natural fit for mutable metadata as players can evolve or power up assets through gameplay. The advantage of doing so through an NFT vs. virtual items in a centralized server is that the metadata can be tracked and stored on decentralized infrastructure, so the record of play persists as the item gets traded.

Supply

Mint and burn mechanics that elevates a subclass of assets into a higher "tier" produces new assets that can only derive from certain combinations of the lower tier (e.g. VV Checks) or inherit traits from its parents (e.g. CryptoKitties, Avastars).

Neolastics and Clovers are generative art projects that introduced dynamic supply and autonomous economics into the collection. For Neolastics, anyone can mint new tokens and inflate the supply with each subsequent mint price increasing according to a bonding curve. 99.5% of each mint cost goes into a community reserve that acts as predictable floor and ready buyer when anyone chooses to burn their Neolastic.

Dynamic supply mechanics can also be introduced via sinks and faucets in games. In Citadel, new ships are brought into the game through the ship Construction Bay via a weekly sealed-bid (blind) dutch auction system. Ships can be destroyed as players travel to more dangerous regions in the game, risking ship destruction for greater rewards. Inflation rates are within a narrow range of new ship quantities offered at auction, while ship destruction will tend to be a percentage based burn of the total supply. As total supply increases, total ship burns will increase until they are relatively equal to the number of new ships introduced at auction.

Onchain functionality

Interesting games can be designed where certain abilities or functions on the NFT contract remain dormant until specific conditions are met. For example, a dynamic collection in which the supply initially only decreases by merging two tokens together could include a function called “birth” that is only callable on NFTs that have reached a certain size by merging (reflected in the contract state). The generation and parent tokenID can be used as the seed for a deterministic renderer to create infinite, visually distinct rounds of a game like VV Checks.

NFTs could also be made changeable up to a period of time before the NFT is "locked" and becomes immutable.

How can creators make Dynamic NFT collections today?

Dynamic Art platforms

• Async Art and OG Protocol have introduced novel mechanics like Master/Layer NFTs and metadata update pipelines specifically to help artists launch multiplayer collections and dynamic NFT projects.

• Transient Labs has been spearheading experiments with interactive and updateable media in art. In ERC-721TL, they outline a way for both the creator and the collector to have a say in metadata updates, allowing collectors to approve or deny any proposed metadata changes by the artist. The standard also offers Story Inscriptions, enabling both creators and collectors to leave their narratives, exhibition provenance, and sales history on an artwork by appending onchain  text  to the NFT. Hidden Stories by Michelle Viljoen is a showcase of this.

Oracles, Decentralized Cloud Functions, and Specialized Middleware

These can be used to feed offchain data to smart contracts, allowing NFTs to respond to offchain events. For more complex interactions such as UGC in virtual land, the NFT represents in-world coordinates and write access. But the content itself is typically processed and stored in hosted servers which the token refers to in its metadata.

• Chainlink Data Feeds can deliver information such as crypto prices and weather to make an NFT “environment aware”. Through their integration with Space & Time, developers can execute a script that queries a game server and push it onchain using Chainlink Functions.

• Using Gelato Web3 Functions or Lit Actions, developers can write any logic to conditionally update NFTs based on offchain events or at specific intervals. Gelato nodes continuously run functions and can trigger metadata changes if specific conditions are met. Similarly, Lit Actions are executed across Lit’s threshold cryptography network, each node independently verifying the result and signing the transaction when it reaches a 2 ⁄ 3 threshold to automatically make onchain updates. For example, fetching live match data from a sports API and upgrading an athlete NFT’s skill traits as they win games.

• Mentaport provides an SDK for “location-aware” smart contracts, enabling time and location-based feature access, mints, and dynamic updates which could be useful for limited time IRL campaigns and proof-of-visit mechanics.

Integrated NFT management platforms

Sparkblox, Evalon, and Metafuse offer all-in-one solutions to launch and manage interactive NFT collections that allow holders to engage with the assets, or dynamic NFTs that evolve based on real world data and connected APIs.

• Kairos provides a GraphQL API to programmatically create, mint, sell NFTs and optionally with updateable metadata, as well as a Dynamic NFT developer tool with metadata and image held in Kairos servers.

• Paima allows game developers to create stateful NFTs that can level up, gain experience, obtain equipment, and develop over time. Paima’s NFT compression protocol mints a minimal set of NFTs on the L1 and evolves them based on the game state on an L2.

• Lync provides out of the box tooling for game developers to integrate web3, including a cross-chain wallet SDK, marketplace SDK, and NFT management tools where updates to in-game assets can be triggered using Chainlink Automation.

• Syndicate metadata API gives creators options to where they want to store NFT metadata and bulk-update traits on collections.

Decentralized Metadata Registries

• Playground is building infrastructure that connects brands to existing NFT communities by issuing traits to collections. By pointing the collection’s tokenURI to Playground’s metadata registry, collection admins can opt in to have their collections be eligible to receive traits to their NFTs, giving holders perks like exclusive rewards and discounts.

Standards & Onchain Primitives

• EIP-4906: Metadata Update Extension offers a standard MetadataUpdate event to make it easy for third-party platforms to update the metadata of NFTs. OpenSea supports this, and serves well for NFTs that update via calls to the contract, but not feasible if the changes are recurring. Juicebox uses cron jobs for the latter use case.

• ERC-721TL by Transient Labs offers a way for creators and collectors to append writing to NFTs onchain (Story Inscriptions), a proposal mechanism for metadata updates (Synergy) in addition to gas-optimized implementations for batch mints and airdrops.

• EIP-721k: Dynamic Onchain Images and Metadata constructs, renders, and evolves NFTs using composable onchain SVGModules and DataStreams. The NFT encodes dynamic instructions which are relayed to SVGElements and DataStreams modules. The SVG rendering engine constructs sub-elements using a public registry, constructing the SVG by encoding/decoding inputs piped in from multiple external smart contract sources in real time.This allows the NFT to be gradually updated and improved. Expansion packs and other gaming features can be easily introduced as the NFT reaches product market fit. See this in action in Web3 Savings Cards and Pixel Pooly.

• EIP-7496 defines methods for setting and getting dynamic onchain traits associated with NFTs. By defining these traits onchain and standardizing how they can be changed, they can be used and modified by other contracts.

• EIP-6551: Tokenbound accounts have been getting a lot of attention this year and, although it doesn’t tackle dynamism within an NFT itself, it enables NFTs to hold other assets and become their own onchain identities. Tokenbound accounts are backwards and forward compatible with any NFT collection out of the box. Any smart account implementation can be deployed to an NFT and start holding assets.

• RMRK’s set of proposed standards for Multi-Asset, Nestable, Composable, Emotable and Soulbound NFTs.

• Zipped-contracts by merklejerk is a nice trick to cheaply deploy contracts that are always called offchain in an eth_call context. Contracts are compressed offchain, and the runtime contract decompresses the zipped contract, deploys it, then forwards the original call to the deployed instance. The result is bubbled up inside of a revert() payload to undo the deployment and avoid permanently modifying state. This approach results in ~50% gas savings and is useful for text-heavy primitives such as composable SVG metadata.

• Hot-chain-svg by w1nter.eth is a toolkit for building onchain SVG projects featuring a barebones rendering engine and hot reloading for developers to quickly visually QA their NFT content.

• Onchain typefaces and design systems are building blocks deployed once and put together more complex onchain media, though it will likely be a while before we see enough of these deployed to give creators a good range of options to choose from.

Conclusion

Dynamic NFTs are an exciting category-agnostic form factor for digital objects. Despite the nascency of supporting infrastructure, the ecosystem has actively been innovating on techniques to bring generative media and metadata onchain, and building composable primitives that can be reused by other projects.

Early dynamic NFT projects have required deep technical skills and bespoke implementations. But as middleware and creator tooling becomes more robust. We expect dynamic NFTs to be as ubiquitous and serve as large a variety of use cases as “vanilla” NFTs themselves. Some open questions…

• How will frontends like marketplaces, portfolio trackers, and wallets add support for NFTs that evolve in real time?

• How can we improve on decentralizing the solutions to rendering or running generative code?

• How can publishers and collectors manage the permissioning and scope of changes that can be made to a particular NFT?

• How generalisable is dynamic NFT infrastructure? Current minting engines and storefront builders were able to provide unopinionated tools to launch collections across media formats. But as dynamism varies even more across contexts, what will dynamic modules look like?

The lightning-in-the-bottle moment for NFTs was arguably not ERC721 itself, but the meteoric rise of CryptoKitties which served as a successful go-to-market for the standard. Predating even Ethereum itself, Colored Coins, Quantum, and Counterparty NFTs have shown us that technical innovation alone isn’t enough to trigger widespread experiments on top of that technology. Instead, it is the versatility, strong community evangelism, and killer application that turn standards into builders’ Schelling points for years to come. For dynamic NFTs, the primitives are already here.

If you’re building a dynamic NFT project, creator platforms and tools to launch dynamic collections, or have additional thoughts on the topic please reach out, would love to chat with you.

Many thanks to Billy Rennekamp, Benny Giang, Jayden Windle, Daniel Volkov, pet3rpan, and Accelxr for their time and feedback on drafts of this.

Special shoutout to JPG for their Dynamic Canon, the most comprehensive database of historical dynamic NFT experiments which served as a great inspiration and resource for this article.

Two years ago, at 1kx, we started to expand our focus into consumer investing.

While infrastructure investing was our core bread and butter, it was evident that the infrastructure stack was maturing, and the path to mass market viable products was only growing clearer.

Our investing so far…

Since then, we’ve made 30+ investments, 13 of which we led, across all categories of consumer, including: virtual worlds, games, creator, community coordination, social, payments and IP. Here is a short overview of our consumer portfolio to date:

What’s next?

While we approach consumer investing with a fairly open mind, we’re actively looking to invest and collaborate with founders operating in the following areas:

Social & Attention economy

As the number of onchain apps, content, and creators grow, we believe that the need for shared social spaces has never been stronger for end users. We’re especially excited about the following ideas:

1. Activity feeds that surface targeted content for end users based on their onchain holdings or connected social graphs. We’re excited about the onchain activity feed becoming an emergent discovery layer for content, creators, and social. The number of non-financially collected nfts are only growing and we believe that this an exciting but under highlighted opportunity.

2. Apps that create new social games around collecting and owning as a core game loop for both cryptonatives with existing onchain content and for mass market audiences and their respective offchain content. We believe digital property rights enable new social expressiveness regardless of the end audience. Super excited about teams building photo sharing, tokenization and collecting especially.

Creator platforms, tools and marketplaces

We’re still incredibly excited by how NFTs and digital property rights are able to empower creators to monetize and build stronger communities, particularly:

1. Vertical creator platforms and marketplaces that focus on onboarding and building a community for new categories of creators onchain. We’ve been a big fan of how SuperRare/KnownOrigin (Art), Artblocks/fxhash (generative art), Sound/Catalog (music) as well as thematic platforms (Metalabel) have pioneered adoption for their respective categories.

2. Dynamic NFT infrastructure that enables deeper more meaningful end holder experiences and turns onchain assets into recurring channels of attention as opposed to the current meta – read our research on dynamic nfts.

3. Generalized tools and infrastructure that help creators get stuff done better onchain such as products that better help manage the issuance of onchain assets, their communities, their royalties, and more. We’re excited about NFT storefronts, mint engines, community coordination tools, capital formation as well as any creator tools that allow for more multiplayer collaboration between each other – read our thesis on multiplayer creation.

IP & composable storytelling

We believe that there’s still a lot of white space in how we can use onchain primitives to create new IP and tell better stories. We think a lot about the design space around:

1. Community owned IP that closely involves community members in the earliest process of creation. While the initial wave of generation 1 cryptonative IP have mostly failed to live up to expectations, we believe that the next meta is being built as we speak – you can read our thesis on minimum viable IP.

2. Platforms and creator tools that allow for easier composable content creations on top of existing IP that already exists as nfts. We’re excited about IP that already has end user and community resonance as starting points for new content.

Virtual worlds and NPCs

We’re excited about spatial and 3D as new mediums for how end consumers will interact with ownership-based virtual experiences, especially:

1. Spatial 3D worlds that are highly interoperable (infrastructure for GLB/GLTF, VRMs, wallet indexing) where content creators are able to seamlessly create regardless of technical skill. Many projects in this area to date have focused too much on scarce land models that limit creator participation - we would like to see projects that remove this barrier to entry while building novel experiences or game loops that bootstrap initial attention. We take a lot of inspiration from Jin and M3.

2. Interactable NPCs and AI agents that leverage digital property rights to create deeper relationships between virtual beings and end users – read our thesis on virtual beings.

Autonomous Worlds & Onchain games

We’re incredibly excited about bringing net new worlds and onchain powered experiences that would otherwise not be possible:

1. Onchain games that financializes player outcomes for traditionally non-financial casual game loops. We’re big fans of projects such as Words3.

2. Worlds that are focused on establishing themselves as developer or creator platforms from day 0 as a playground for permissionless extendability – you can read our thinking on autonomous worlds + check out our community compilation.

Other areas of interest and research:

1. Mass market consumer experiences and goods (we find inspiration from Kiki’s work with women’s beauty and Pudgy Penguin’s toy launch)

2. Luxury goods and consumer experiences (we get excited with work such as Trame’s work with translating generative art into physicals)

Working with us

Through digital property rights enabled by blockchains, we believe that crypto provides a new opportunity to re-think and re-imagine what the consumer internet experiences look like. Today, we have never been this excited about consumer and believe the future is bright with white space for experimentation and opportunity.

We are excited to continue deepening our focus on consumer investing.

If you’re building something exciting, don’t be afraid to reach out!!

Consumer 🤝 1kx

@pet3rpan_ + @nichanank + Mikey0x_ + @accelxr

Today, onchain generative algorithms are primarily used in the visual arts domain, as artists write code to create digital and interactive pieces, animations, and prints. However, art is likely just the first applicable medium for this new onchain creative process. We believe onchain generative media will proliferate across every other consumer and luxury goods vertical, and this artistic process will uniquely enable a new class of physical productions through generative manufacturing.

The allure of generative collections is evident: consumers yearn for distinctive products that reflect their unique identities while also tethering them to a larger community. Through the 1/1/x model, generative algorithms accomplish this by creating singular pieces in a wider collection with cohesive aesthetics. These unique creations cater to individuals' specific tastes, allowing for granular expression within a tribe, and its success in this endeavor is showcased by the burgeoning popularity of PFP markets and the niche communities sprouting around specific generative traits.

Interestingly, generative algorithms and 1/1/x rarity distributions address the tension between mass production and customization as well. In traditional manufacturing, creating customized products on a large scale is often impractical and costly. Generative algorithms however can be directly integrated into fabrication hardware like 3d printers, CNC machines, laser printers, automatic looms, and other devices, providing feasibility in terms of production and distribution while offering scarcity and uniqueness.

This interplay between social dynamics and rarity, and digital creation and physical production, sets the stage for a new class of consumer and luxury goods that combine algorithmic randomness, end-user parameterization, and verifiable uniqueness to meet the desires of consumers.

History of Generative Manufacturing

Artists have consistently turned to technology as a means to explore new dimensions in creativity. Over time this relationship has evolved markedly, moving from pure artistic endeavors to the confluence of art and manufacturing.

• 1960s - Early Generative Art: Artists began experimenting with algorithmic processes to create artworks. Using early computers and programming languages, along with tools such as the pen plotter, artists like Manfred Mohr, Vera Molnár, and Harold Cohen began creating algorithm-driven artworks.

• 1980s - Personal Computers & Software Revolution: The emergence of personal computers made digital tools more accessible. This enabled a wider range of artists to experiment with these novel artistic processes.

• 1990s to 2000s - Birth & Expansion of Additive Manufacturing: As 3D printing emerged and evolved, artists saw new opportunities. Generative artists began to experiment with these tools, creating sculptures and installations directly from their software-driven designs.

• 2000s to 2010s - Digital Art Meets Digital Fabrication: As both fields matured, digital artists would collaborate with makers, architects, and designers to realize large-scale installations. Projects like the Hy-Fi tower by The Living, which used organic bricks cultivated from corn stalks and mushrooms, incorporated generative design principles in its conception and used modern manufacturing methods for its creation. It was around this time that software tools tailored to artists, like Processing, empowered them to create complex programmatic art without needing deep programming knowledge as well.

• 2010s - Maturation of Tools & Methods: Generative art platforms and frameworks such as openFrameworks and TouchDesigner grew in popularity. These tools, combined with more accessible and sophisticated 3D printing, laser cutting, and CNC milling technologies, enabled seamless production. For instance, artists like Nervous System used generative algorithms to design unique jewelry and apparel items that were subsequently 3D printed.

• 2020s - Convergence & Collaboration: The line between art, design, and manufacturing has become increasingly blurred. Art installations, architectural structures, and even everyday objects now showcase the unique aesthetics and capabilities that the combination can produce. Notably, onchain art renewed interest in the field of generative art at this time, using cryptographic inputs as random seeds for onchain collections. Coupled with new primitives in the digiphysical space, we are reaching a new beachhead in the marriage of digital creation and physical production.

Today's generative artists aren’t just producing art; they’re redefining consumer goods, merging aesthetic value with functional design, and pushing the frontiers of what's possible in both art and industry.

Web3 Experiments

Within web3, there are a variety of early experiments with generative manufacturing.

Trame x CPG’s Craft Nouveau is a series of collections focused on combining traditional crafts with generative art, showcasing the ability of generative code to preserve the art styles of cultures across the globe. Navette by Alexis André was the debut collection of Craft Nouveau, wherein Alexis wrote an algorithm that generates images that are designed to be woven automatically by the Neolice Loom - an automatic loom that can ingest code to weave physical pieces.

The fx(hash) ecosystem has seen a lot of experimentation on the manufacturing side - likely due to its permissionless approach to self-publishing. Klangteppich, an evolving dynamic NFT provides instructions for a loom and allows the collector to receive a physical piece of any generated frame. Mini Dahlias hold within the metadata of the NFT the instructions to create a 3.5" x 2.0" pocket sculpture from 14 layers of laser cut alpha-cellulose mat board. Nuages possible creates variations on ​​Joanie Lemercier’s cloud series with the outputs from the code recreated by a plotter machine in physical space.

Outside of craft and art, fashion is one of the most explored avenues for generative manufacturing. 9dcc’s production of Iteration-002 is an early example of combining generative designs and physical products. The Iteration-002 shirts were produced live using a printer that was plugged into SnowFro’s Squiggles algorithm. The printer relied on the algorithmic randomness of the source code to determine the traits of the design printed on the shirt, and it followed the same trait distribution as the original 10k collection.

Tribute Brand also recently remixed the Chromie Squiggle algorithm to create manufactured clothing. Chromie Squiggle holders could generate personalized sweaters using their unique Chromie Squiggle, while others can generate one-of-a-kind sweaters through the original Chromie Squiggle algorithm. The drop included both a digital and physical ODDS sweater derived from the source code of Chromie Squiggle. The digital objects serve as blueprints for future sweater editions and can be used as skins in immersive environments, and each unique ODDS digital object can be redeemed for a corresponding ODDS physical sweater, handmade by Waste Yarn Project.

Other notable generative fashion projects include mmERCH and RSTLSS, both of which plan to experiment around algorithmic randomness and designs.

Taking a slightly different approach to generative goods, Deep Objects used a community curation engine to whittle down one million designs generated by their GAN AI model to a single piece. This final piece is now set to be 3d printed in a showcase of community-powered generative product creation.

The Generative Manufacturing Stack

The generative manufacturing stack can be broken down into 5 layers:

• Creation: The initial stage where a design or concept is generated using algorithmic or AI processes.

• Curation: The process of selecting and refining the generated designs to achieve desired outcomes or specifications.

• Translation: The conversion of the digital design into machine-readable instructions or code to be used by the manufacturing equipment.

• Fabrication: The physical production or manufacturing process of the design using various materials and equipment.

• Authentication / Linking: The verification of the authenticity of the manufactured product and linking it to its digital twin to ensure provenance.

Creation Layer

The creation of a generative good starts with code. Libraries like p5.js and Processing provide powerful tools for artists and designers to create generative art. These libraries are extended through onchain randomness using seeds generated from tx hashes, token data, block headers, and more. Onchain art engines like ArtBlocks Engine and fx(hash) allow artists to easily plug these random seeds into their code and mint an artwork directly onchain.

For AI artists, this layer is focused on model development and fine-tuning to create a desired aesthetic. They typically start by selecting a pre-existing AI model, such as a generative adversarial network (GAN) as a foundation. Through backpropagation, the model weights are gradually improved to generate art that aligns with the desired style. The artist provides feedback by curating the most appealing outputs and incorporating them back into the training dataset. This iterative process continues, refining the model's performance and allowing the artist to explore different possibilities. Outside of custom models or Stable Diffusion LoRAs, there are also tools to simplify this process, such as Scenario.gg.

Curation Layer

Following the creation layer, the outputs of the code can be further refined to match a user’s preference. In the creative coding context, this is typically in the form of multiplayer parameterization like what is available with fx(hash)’s fx(params).

In the context of AI generative models, curation has typically been done via a broader community of tokenholders as is the case with Botto’s generative algorithm and the Deep Objects’ community design process.

Studios or self-publishing is the final leg of the curation process. This is where generative studios like Trame and ArtBlocks come in to present pieces to the public, or fx(hash) as a self-publisher.

Translation Layer

Once the algorithm and design is set, the generative good must be translated into machine-readable instructions for fabrication hardware. Translation is a relatively straightforward process with the goal to recreate a piece in physical space as accurately to the original as possible.

Translation can be done in a few different ways including:

• Artist / Collector Interpretation. The simplest way to translate an object is to leave the physical design specs up to the artist or collector. They would make decisions about how a piece should be made, the material to be used, the specific dimensions, etc.

• Embedded Traits. A much more scalable and interesting approach is to embed the physical information needed for fabrication into the NFT itself. Traits within the metadata of the NFT define the areas of interpretation (e.g. the thread count, thread size, weave instructions, and more for a tapestry piece).

• Direct Instantiation. A third approach is the generation of the interpretable asset directly: the generative algorithm is already adapted for the fabrication hardware, or the output of the algorithm is a 3d-printable file or the vertices of a 3d mesh.

Fabrication Layer

After translation, the generated goods are fabricated. ​​The fabrication phase is a critical step that involves converting the virtual design into a physical object. Several techniques, such as 3d printing, CNC milling, laser cutting, robo-printers, and automatic weaving, are used to create objects with different materials and shapes.

For Trame’s first drop with Alexis Andre mentioned above, the Neolice Loom was used as the fabrication hardware. The Neolice Loom takes in the artist’s custom script and reinterprets the code into 3d space through weaving. Trame is extending into new mediums as well - the above picture highlights an experiment with generative pottery.

While specific to generative art production today, Artmatr highlights what advanced fabrication tooling can do for physical production of digital items. Artists work with the Artmatr team to submit various digital file formats such as code, 3D models, PSD files (Photoshop), vectors, and animations. Next, they define the physical ‘threads’, including parameters such as the medium (oil, UV, acrylic), substrate, dimensions, and more. Lastly, this is materialized through the use of machines like robotic arms and 6-axis printers. Different techniques such as inkjet printing, airbrushing, and extrusion are utilized, and the resulting topologies can be 2D, 2.5D, or 3D.

Authentication / Linking Layer

After the physical object is created, it needs to be linked back to its digital twin. This would be similar to digiphysical processes in other spaces like fashion. Use of near-field communication chips like those manufactured by Kong and IYK, stenography, and QR codes are just some of the techniques for linking the digital to physical and providing certification around provenance.

Check out our recent post on digiphysical infrastructure here, or our piece on digital fashion which dives into digiphysical strategies specific to fashion.

Future Possibilities

Going forward, we anticipate existing onchain generative art will act as programs for derivatives. We’ve seen this through the use of Squiggles in a variety of fashion projects, and another early example of this is Terraflows, a flow field study built on top of the Terraforms art program. This kind of networked art can lead to fascinating reinterpretations of generative art in physical space. For example, one could use the Fidenza art script to create architectural layouts for 3d printed homes.

Another interesting future possibility is the tokenization of decentralized manufacturing facilities for producing generative goods at scale in a kind of physical infrastructure network. Hobbyists and commercial manufacturers with the appropriate equipment can print or create pieces by bidding for jobs posted by collectors or artists. A token can meter this hardware network and assist with bootstrapping the initial costs of manufacturing facilities. This pairs especially well with a CC0 paradigm coupled with onchain code.

Looking even further out, synthetic biology and/or chemistry manufacturing could be an interesting avenue as well for generative qualities: for example, generative code could be used to algorithmically determine the traits of lab-grown crystals, phenotypes of plants, and more.

—

We firmly believe in a digiphysical future where online networks port into physical spaces. Generative design, which includes algorithmic randomness, end-user parameterization, and verifiable uniqueness, is one such program we expect to port over - and existing fabrication hardware is capable of bringing this to fruition.

If you are building at the intersection of art and technology + digital and physical, please reach out to us.

To date, attempts to represent physical goods onchain have been spearheaded by cryptonative brands for high-end fashion items for authenticity and provenance. Meanwhile, the vast majority of onchain actions such as NFT minting, attesting, trading, and equipping remain limited to digital contexts.

We see a big potential for onchain interactions to be brought to physical environments through digiphysical consumer goods, and believe that digiphysical infrastructure will soon bring programmable utility to any physical product.

Cryptonative digiphysicals are goods that exist in physical form, and maintain a cryptographic link to its digital (onchain) record via NFC technology. This record can be in the form of an NFT, a registry entry, or an onchain identity with signing capabilities.

Chip technology coupled with onchain programmability gives wallet-like abilities (e.g. token-gated access, signing, and asset ownership) to any physical object, unlocking end consumer experiences that were not previously possible.

We are excited about this next generation of digiphysicals because they:

• Bring onchain interactions to physical environments. Embedded scannable hardware allows consumers to verifiably prove their proximity to an object or presence in a physical space. This proof can arbitrarily unlock location-sensitive utility such as live mints, or serve as a record of participation that can qualify the holder for future rewards.

• Give agency and reputation to any consumer product. Chips with signature-generation capabilities let any physical good build its own onchain history and accrue value as its owners engage with brands and communities IRL. Because an item’s onchain record is decoupled from the owner, the item can be sold with its history intact and priced accordingly.

• Create a new distribution channel for issuers. Brands, communities, and creators can give ongoing, dynamic utility to their products and spaces. IRL interactions become a metric used to extend differentiated rewards to the most engaged members, who are in turn incentivized to engage more with the community.

• Unlock fullstack utility for consumers. Owners of digiphysical products can derive value from the good both in the physical (wearing the item) and digital (flex ownership, early access rights, exclusive content) realm.

What did digiphysical products look like in the past?

The first digiphysicals existed prior to crypto. We can think of them as physical products that came “bundled” with its digital form. Think vinyl records that come with digital download codes, or physical books sold with its pdf format. While these bundled products allow buyers to consume the content in multiple ways, the digital and physical counterparts have no strict association with one another and are treated as add-ons, which often go unutilized.

Within crypto, early experiments include selling unchipped “Digital Twin” products through onchain mechanics like burn-to-redeem. Here, the owner has to burn/replace the NFT in order to claim a physical product, or claim an NFT add-on after making a physical purchase. Like bundles, early “digital twin” products treat each counterpart as a secondary offering as opposed to product components that are deeply integral to one another.

While both forms of physical-digital bundling offer incremental benefits to the consumer, there is only a loose coupling between the physical and digital product. Each good can theoretically be sold separately without losing its own intrinsic value or utility.

Embedded chips, coupled with onchain programmability, allow us to achieve tight coupling between a physical object and its onchain record and property rights. This means that the physical and digital are no-longer distinct products, but two sides of the same coin. The physical object is a means for the holder to read or write to its onchain record or exercise the rights that come with it through IRL interactions.

The prevailing mental model for digiphysical goods are IRL fashion items with an NFT counterpart wearable in virtual worlds. While this is one use case, it only scratches the surface of the design space that has opened up.

A better analogy would be a (physical) credit card and the (digital) financial identity, purchasing power, and access rights it grants. The credit card is a means for the holder to make purchases in the physical world, while the bank holds a digital record of card usage and any perks that come with it. The card issuer secures partnerships with vendors to give cardholders ongoing utility, such as airport lounge access and hotel discounts.

The digital record is not necessarily a “mirror” of the physical card, but an important part of its function as third parties read from and update its record through IRL interactions.

Another example is a festival wristband that attendees can use to enter the venue, pay vendors, and claim rewards. The wristband itself makes for a convenient method to pay, but its digital record with access rights and spendable balance is inherently a part of what makes it useful.

**Cryptonative digiphysicals bring this programmable interactivity to any consumer product to be used in a variety of financial, social, and entertainment use cases. **They can be objects that are carried or worn on-person (e.g. clothing items, POAP cards) or tied to a stationary object (e.g. artwork in a gallery, collectables), scanned to pull up its ownership history, proof-of-authenticity, historical interactions, and associated utility. Its onchain record makes the interactivity composable - anybody can distribute rewards and grant rights to any the goods without gated API access.

What use cases do cryptonative digiphysicals enable?

While the popularized notion of the metaverse is a Ready Player One-like VR environment where humans and objects have “digital twins” operating separately in immersive virtual worlds, we believe that it will look more like physical objects and spaces having an onchain record and fullstack utility across physical and digital environments.

Creator Monetisation & Fan Engagement

• Extended Merch Interactivity. Creators can now turn any physical merch into loyalty cards with post-purchase utility. Merch becomes a direct channel to fans and fans to each other, allowing creators to build a deeper relationship with their fans and within the fan base itself. The experience that comes with the merch is programmable so the interactivity is ongoing.

• Fan Segmentation. Today, there is a large information gap for artists about their fanbase as ticketing and merch sales are conducted through third parties who control how much data they share with the artist. By distributing POAPs at events and building a record of IRL interactions like taps and scans, creators get a granular view of who their most engaged fans are.

• Targeted Rewards. Using chipped merch and scannable kiosks at an event, creators are able to reward fans directly. Fans can build their reputation within the community as IRL and digital participation can be recorded onchain. Wearing chipped merch to multiple events can accrue loyalty points to the merch, which can be scanned to unlock access to presales for future shows.

Tokenized Communities

• Proof of Connection & Social Games. Products designed for individuals to have on their person, such as a piece of clothing or a card, can be a way for communities to connect with other members as they meet each other in person. It also enables new types of games around connecting in person e.g. 9dcc Network Points

• Social Graphs. “I met” POAP transfer activity creates a bottom-up social graph that can be used to see which events community members are going to, and the most “connected” members. This can serve as a primitive data point to distribute rewards to members who are most active in evangelizing the community at events.

• Payments & Co-Signing. Because a physical item can now produce cryptographic signatures, it could be used as a signer for a multisig for contactless payments as a DAO member. This could be a membership card, a hoodie, bracelet or anything else.

Decentralized Authentication & Proof-of-Ownership

• Luxury Goods & Collectables. While traditional authentication systems for luxury goods, art, and high-end collectibles rely on centralized servers, onchain authentication standards using chips enable physical products to self-attest their authenticity even if their issuers go away. Once scanned, there is a transparent record of who manufactured the chip, which brand issued the product, and what rights come with it.

• Fullstack Flex. Using standards like physical-backed NFTs, brands can tie a physical garment to a digital wearable that owners can take across the physio-virtual stack. The token standard ensures ownership of the NFT and physical items remain tightly coupled, allowing owners to verifiably flex ownership across both digital and physical spaces.

• Cryptographic Autographs. KrauseHause Metafactory jacket that basketball players can tap and sign digitally.

Digiphysical Interactions

• XRL Gameplay. On-chain games can add IRL quests by chipping physical products and updating its record as players engage with one another in the real world. Loot LARP was an XRL larp experience at NFT NYC 2022 where players fight each other with weapons embedded with Kong chips. Points were recorded to the weapons after the interaction.

• Live Minting. Galleries can design engaging, interactive minting experiences at physical locations by having guests scan a chipped disc to purchase editions of displayed artwork or mint a generative piece.

• Tokenized Guestbooks. Audiences can record their “proof of presence” in a digital work. NFT metadata could update with each message from someone who has been in close proximity to it.

Digiphysical Experiments in Web3

2019 - 2021: Early attempts to bridge the onchain and the physical

• Notable experiments using chips are Kong Cash, Signed Star Trek Collectable and MetaFactory’s Genesis Bomber (February 2020) which utilized SiLos chips.

• Redeem mechanics using ERC-20s (Saint Fame $FAME, Unisocks $SOCKS, Zora $TAPE) which traded in the open market prior to a date in which token holders could burn/replace the token in return for the physical product. Aside from showing proof of burn or the redemption token as receipt, there was no way for the owner of the physical product to prove ownership of it in real time, nor can the issuers continue to deliver perks as there isn’t an onchain linkage.

• POAP launched as a way for individuals to have an onchain record of their involvement at events.

2021: New layers of infrastructure

• IYK launched tools for issuers to program utility on top of chips, such as tap-to-own and tap-to-mint. This later enabled direct-to-item messaging in the VÉRITÉ Crewneck and POAP mints from 9dcc ITERATION-001 shirts.

• Vaulting solutions like 4k and Courtyard also launched around this time, providing a solution to securing physicals prior to their redemption or during a trade.

• Along with the rise of NFTs and generative art, Bright Moments launched to bring onchain art to physical spaces and soon started collaborating with platforms like ArtBlocks and fx(hash) to bring generative art mints/reveal experiences to the physical environment.

2022 - early 2023: Early applications on top of protocols

Seeded with the hardware and programmability layer, over the past year we saw adoption of digiphysicals in more markets, notably around tokenized communities and creator-fan engagement.

• On-chain communities like Nouns began to issue merch, working with Dopamine to launch chipped streetwear that buyers can scan to gain access to digital rewards. The DAO also approved a budget to launch Nouns Vision, with Nouns members getting pre-mint access to an NFT redeemable for luxury Nouns sunglasses.

• VÉRITÉ collaborated with IYK to launch the VÉRITÉ Crewneck, which can be tapped to gain early access to her next record and other exclusive experiences.

• Token-gating, a mechanism pioneered by Collab.Land to restrict access to Discord and Telegram channels, started being applied to physical spaces. Organizations can use protocols like tokenproof to let attendees verify asset ownership to enter a venue.

• Cryptonative brands outside of fashion such as Azuki made their foray into digiphysicals by utilizing Kong chips in their Golden Skateboards, and authoring a standard for physical-backed tokens (PBT).

• As more cryptonatives got their hands on chipped products, social games emerged around meeting as many people wearing the collection as possible. POAP mints from 9dcc ITERATION-001 shirts are observable onchain as social graphs for the community at Art Basel in Miami. At ETH Denver, Bead DAO formed as a hackathon project where beaded lizards were embedded with Kong chips, and new members could get onboarded as they met individuals carrying lizards.

• We saw the first loans taken out against digiphysicals, with gmoney taking out a loan against a 9dcc IT-001 shirt, vaulted on 4k. In April 2023, a 4k-issued PBT was used in a DeFi loan on a Rolex

The Web3 Digiphysical Stack

Infrastructure Layer

Hardware. The physical interface, such as specialized NFC chips, dynamic QR codes, or stenography, provides a linkage and means of communication to the blockchain to update a digiphysical’s onchain record. It’s important to note that not all chips are made equal. For example Kong*’s latest HaLo chips, unlike its predecessor SiLo chips and typical NFC tags, can generate its own signatures and thus behave like a hardware wallet with its own identity (onchain address) - enabling the payment and co-signing use cases.

Creator Tools, Minting & Generative Engines. The programmability layer on top of hardware infrastructure enable creators to design digiphysical experiences. IYK* provides chip-agnostic software modules for issuers, including post-distribution tooling like tap analytics. Causality is middleware for linking chipped physical goods to digital environments. Metalabel* is a platform for creators to discover each other and co-release work in the form of digiphysical record bundles as well as IRL experiences. fxhash, ArtBlocks provide minting engines for generative art, enabling live minting experiences at galleries and IRL events.

Vaulting, Escrow & Commerce. Escrow protocols like 4k, Courtyard, and Mynaswap provide shipping and storage services for physical products prior to NFT redemption or during a trade. This involves nftokenizing the product through each platform’s own minting service. Americana goes one step further to attach its own chip technology to physicals shipped in to be vaulted. These protocols are critical infrastructure for representing ownership of a physical asset onchain, as token owners must have confidence that the underlying asset exists in the exact form and quantity represented.

Supply & Value Chains. Protocols like Permet and Crowdmuse provide contribution tracking services along the supply chain of physical products from designers to manufacturers, which helps with subsequent value distribution when primitives are reused for future projects.

RWA Tokenization & Digitization protocols serve to give existing physical goods a representation onchain, ensuring linkage either by chipping the object (Mattereum’s Star Trek collectables, Sunya bracelets), or minting a “Physical-Backed Token” and securing the physical object through a decentralized network of Guardians (4k) or centralized service (Courtyard). The PBTs can then be traded on open markets like OpenSea or used as financial instruments in DeFi protocols. Digitization platform Altr gives physical brands tools to upload their archives as high-fidelity 3D assets, enabling virtual try-ons, wearables, and digiphysical redemption options.

NFT Financialization. Physical assets, once represented onchain as NFTs, can be plugged into the existing NFT finance infrastructure. They can be traded like any other NFT on OpenSea, and be used to collateralize loans on platforms like NFTfi* and Arcade. There are also specialized marketplaces such as unxd for tokenized luxury goods, Tectyle for tokenized textiles and fine art, and mntge for vintage clothes.

Generative Manufacturing / Digital-to-IRL. The adoption of generative engines has thus far been in the digital art market. As consumers look to engage with their collections IRL, projects like Trame and Artmatr provide a manufacturing pipeline for creators to transform digital files into physical art. Beyond the art market, these manufacturing pipelines could eventually serve as infrastructure for any generative consumer product. For example, a generative fashion item could first be minted as an NFT, which the holder could then use to purchase the corresponding physical product.

Standards & Registries. Physical-Backed Tokens (PBT) is a standard by which chipped physical goods can assert its own authenticity and maintain a tight coupling with its NFT counterpart onchain. This is achieved through chips that can generate its own signature, and only allowing NFT transfer if the signature of the chip is supplied to the contract (guaranteeing that a token cannot be transferred without consent from the owner of the physical item). Ethereum Reality Service (ERS), spearheaded by Arx (of Kong chips), proposes canonical registries for chips, manufacturers, and service providers which attests to the function of chips and the projects/companies embedding them in objects. It also proposes schemas whereby developers can indicate what claims and functions are possible against chips.

Application & Utility Layer

Token-Gating, Ticketing, & Memberships. Protocols like Collab.Land, Guild, Tokenproof, and Unlock allow communities to build custom token-based logic for who is allowed to access digital spaces, IRL events, and digiphysical perks (e.g. Shopify discounts). In addition to a collector’s onchain history and asset ownership, these tools could be used to extend eligibility to people who own physical products as well e.g. scan your hoodie for early access to ticket sales.

Live Minting. Communities like Bright Moments and VerticalCrypto Art host IRL events around the world centered around showcasing generative art, typically working with artists to design the physical reveal and minting experience for ticket holders and DAO members. Some galleries, like imnotArt, also has virtual world real estate where users can visit and bid on curated artwork.

Community Engagement. While POAP* is widely known to be badges that prove you attended an event, the protocol becomes more powerful when individuals are wearing chipped community merch as they can become POAP issuers themselves. TYB helps brands leverage web3 memberships to drive product development and activation, rewarding in-store visits, token-gating Shopify discounts, and enabling community co-creation.

Conclusion

We look forward to seeing more experiments in digiphysical consumer goods and experiences and anticipate that

• Chips will become cheaper and economically viable to be used in products beyond high end fashion, collectables and art

• Consumers will soon expect that creator and community merch be chipped and able to interact with them in IRL events to access future rewards

• DAOs and cryptonative communities will start using digiphysicals in the form of cards to perform operations like co-signing and payments

• NFTs will increasingly be used to represent ownership of a physical asset thanks to tokenization and vaulting protocols, bringing the cost and speed of trading down to that of purely digital assets. We’ll also see tokenized RWAs like collectables, high end fashion items being leveraged in NFT finance protocols

If you are building something in this space or have additional thoughts on the topic, would love to chat! Please reach out on Twitter @nichanank.

*denotes 1kx portfolio companies.

Many thanks to pet3rpan, Chris Lee, gmoney, Cameron Robertson, Drew Harding, and Richard Li for their time and feedback on drafts of this.

Fifty four individuals from 17+ teams attended: Playmint, Citadel, Lattice, Mithraeum, AW House, Briq, Aperture Labs, Worldrunning.guide, Cometh, Realms, Cartridge, Mask Network, Great Wyrm, ID Theory, Matchbox DAO, CafeCosmos, Caldera, Matterlabs, Conquest, Poseidon ZKP, Dip, and other great minds.

The Summit's goal was to curate a space for the builders of onchain games today to share their learnings, debate and discuss critical themes such as indexing, application frameworks, developer standards, monetization, onchain value propositions, and governance.

A key part of coming together was to share what we were working on as a community, what was on our minds, lessons we’ve learned, and more. Here are the recordings of the presentations presented by community members on the first day:

Jay Springett – Worlds & their History Done Quick

Worlds & their history done quick | Jay Springett

The Summit, hosted by 1kx. All mint proceeds go towards future community events.

https://glass.xyz

Citadel – Automatic Indexing with Codemods

Automatic Indexing with Codemods | Citadel

The Summit, hosted by 1kx. All mint proceeds go towards future community events.\nPresenter: citadel.game

https://glass.xyz

Briq – Building with Briq

Building with Briq | Briq

The Summit, hosted by 1kx. All mint proceeds go towards future community events.\nPresenter: briq.construction

https://glass.xyz

Playmint – Downstream - A Composable MMO

Cueing up...

Make it happen.

https://glass.xyz

Cometh – Lessons learned after two years of onchain games

Lessons learned after two years of onchain games | Cometh

The Summit, hosted by 1kx. All mint proceeds go towards future community events.

https://glass.xyz

Mithraeum – On-chain Gaming: Principles for Building Viable Systems

Onchain Gaming: Principles for Building Viable Systems | Mithraeum

The Summit, hosted by 1kx. All mint proceeds go towards future community events.

https://glass.xyz

1kx – Onchain Games in History

Onchain Games in History | 1kx

The Summit, hosted by 1kx. All mint proceeds go towards future community events.

https://glass.xyz

Cartridge – Building with Dojo

Building with Dojo | Cartridge

The Summit, hosted by 1kx. All mint proceeds go towards future community events.

https://glass.xyz

What’s next

We will be looking to do more as a group ✌️

At 1kx, we’re excited to invest in onchain games, playtest alpha releases, connect teams with grants programs, and help foster the best community for builders.

If you are building an onchain game and want to join a core community to get feedback from, collaborate with, or attend future onchain gaming events. Get in touch.

Photo Album.

By adding ML capabilities, smart contracts can be made more autonomous and dynamic, allowing them to make decisions based on real-time on-chain data, rather than just static rules. Smart contracts would be flexible and adaptable to various scenarios, including those that may not have been anticipated when the contract was initially created. In short, ML capabilities will broaden the automation, accuracy, efficiency, and flexibility of any smart contract we put on-chain.

In many ways, it is surprising that smart contracts do not use embedded ML models today given the prominence of ML in a majority of applications outside of web3. This absence of ML was largely due to the high computational cost of running these models on-chain. For example, FastBERT, a compute-optimized language model, uses ~1800 MFLOPS (million floating point operations), which would be infeasible to run directly on the EVM.

When considering the application of ML models on-chain, the primary focus is with the inference phase: applying the model to make predictions over real-world data. In order to have ML-scaled smart contracts, contracts must be able to ingest such predictions, but as we mentioned earlier, it’s infeasible to run the model directly on the EVM. zkSNARKs give us a solution: anyone can run a model off-chain and generate a succinct and verifiable proof showing that the intended model did in fact produce a particular result. This proof can be published on-chain and ingested by smart contracts to amplify their intelligence.

In this piece, we will:

• Review the potential applications and use cases for on-chain ML

• Explore the emerging projects and infrastructure building at the heart of zkML

• Discuss some of the challenges of existing implementations and what the future of zkML could look like

A Quick Primer on ML

Machine learning (ML) is a subfield of artificial intelligence (AI) that focuses on developing algorithms and statistical models that enable computers to learn from and make predictions or decisions based on data. ML models typically have three primary components:

1. Training Data: A set of input data that is used to train a machine learning algorithm to make predictions or classify new data. Training data can take many forms, such as images, text, audio, numerical data, or a combination of these.

2. Model Architecture: The overall structure or design of a machine learning model. It defines the types and number of layers, activation functions, and connections between nodes or neurons. The choice of architecture depends on the specific problem and data being used.

3. Model Parameters: Values or weights that the model learns during the training process to make predictions. These values are adjusted iteratively through an optimization algorithm to minimize the error between predicted and actual outcomes.

Models are produced and deployed in two phases:

1. Training Phase: During training, the model is exposed to a labeled dataset and adjusts its parameters to minimize the error between predicted and actual outcomes. The training process typically involves several iterations or epochs, and the accuracy of the model is evaluated on a separate validation set.

2. Inference Phase: The inference phase is when a trained machine learning model is used to make predictions on new, unseen data. The model takes in input data and applies the learned parameters to generate an output, such as a classification or regression prediction.

zkML today is primarily focused on the inference stage of ML models rather than on the training phase primarily due to the computational complexity of verifying training in-circuit. zkML’s focus on verifying inference is not a limitation though: we expect some very interesting use cases and applications to be produced from the inference phase.

Verified Inference Scenarios

There are four possible scenarios for verified inference:

• Private Input, Public Model. A Model Consumer (MC) may wish to keep their inputs private from a Model Provider (MP). For example, a MC may wish to prove the result of a credit-scoring model to a lender without disclosing their personal financial information. This could be done using a pre-commitment scheme and running the model locally.

• Public Input, Private Model. A common issue with ML-as-a-Service is that the MP may wish to hide their parameters or weights to protect their IP, and the MC wants to verify that the generated inferences do in fact come from the specified model in an adversarial setting. Think of it this way: a MP has an incentive to run a lighter model to save on costs when serving inferences to a MC. Using a commitment of the model weights on-chain, a MC can audit a private model at any time.

• Private Input, Private Model. This scenario arises when the data being used for inference is highly sensitive or confidential, and the model itself is hidden to protect IP. An example of this might include auditing a healthcare model using private patient information. Compositional techniques in zk or use of multi-party computation (MPC) or variations of FHE can be used to service this scenario.

• Public Input, Public Model. When all aspects of the model can be public, zkML solves for a different use case: compression and verification of off-chain computation to an on-chain environment. For larger models, it is more cost-effective to verify a succinct zk proof of an inference than to re-run the model themselves.

Applications and Opportunities

Verified ML inference opens up a new design space for smart contracts. Some crypto-native applications include:

DeFi

• Verifiable Off-Chain ML Oracles. Continued adoption of generative AI may help push industries to implement signature schemes for their content (e.g. a news publication signing an article or image). Signed data is ZK-ready and makes the data composable and trustworthy. ML models can process this signed data off-chain to make predictions and classifications (e.g. classifying the outcome of an election or a weather event). These off-chain ML oracles could be used to trustlessly settle real-world prediction markets, insurance protocol contracts, and more by verifying the inference and publishing the proof on-chain.

• ML-parameterized DeFi applications. Many facets of DeFi could be more automated. For example, a lending protocol could use an ML model to update parameters in real time. Today, lending protocols primarily trust off-chain models run by organizations to determine the collateral factor, LTV, liquidation threshold, etc., but a better alternative could be community-trained open-source models that can be run and verified by anyone.

• Automated trading strategies. A common way of showcasing the return profile of a financial model strategy is for the MP to provide various backtests to investors. However, there is no way to verify the strategist is following the model when executing a trade - instead, the investor must trust the strategist is actually following the model. zkML offers a solution where the MP can provide proof of the financial model inference when deploying into a specific position. This could be particularly useful for DeFi managed vaults.

Security

• Fraud monitoring for smart contracts. Rather than having slow human governance or centralized actors control the ability to pause a contract, an ML model could be used to detect possible malicious behavior and enact a pause.

Traditional ML

• A decentralized, trustless implementation of Kaggle. A protocol or marketplace could be created that allows a MC or other interested party to verify the accuracy of a model without the MP disclosing the model weights. This can be useful for selling models, running competitions around model accuracy, etc.

• Decentralized prompt marketplaces for generative AI. Prompt creation for generative AI has evolved into a sophisticated craft where the best output generating prompts are often complex with many modifiers. External parties could be willing to purchase these sophisticated prompts from a creator. zkML can be used in two ways here: 1) verifying the outputs of a prompt to ensure to a potential purchaser that the prompt does in fact create the desired images, and 2) allowing the prompt owner to maintain ownership of the prompt after a purchase by keeping it obfuscated from the purchaser while still generating verified images for them.

Identity

• Replacing the private key with privacy-preserving biometric authentication. Private key management remains one of the largest frictions in web3 UX. Abstracting the private key via facial recognition or other unique factors is one possible solution from zkML.

• Fair airdrops and contributor rewards. An ML model could be used to create detailed personas of users that determine airdrop allocations or contribution rewards based on multiple factors. This can be particularly powerful when combined with an identity solution. In this scenario, one possibility could be having users run an open-source model that evaluates their participation in an application along with higher context participation like governance forum posts to infer an allocation for them. They then provide this proof to the contract to receive their token allocation.

Web3 Social

• Filtering for web3 social media. The decentralized nature of web3 social applications will result in higher volumes of spam and malicious content. Ideally, a social media platform could use an open-source ML model that is agreed upon by the community and publish proofs of the model’s inference when electing to filter a post. For more on this, check out Daniel Kang’s zkML analysis of the Twitter algorithm here.

• Advertising / Recommendations. As a social media user, I may be willing to view personalized advertisements but wish to keep my preferences and interests private from advertisers. I could elect to run a model on my tastes locally that feeds into media applications to serve me content. In doing this, advertisers in this scenario may be willing to pay for the end-user to do this, however it is likely these models would be much less sophisticated than targeted advertising models in production today.

Creator Economy / Gaming

• In-game Economic Rebalancing. ML models could be used to dynamically adjust token issuance, supply, burn, voting thresholds, etc. One possible model would be an incentivized contract that rebalances an in-game economy if a certain rebalancing threshold is met and a proof of inference is verified.

• New types of on-chain games. Co-operative human versus AI games and other innovative takes on on-chain games could be created where a trustless AI model serves as a non-playable character. Each action the NPC takes is published to the chain with a proof that anyone can verify to determine the correct model is being run. In the case of Modulus Labs’ Leela vs. the World, the verifier wants to ensure the stated 1900 ELO AI is picking the chess moves and not Magnus Carlson. Another example is AI Arena, a Super Smash Brothers style AI fighting game. Players in high stakes competitive environments would want to be sure that there is no interference or cheating involved with their trained models.

Emerging Projects and Infrastructure

The ecosystem for zkML can be broadly broken down into four primary categories:

• Model-to-Proof Compilers: Infrastructure that compiles a model from an existing format (e.g. Pytorch, ONNX, etc) into a verifiable computational circuit.

• Generalized Proving Systems: Proving systems built to verify an arbitrary computational trace.

• zkML-Specific Proving Systems: Proving systems specifically built to verify an ML model’s computational trace.

• Applications: Projects working on unique zkML use cases.

Model-to-Proof Compilers

When looking at the zkML ecosystem, most attention has been given to creating model-to-proof compilers. Generally, these compilers translate high-level ML models written in Pytorch, Tensorflow, or the like into zk circuits.

EZKL is a library and command-line tool for doing inference for deep learning models in a zk-SNARK. With EZKL, you can define a computational graph in Pytorch or TensorFlow, export it as a ONNX file with some sample inputs in a JSON file, and point EZKL to these files to generate a zkSNARK circuit. With the latest round of performance improvements, EZKL can now prove an MNIST-sized model in about 2 seconds and 700MB of RAM. EZKL has seen some significant early adoption thus far, being used as infrastructure for various hackathon projects to date.

Cathie So’s circomlib-ml library contains various ML circuit templates for Circom. Circuits include some of the most common ML functions. Keras2circom, also built by Cathie, is a python tool that transpiles a Keras model into a Circom circuit using the underlying circomlib-ml library.

LinearA has developed two frameworks for zkML: Tachikoma and Uchikoma. Tachikoma is used to convert neural networks into an integer-only form and generate a computational trace. Uchikoma is a tool that converts TVM's intermediate representation into programming languages that don't support floating point operations. LinearA plans to support Circom, which uses field arithmetic, and Solidity, which uses signed and unsigned integer arithmetic.

Daniel Kang’s zkml is a framework for constructing proofs of ML model execution in ZK-SNARKs based on his work in the Scaling up Trustless DNN Inference with Zero-Knowledge Proofs paper. As of the time of this writing, it is capable of proving an MNIST circuit using around 5GB of memory and around 16 seconds to run.

On the more generalized model-to-proof compilers, there is both Nil Foundation and Risc Zero. Nil Foundation’s zkLLVM is an LLVM-based circuit compiler and is capable of verifying computational models written in popular programming languages such as C++, Rust, and JavaScript/TypeScript, among others. It is generalized infrastructure versus some of the other model-to-proof compilers noted here, but it is still suitable for complex computations like zkML. This can be particularly powerful when combined with their proof market.

Risc Zero has built a general-purpose  zkVM tagetting theopen-source RISC-V instruction set, hence supporting  existing mature languages such as C++ and Rust, as well as the LLVM toolchain. This allows for seamless integration between host and guest zkVM code, similar to Nvidia's CUDA C++ toolchain, but with a ZKP engine in place of a GPU. Similar to Nil, it is possible to verify the computational trace of an ML model using Risc Zero.

Generalized Proving Systems

Improvements in proving systems have been the primary enabler in bringing zkML to fruition, specifically the introduction of custom gates and lookup tables. This is primarily due to ML’s reliance on nonlinearities. In short, nonlinearities are introduced through activation functions (e.g. ReLU, sigmoid, and tanh), which are applied to the outputs of linear transformations within a neural network. These nonlinearities are challenging to implement in zk circuits due to limitations around mathematical operation gates. Bitwise decomposition and lookup tables can help solve this problem by precomputing the possible outcomes of a nonlinearity into a lookup table, which interestingly is much more computationally efficient in zk.

Plonkish proof systems tend to be the most popular backends for zkML for this reason. Halo2 and Plonky2 with their table-style arithmetization scheme can handle neural network non-linearities well via lookup arguments. In addition, the former has a vibrant developer tooling ecosystem coupled with flexibility, making it the de facto backend for many projects including EZKL.

Other proof systems have their benefits as well. R1CS-based proof systems include Groth16 for its small proof sizes and Gemini for its handling of extremely large circuits and linear time prover. STARK-based systems like the Winterfell prover/verifier library are also useful especially when implemented via Giza’s tooling that takes a Cairo program’s trace as an input and generates a STARK proof using Winterfell to attest to the correctness of the output.

zkML-Specific Proving Systems

Some progress has been made in designing efficient proof systems that can handle the complex, circuit-unfriendly operations of advanced ML models. Systems like zkCNN, which is based on the GKR proving system, or the use of compositional techniques in the case of systems like Zator are often more performant than their generalized counterparts as evidenced by the Modulus Labs’ benchmarking report.

zkCNN is a method for proving the correctness of convolutional neural networks using zero knowledge proofs. It uses a sumcheck protocol to prove fast Fourier transforms and convolutions with a linear prover time that is faster than computing the result asymptotically. Several improvements and generalizations have been introduced for interactive proofs, including verifying the convolutional layer, the ReLU activation function, and max pooling. zkCNN is particularly interesting given Modulus Labs’ report on benchmarking where they found it outperformed other generalized proving systems in both proof generation speed and RAM consumption.

Zator is a project aimed at exploring the use of recursive SNARKs to verify deep neural networks. The current constraint for verifying deeper models is fitting the entire computation trace into a single circuit. Zator proposes verifying one layer at a time using recursive SNARKs, which can verify N-step repeated computation incrementally. They use Nova to reduce N instances of computation into a single instance that can be verified at the cost of a single step. With this approach, Zator was able to snark a network with 512 layers, which is as deep as most production AI models today. Proof generation and verifying time for Zator are still too large for mainstream use cases, but their compositional technique is interesting nonetheless.

Applications

Much of the focus of zkML given its early stage has been on the above infrastructure. However, there are a few projects working on applications today.

Modulus Labs is one of the most diverse projects within the zkML space, working on both example applications and relevant research. On the application side, Modulus Labs has demonstrated zkML use cases through RockyBot - an on-chain trading bot - and Leela vs. the World - a chess game where all of humanity plays against a verified, on-chain instance of the Leela chess engine. The team has also branched into research, writing The Cost of Intelligence which benchmarked various proving systems for the speed and efficiency across differing model sizes.

Worldcoin is applying zkML in its attempt to make a privacy-preserving proof of personhood protocol. Worldcoin is using custom hardware to process high-resolution iris scans which are inserted into their Semaphore implementation. This can then be used to perform useful operations like membership attestation and voting. They currently use a trusted runtime environment with a secure enclave to verify the camera-signed iris scan, but they ultimately aim to use a ZKP to attest to the correct inference of the neural network for cryptographic-level security guarantees.

Giza is a protocol that enables the deployment of AI models on-chain using a fully trustless approach. It uses a technology stack that includes the ONNX format for representing machine learning models, the Giza Transpiler for converting these models into the Cairo program format, the ONNX Cairo Runtime for executing the models in a verifiable and deterministic way, and the Giza Model smart contract for deploying and executing the models on-chain. While Giza could also fit into the Model-to-Proof compiler category, their positioning as a marketplace for ML models is one of the more interesting applications out there today.

Gensyn is a distributed hardware provisioning network used to train ML models. Specifically, they are engineering a probabilistic audit system based on gradient descent and using model checkpoints to enable a decentralized GPU network to service the training of full-scale models. While their zkML application here is highly specific to their use case - they want to ensure that when a node downloads and trains a piece of a model that they are honest about the model updates - it showcases the power of combining zk and ML.

ZKaptcha is focused on the bot problem within web3, providing captcha services for smart contracts. Their current implementation has the end user essentially producing a proof of human work by completing the captcha, which is verified by their on-chain verifier and accessible by a smart contract via a couple lines of code. Today, they are primarily only relying on zk, but they intend to implement zkML in the future similar to existing web2 captcha services that analyze behaviors like mouse movements to determine if a user is human.

Given how early the zkML market is, a lot of applications have been experimented with at the hackathon level. Projects include AI Coliseum, an on-chain AI competition built using EZKL that uses ZK proofs to validate machine learning outputs, Hunter z Hunter, a photo scavenger hunt using the EZKL library to validate an image classification model’s outputs with halo2 circuits, and zk Section 9, which converted an AI image generating model into a circuit for minting and verification of AI art.

The Challenges of zkML

While improvements and optimizations are being made at light-speed, some core challenges remain for the zkML space. These challenges range from technical to practical and include:

• Quantization with minimal accuracy loss

• Circuit sizes especially when a network is composed of many layers

• Efficient proofs for matrix multiplication

• Adversarial attacks

Quantization is the process of representing floating-point numbers, which most ML models use to represent model parameters and activations, as fixed-point numbers, which is necessary when dealing with the field arithmetic of zk circuits. The impact of quantization on accuracy for machine learning models depends on the level of precision that is used. Generally, using lower precision (i.e., fewer bits) can result in reduced accuracy because it can introduce rounding and approximation errors. However, there are several techniques that can be used to minimize the impact of quantization on accuracy, such as fine-tuning the model after quantization and using techniques like quantization-aware training. In addition, Zero Gravity - a hackathon project at zkSummit 9 - has shown that alternative neural network architectures developed for edge devices such as the weightless neural network can be useful for avoiding the problems of quantization in-circuit.

Outside of quantization, hardware is another key challenge. Once a machine learning model is properly represented via a circuit, it is cheap and fast to verify proofs of its inferences due to the succinctness property of zk. The challenge here is not on the verifier, but on the prover since RAM consumption and proof generation time scales quickly as the model size grows. Certain proving systems (e.g. GKR-based systems that use sumcheck protocols and layered arithmetic circuits) or compositional techniques (e.g. wrapping Plonky2, which is efficient at proving time but poor at efficient proof sizes for large models, with Groth16, which doesn’t see a growing proof size with complexity of the model) are better suited for handling these issues, but managing tradeoffs is a core challenge for projects building in zkML.

On the adversarial side, there is also still work to be done. For one, if a trustless protocol or DAO elects to implement a model, there is still risk around adversarial attacks during the training phase (e.g. training a model to behave a specific way when it sees an input which could be used to manipulate later inferences). Federated learning techniques and training phase zkML could be one way of minimizing this attack surface.

Another core challenge is the risk of model-stealing attacks when a model is privacy-preserved. While a model’s weights can be obfuscated, it is theoretically possible to reverse engineer the weights given enough input-output pairs. This is mainly a risk for smaller scale models, but it is a risk nonetheless.

Scaling Smart Contracts

Although there have been some challenges with optimizing these models to operate within the constraints of zk, improvements are being made at exponential speed and some anticipate we will reach parity with the broader ML space soon assuming further hardware acceleration. To underscore the pace of these improvements, zkML has gone from 0xPARC’s zk-MNIST demonstration in 2021 showing how a small-scale MNIST image classification model could be performed in a verifiable circuit to Daniel Kang’s paper doing the same for ImageNet-scale models just under a year later. In April of 2022, this ImageNet-scale model was further improved from 79% accuracy to 92% accuracy, and networks as large as GPT-2 are potentially feasible in the near-term albeit with slow proving times today.

We view zkML as a rich and growing ecosystem that looks to scale the capabilities of blockchains and smart contracts to be more flexible, adaptable, and intelligent.

While zkML is still in its early stages of development, it has already begun to show promising results. As the technology evolves and matures, we can expect to see even more innovative use cases for zkML on-chain.

For more information on the zkML space, I highly recommend the zkML Community telegram group and the associated resource Github.

If you are building zkML applications or infrastructure, DM me - we would love to connect with you!