That framing is breaking, and two things changed at once. World models now learn “what happens next” from internet-scale video, giving robots physics priors without hand-coded simulators or armies of teleoperators. And a thin layer of real demonstrations now bootstraps enormous synthetic scale. NVIDIA’s Manipulation-Augmented dataset turns 10 human teleoperated demos into 1,000 domain-randomized examples. The real world becomes the seed, not the substrate.

The data confirms the shift. Robotics went from 1,145 datasets on Hugging Face in 2024 to 26,991 in 2025, climbing from rank 44 to 1, with synthetic generation as the primary driver. And this week, NVIDIA launched Cosmos 3, an open model that collapses vision reasoning, world generation, and action prediction into a single system. Cosmos runs as a physics-grounded simulator that predicts approaches, evaluates them in a closed loop, and converges on behavior without real-world risk.

We think the moat is moving. Not to whoever owns the best model or the most teleoperators, but to whoever builds the best learned simulator. Reality becomes the verification step, and the data flywheel that used to live on customer floors now runs in software. The teams that own the simulator own what comes out of it.

What beat it is API plus CLI.

Anthropic's Claude Code has become the fastest-onboarding developer surface the company has shipped, despite running entirely in a text terminal that looks like it was designed in 1985. OpenAI killed Operator as a standalone product and now ships its computer-use model through the API instead. Anthropic's Quick Mode for Claude in Chrome ships a stripped-down agent loop that swaps structured JSON for single-character commands and delivers a roughly threefold speedup on real browsing tasks. The public benchmarks we've found point the same direction: agents that work through APIs and command lines are pulling ahead of agents that work through pixels.

The reason is unit economics, and it gets worse for vision every quarter. A browser-based agent pays to tokenize a screenshot and a DOM tree on every single step, feeding a model big enough to reason over a megabyte of UI noise just to find one button. A CLI agent pays for a few hundred structured tokens per step and gets a structured answer back. The gap isn't one click – it's tokens-per-step multiplied across every step of every task. Frontier model pricing for structured reasoning is collapsing faster than vision pricing, and the gap is widening. The cost curve is bending against pixels, and the bend is steepening.

The implication for SaaS is the part most software companies have not yet priced in. If the dominant user of your product over the next decade is a fleet of agents, then the surface that matters for distribution is your schema, not your screen. Salesforce, Notion, Linear. Every category leader is about to be evaluated on a single question: how cleanly can a model consume your product without rendering a pixel. The companies that ship a real, documented, agent-readable API as a first-class product become rails – Stripe proved the model years ago. The companies that gate their primitives behind a UI become the legacy layer that the next generation of startups quietly automates around.

There is a real counter to this view: the browser is also the universal abstraction over the long tail of legacy software that will never expose a clean API. State government portals, hospital admin systems, niche industry tools with three customers and a 2008 codebase. That ~15% of the workflow surface is a real market, and vision agents will serve it for a long time. But 15% is a fallback business. The platform fight is the other 85%.

The bigger compounding effect sits at the infrastructure layer. A human can run one workflow at a time. An agent can run a hundred in parallel, each making thousands of API calls. Our bet is that machine-initiated traffic comes to dwarf human-initiated traffic by an order of magnitude, and the inference cycles, bandwidth, and storage required to support that pattern are nowhere in the current hyperscale capex curve. The most underwritten thing in markets right now is not the model layer or the application layer. It is the silicon, fibre, and power required to carry the load that an agent-native software stack actually generates.

Our view, plainly: the SaaS layer bifurcates. The companies that ship clean, agent-readable APIs become the rails. Everything else becomes vision-agent fallback or a wrapper around someone else's schema. The infrastructure layer captures the spillover, and the spillover is arguably the biggest single capex flywheel in history.

The next decade of software is being built for readers that do not have eyes. Plan for them.

The gap isn’t capability. The models are good enough. The gap is that an agent, no matter how sophisticated, cannot yet participate in commerce as a real economic actor. It can read a contract better than most lawyers. It cannot sign one. It can manage a treasury. It cannot own one. It can run your marketing, your procurement, and your customer support, but you won’t hand it the corporate bank account. There’s no way to scope what it can touch, no audit trail if something goes wrong, and no kill switch that actually works in real time. So the most capable employee at every fast-growing company is locked out of the most leveraged tool.

This is the actual bottleneck in enterprise AI adoption, and it runs deeper than most discussions acknowledge.

Making an agent genuinely autonomous requires solving three distinct planes of infrastructure simultaneously. The first is trust: does the counterparty know who this agent is, what it’s authorized to do, and what its track record looks like? The second is market: can the agent discover work, quote a price, sign a contract, settle payment, and resolve a dispute without human intervention? The third is control: can the organization deploying the agent prove, to a regulator or a board, exactly what the agent was authorized to do and what it actually did?

None of these planes exists in mature form today.

On the trust side, the identity problem alone is unsolved. Existing card frameworks like EMV, 3-D Secure and chargeback rules assume a human clicked a button. KYC verifies a face and an ID, not the autonomous code now acting on a person’s behalf. Once an agent executes a payment, the liability question becomes unresolved: is it the issuer, the merchant, or the model? t54 is one of the more interesting attempts to answer this, building what they call a Know Your Agent layer, binding human identity to agent identity and producing cryptographic proof of intent that financial institutions can actually consume.

On the market side, even getting an agent access to a payment account is harder than it sounds. The obvious solution is to give it access to the company bank account. The obvious problem is that one bad decision wipes the treasury. Bank of Bots is building around this constraint: a segregated, scoped account that sits alongside your existing bank, where each agent gets hard spending limits and merchant whitelists. The blast radius is bounded by design.

And on settlement, fast, cheap, and programmable stablecoins look like the natural rails for agent-to-agent commerce. Except they’re irreversible. No chargebacks, no dispute window. Anchor is building the risk pricing layer that sits in front of stablecoin transactions, evaluating and pricing each one before it broadcasts. The analogy is interchange: credit cards always worked because a chunk of every swipe funded the fraud and dispute machinery. Stablecoins skipped that and skipped the protection with it.

What’s notable is that none of these companies is trying to build the whole stack. Each is one layer. And that’s fine, each layer is genuinely hard. But it also means the market is still very early and very fragmented.

The processing layer itself won’t be where the money is made. Stripe, Coinbase, Circle, and Visa will commoditize that fast. The venture opportunity is in what gets built on top of verified agent payment flows: credit, insurance, and reputation data. The more agents you’ve scored and lent to, the better your underwriting model, and the harder you are to displace. That’s a real moat.

The identity and trust layer is probably the most interesting place to be right now. Whoever builds the credit bureau for software, scoring agents, verifying what they claim about themselves, pricing their risk for merchants and lenders, wins a category that didn’t exist three years ago and that no incumbent is well-positioned to own.

Last week we wrote about where the value gets built in agentic payments. This is why it matters. The models are commoditizing. The infrastructure above them is not built yet. That’s the opportunity.

How much does this SAFE actually cost us? What if we stack another at a higher cap? How much do we lose if we top up the option pool to 15% at the Series A? At a $1B exit, what do we actually walk away with?

They couldn’t answer those quickly. We couldn’t either, sitting on a call, without opening a spreadsheet.

The math compounds in non-obvious ways. Each post-money SAFE locks in investment divided by cap at conversion. Stacked SAFEs at different caps don’t dilute each other. Founders absorb all of it. Pre-money option pool top-ups come out of existing equity, not the new round. Liquidation preferences hurt at low exits and evaporate at high ones.

Founders sign these terms in moments of high emotion. The lead just committed. The lawyer wants to close. You want to get back to building. The second-order effects don’t surface until years later, by which point your ownership is already where it is.

So we built it. A free, single-page tool that lets founders stack SAFEs at any cap, add priced rounds, top up option pools, set exit valuations, and watch the waterfall move in real time. No login, no spreadsheet.

dilutionlab.canonical.cc

Try it on your own cap table. Drag the exit slider. See exactly how much you walk away with.

We vibe coded it in <30 minutes. The source is on GitHub. If anyone wants to add features or fix corner cases, send a PR.

These products are real, but they’re narrow. Each one lets an agent transact on behalf of a human, usually with virtual card credentials, spending limits, and merchant whitelists. The agent gets to act. But the human is still the one who signed up for the card, got KYC’d, approved the merchant, and eats the loss if something goes wrong.

The agent looks like the buyer. The human is still the buyer.

That works today, when a person tells an agent to book a flight or order groceries. Volumes are small, the agent has a human to fall back on, and credit cards do the job.

It breaks when agents start transacting with other agents on their own. Imagine one agent paying another agent a fraction of a cent for an API call, a thousand times a minute, with no human in the loop. Credit cards were never designed for that. The amounts are too small. Settlement is too slow. The chargeback process needs a human to file a dispute.

Stablecoins are the only payment rail ever built for this kind of activity. They’re programmable. They settle in seconds. They have no chargeback layer. An agent can hold a balance and spend it directly, in any size, without a human approving each step.

@circle made this concrete today with the launch of Circle Agent Stack. Agent wallets funded with USDC. A CLI that lets agents in Codex, Claude Code, and Cursor discover and pay for services programmatically. Nanopayments through Circle Gateway so an agent can pay an API in real time, in any amount, with no checkout page.

Blessing Adesiji

@bleso_a

Today, we launched Circle Agent Stack, and I want to explain the core problem this solves.

AI agents can browse, reason, write code, call tools, and complete tasks. But they still struggle with one very basic thing: paying for services on the internet.

Let’s say an agent needs

agents.circle.com

Circle for Agents

Financial infrastructure for autonomous economic actors. Wallets, a marketplace, and a CLI for agents to hold funds, discover services, and transact in USDC.

Circle

@circle

Introducing Circle Agent Stack: financial infrastructure for the agentic economy.

Agent Stack gives agents the tools to:
→ Hold and move USDC through Agent Wallets
→ Discover services through Agent Marketplace
→ Execute repeatable financial actions through Circle CLI

All

46

6:16 PM • May 11, 2026

All the incumbents see this: Stripe paid $1.1B for Bridge, Mastercard paid $1.8B for BVNK, @stripe‘s Machine Payments Protocol, built with @tempo, routes payments natively to stablecoins. They aren’t betting against stablecoins. They’re buying their way in.

Which raises the real question for builders. If Stripe, Visa, Mastercard, and now Circle are going to set the standard for how agents and merchants talk to each other, and they will because they own distribution, then where is the actual opportunity for a startup?

Not in the coordination layer. Any startup building an agentic checkout or an agent wallet is competing with the next Stripe or Circle product release. Looks good today. No moat tomorrow.

The wedges that matter sit above and below.

Below the coordination layer: payment rails purpose-built for agents, not for humans using agents. @coinbase‘s x402 protocol moved $3M in its first seven days after eighteen months of moving $80K – demand is starting to flip.

Above the coordination layer: trust and identity for agents. The hard question isn’t how two agents transact. It’s whether either one should believe the other is real, has a track record, and will actually pay. No one scores this today. ERC-8004 is the leading onchain standard for agent identity. Visa has a centralized version. But standards aren’t products. Whoever builds the credit bureau for software wins a category that didn’t exist three years ago.

Highest leverage: credit, insurance, and yield on top of agent payment flows. Payment processing is a low-margin business. The money has always been in what sits above the rails. Our portfolio company @raincards proves this with stablecoin cards today. The payment is the entry point. The credit and rewards are the business. Agentic payments will follow the same shape.

Honest pushback: most agent builders today have credit cards, not stablecoins. So Visa and Mastercard are the easier on-ramp for the next year. That’s true. It’s also why most of the early agent payment startups will get acquired or shut down by Stripe before they ever build a defensible business.

Real adoption in payments has always started with a killer app, not infrastructure. Stripe scaled because Shopify and Uber needed payment processing first. The agentic killer app hasn’t shown up yet. Building infrastructure ahead of demand is fine. Building it on the wrong rails isn’t. If the agent that everyone ends up using lives on stablecoins, the startups built on Visa can’t swap their foundation. They die.

The bet is straightforward. Stablecoin rails for agents. Trust and reputation infrastructure where crypto already has the lead. Credit and insurance on top of agent payment flows.

Everything else is at risk of being turned into a feature by the platforms that already own distribution.

We use AI to buy back attention, not outsource judgment. The goal is to automate repetitive, high-volume parts of venture - both things we previously did and found monotonous and things we never had the bandwidth to do - first-principles thesis work, founder support, deep diligence and reference checks, and high-context relationship building.

How we use AI at Canonical today:

We run an agent that monitors newly published academic papers across frontier tech and AI. It tracks them in a living spreadsheet, tags them by category, and flags the ones that matter based on our internal thesis. This is a deal flow tool: the best founders often show up as authors before they appear in your inbox. Over time, we want this workflow to automatically pull context on authors and labs so outreach becomes faster and higher signal.

We run an agent over our email deal flow. Any deal we get sent gets triaged: the system detects whether it is an investable opportunity, extracts key fields into our CRM, and produces a structured readout against our deal criteria. The objective is not to “decide for us,” but to ensure every inbound gets a consistent first pass, and that nothing slips through cracks when volume spikes.

We run an agent over call notes and transcripts. It routes them into the right CRM entities (founders, companies, LPs), updates the record, and pulls out the cleanest takeaways. This has become one of the most leveraged parts of our process because it turns conversations into compounding institutional memory. It also feeds directly into how we generate these weekly entries.

Where this is going next is even more interesting: agent identity. Most “agentic” systems today still assume the agent is just an extension of the user, operating with the user’s permissions. That model worked for cloud software because it made integrations simple. But systems like OpenClaw reveal the next phase: agents that operate more like colleagues, running independently and in parallel. With this, firms will need scoped access, sandboxes, auditability, and clear blast-radius controls. The winners will not just adopt agent tooling; they will train a small set of firm-specific agents and govern them like real actors inside the org.

For most of crypto’s history, credit was something to be avoided. Lending systems were designed to eliminate risk rather than price it, relying almost exclusively on overcollateralization and reflexive demand to function. Yield, when it appeared, was a byproduct of leverage and incentives, not durable cash flows.

That is starting to change.

Over the past year, a new class of protocols has emerged that run credit explicitly onchain. Instead of trying to engineer risk away, these systems generate yield by financing real economic activity: lending to specific counterparties, underwriting real-world risk, or funding assets with predictable revenue. The mechanics differ, but the common thread is simple: yield is increasingly tied to who is borrowing, why they need capital, and how losses are handled when things go wrong.

Where Stablecoin Yield Actually Comes From

To understand why this shift matters, it helps to separate stablecoin yield into its underlying components. Despite the diversity of products and narratives, much of the new onchain yield emerging today comes from a small number of economic sources.

1. Capital Where Banks Can’t Go

This exists wherever there is a structural mismatch between who can lend cheaply and who needs capital. In many markets, traditional lenders are constrained by regulation, risk-weighting, or operational and compliance friction, even when the underlying assets generate predictable cash flows. The result is a higher cost of capital driven by regulatory and operational constraints.

Stablecoins efficiently step into these gaps. Protocols like USD.ai finance infrastructure assets like GPUs that banks are poorly positioned to underwrite, while others like Daylight target energy and infrastructure financing more broadly. In these cases, yield compensates investors for supplying capital where legacy balance sheets cannot, or will not, operate.

2. Underwritten Credit

This is where yield is explicitly compensation for default risk. Capital is lent to identifiable borrowers, and losses are explicitly planned for, rather than treated as tail events to be engineered away.

What matters most in these systems is not the headline rate, but how losses are allocated. Who absorbs first loss? How much protection exists above them? And how quickly can capital exit when conditions change? Overcollateralization was the dominant answer in earlier DeFi systems, but it is no longer the only one.

Protocols like Wildcat extend undercollateralized credit to institutional counterparties based on reputation and relationship-driven risk assessment, while 3Jane focuses on smaller operators and merchants with measurable cash flows. In both cases, yield is inseparable from borrower quality and loss structure.

3. Insurance & Risk Pools

These systems generate yield by absorbing volatility that others prefer to avoid. Participants earn premiums in exchange for taking on exposure to specific categories of risk.

This is structurally different from lending. Returns tend to be uneven, with long periods of stability punctuated by drawdowns. Returns depend less on steady utilization and more on whether losses arrive independently or all at once, which is why leverage can dramatically amplify downside.

Reinsurance protocols like Re (auto, homeowners, small business) and OnRe (catastrophe) both sit in this category, but with very different volatility profiles depending on the risks they underwrite. The distinction between them is not whether yield is “real,” but which risks are being priced, and how transparent those risks are to capital providers.

From Tokenization to Credit Design

What’s interesting about this moment isn’t that assets are being brought onchain. Tokenization by itself is not new, and it’s rarely the hard part. The harder problem is designing credit systems around those assets that can scale, survive stress, and allocate losses clearly.

The real innovation is not the wrapper, but the structure. How underwriting is done. Where first loss sits. How liquidity is gated. How quickly risk becomes visible to capital providers. Two protocols can both advertise “real-world yield” while one behaves like senior secured financing and the other like a volatility-selling strategy.

Seen this way, the recent wave of “stablecoin yield” and “real-world assets” is less about a single onchain credit narrative and more about a growing range of distinct designs. Yield now reflects how credit is structured, including who borrows, who absorbs losses, and how capital moves under stress. Different protocols express different balance sheets onchain, each with its own assumptions about risk, trust, and time.

That is the shift. Crypto is no longer just moving dollars or wrapping assets. It is starting to design credit.

What Unlocks Onchain Credit

What makes this moment interesting is not just that credit is being rebuilt onchain, but that crypto is beginning to expand the frontier of what credit systems can do. Blockchains, stablecoins, decentralized exchanges, and cryptographic primitives are converging into a stack that allows capital to move, price risk, and settle globally in ways that traditional finance simply does not offer.

This is why emerging technologies like zkTLS matter. Credit depends on facts: revenues, utilization, balances, insurance coverage. zkTLS makes it possible to verify those facts without fully revealing them or routing them through centralized intermediaries, allowing real economic activity to be underwritten onchain while preserving privacy and minimizing trust assumptions. Protocols like 3Jane use this to enable unsecured lending based on cash flows rather than collateral, while others apply similar primitives to infrastructure financing, institutional credit, or insurance underwriting, as seen in USD.ai, Wildcat, and Re. The common thread is the expansion of what can be verified, priced, and financed onchain.

More broadly, these systems point toward a future where sophisticated credit products are no longer gated behind institutional balance sheets or geographic boundaries. Retail capital can participate directly in financing infrastructure, underwriting risk, or extending credit to real businesses, with transparency and control that rarely exists in traditional markets. If this trajectory holds, onchain credit will not just compete with existing financial rails. It will materially widen the frontier of finance itself.

Credit Is the Real Test of Crypto

Credit is where financial systems are forced to confront reality. Defaults happen. Liquidity disappears. Risk concentrates in uncomfortable places. For most of its history, crypto avoided these constraints by design, leaning on overcollateralization and reflexive demand to scale quickly without confronting loss. What’s different now is that onchain systems are beginning to engage with these realities directly, not as failures to be patched over, but as inputs to be priced and structured.

If crypto can build credit systems that survive stress, allocate losses transparently, and expand access to sophisticated financial activity beyond traditional gatekeepers, then everything else follows. Not because yield is higher or rails are faster, but because the system is doing something genuinely new. Onchain credit is hard. That’s precisely why it matters.

Special thanks to Armen Ter-Avetisyan and Conor Moore for helping me with this post.

TLDR

AI is scaling into real-world constraints (chips, capital, governance, trust). Crypto gives us open markets for compute, data, identity, and payments: the primitives needed to build AI that’s cheaper, more resilient, and less centralized. There are two arcs:

1. Decentralized AI (DeAI): use blockchains and crypto incentives to build and coordinate AI infrastructure

2. Onchain AI: use AI to make blockchains useful to developers, users, and autonomous agents

I’m most excited about the agent stack: the frameworks, wallets, identity, coordination, and reputation layers that let autonomous systems transact and evolve onchain.

Why Now?

AI research, consumer and enterprise AI adoption, the discovery of new methods to apply AI in different contexts, and investment in AI are all exploding in one big positive feedback loop.

What many have recognized as the greatest technological platform shift in modern history, however, is starting to bring with it several challenges across the “AI stack”, including:

• Capex & supply: data centers are extremely capital-intensive, GPUs are increasingly scarce and politicized

• Governance & trust: compute and power are concentrated in the hands of a few companies, arising in governance risks, blind trust that AI outputs haven’t been manipulated, and increasingly gamed benchmarks

• Provenance & safety: deepfakes and data misuse erode trust; verifying authenticity and origin matters

• Access & limits: most commercial APIs throttle usage; developers hit rate limits and permission walls

Meanwhile, modern blockchains have matured. They’re open, permissionless, programmable, and verifiable – now fast enough, cheap enough, and scalable enough to support mainstream applications.

Crucially, they’ve found product-market fit: BTC as a store of value; stablecoins and DeFi for global, 24/7 settlement; perps and structured products for trading; and permissionless capital formation. Tokens have also proven they can coordinate resources and incentives at internet scale – the Bitcoin network, which coordinates and incentivizes BTC miners globally, is the first example of this, but there are many more.

These properties aren't just technical achievements; they’re the missing primitives needed to solve AI's most pressing challenges. Blockchains provide credibly neutral infrastructure for open markets, verifiable provenance, and decentralized governance – the things modern AI systems currently lack.

To better understand this convergence, let's explore the two primary arcs that define the Crypto x AI landscape today.

Two Arcs in Crypto x AI

Decentralized AI (DeAI)

DeAI – building AI infrastructure on modern blockchain networks – aims to solve today’s bottlenecks in AI and, in some cases, enhance AI’s capabilities.

Before the LLM wave, Ocean Protocol (2017) launched a decentralized data exchange, and Fetch.ai (2017) explored blockchain-native ML to power onchain autonomous agents. Post-ChatGPT, we’ve seen an explosion in DeAI. Company formation, talent density, mindshare, dollars invested, and token appreciation have all surged.

Over the past ~three years, we’ve seen companies across the AI stack attempt to tackle some of the aforementioned problems, including:

• Compute capacity – Akash & Aethir (permissionless marketplaces), io.net & Hyperbolic (decentralized networks), Exo (edge)

• Decentralized training & model collaboration – Prime Intellect, Nous Research, Bittensor, Sentient, Fortytwo

• Data layer – Sahara & Ocean (markets), Vana & Grass (collection), Story Protocol (provenance), Sapien (labeling)

• Verifiable inference – Inference Labs & Open Gradient

• Identity & Sybil-resistance – Worldcoin & Humanity Protocol

Projects have shattered expectations – decentralized training, for example, was once considered impossible, including by AI experts. However, breakthrough research like OpenDiLoCo (Prime Intellect), DeMo (Nous Research), and Protocol Models (Pluralis Research) together show that, for the first time, high-performance AI training is truly viable on decentralized networks. Many also doubted there’d be demand for open compute or onchain identity; io.net crossed $16M in network earnings with 17 million compute hours delivered, GPU usage on Akash is now at 77% (whereas Hyperscale Cloud GPU utilization averages ~50%), and Worldcoin’s onboarded 31M users.

FYI: I recently wrote a non-technical deep-dive on decentralized AI model training.

Onchain AI

While DeAI focuses on using blockchains to improve AI infrastructure, Onchain AI takes the reverse approach: it uses AI to help blockchains find PMF. It plays out in two ways:

1. Better experience for developers of protocols, apps, and agents

2. Better experience for users & agents – humans, enterprises, and autonomous agents operating onchain

Most funding and attention in Crypto x AI has gone to DeAI (infrastructure). Onchain AI is newer, sits higher up the stack, and is where I expect the first real “ChatGPT moment” for Crypto x AI that makes this all click.

1/ Developers

Coding agents are some of the best examples of PMF in AI, and the same will go for crypto. We’ll see:

• Fine-tuned models for Solidity/Rust/Move

• AI-native build surfaces: open, queryable SDKs and MCP-style servers that turn docs into conversational workflows

• Opinionated agent frameworks/toolkits with seamless integrations for robust onchain agents

Combined, this will dramatically lower the barrier to shipping crypto apps and agents.

2/ Users & Agents

The simplest-to-understand category here is better apps – apps that embed AI models and agents into their core functionality to make better products. For example, AI integrated into a DeFi app to maximize yield, improve execution, or better manage risk. Or agents that predict, at scale, whether wallets are sybil (see Vitalik’s post, AI as the engine, humans as the steering wheel).

More emergent is onchain AI agents – agents with wallets and stablecoins that follow programmable rules, transact with humans and other agents, and accumulate reputation/memory. The long-term view here is that these agents scale into the trillions and function like new, 24/7 “users,” driving more transactions and demand for blockchain rails. This is the agentic web, and blockchains matter here because they provide reliable money, self-custody wallets, and a verifiable, trustless environment with programmable rules for agents to operate in.

Onchain AI is nascent, but areas worth noting:

• Agent frameworks (Eliza OS, GAME)

• Agent launchpads and capital formation (Virtuals, Creator Bid)

• Agent payments (Nevermined, Payman)

• Agent orchestration (Naptha, Theoriq)

• DeFi agents (Bankr, Mamo)

• DeFAI apps (GIZA, Fungi)

Note: I recently wrote about the 2024 inception story of onchain AI agents.

The AI Agent Stack: My Investment Thesis

AI is the most powerful platform shift of our time, but it’s being shaped by a small group of companies racing to control every layer of the stack: compute, data, models, and distribution. I believe the future of AI shouldn’t be monopolized, and that the most transformative systems will be built on open, decentralized infrastructure.

AI agents are becoming active participants in software, and they need environments where they can hold assets, follow rules, and transact. Blockchains, which are now fast, cheap, and expressive enough to support real AI infrastructure, provide that foundation. Decentralized, credibly neutral, and user-owned, they offer exactly what today’s AI systems lack. And for the first time, they’re ready.

Just a year ago, decentralized training was dismissed as technically unfeasible, even by leading AI researchers. But breakthrough research like OpenDiLoCo, DeMo, and Protocol Models are flipping that assumption. For the first time, high-performance AI training is not only possible on decentralized networks – it’s working.

Exceptional teams are focused on core decentralized AI infrastructure. While I believe there will always be opportunities in improving compute, training, and data layers, I see the most practical wedge in agent infrastructure: the frameworks, wallets, and coordination and reputation layers that enable autonomous systems to interact, transact, and evolve. This is where the Crypto x AI stack starts to become real.

As primitives for identity, payments, memory, and coordination mature, we’ll move from isolated agents to persistent, interconnected networks of agents cooperating, competing, and evolving onchain. These agent networks will become the foundation for new types of products, experiences, and organizations. Over time, I expect this to unlock fully autonomous agent swarms, governed by cryptographic rules, not corporate APIs.

I’ve spent the past few years working closely with companies in decentralized AI, and I'm excited to continue to back ambitious, pre-product founders who are building years ahead of the market, starting from day zero. My DMs are always open.

Seven Predictions for the Crypto x AI Future

1. Decentralized training won’t beat SOTA (and that’s fine): Through 2027, decentralized training won’t beat centralized frontier models; its wins will be cost, resilience, transparency, and ownership. 

I think decentralized setups make most sense in terms of enabling domain-specific models, frequent fine-tunes, and in a regulated context that values auditability and control. What I’m less sure about is how decentralized training competes with Chinese open-source. 

2. Small models: By 2027, domain-specific models will power most real-world usage of decentralized training networks.

Related to the above, I think decentralized training and wins by fine-tuning lean models – or perhaps even incentivizing private pools of data to co-create small models – focused on specific use cases (biomed, legal, DeFi, etc). Small models are cheaper to train, easier to verify, and don’t require hyperscale infra. Their edge is performance-per-dollar and the ability to run on edge devices and agent wallets.

3. Provable inference becomes default in DeFi: By 2027, most major onchain DeFi protocols will require verifiable inference (TEE/zk) for any model-driven decisions.

Cheap TEEs and improving zk tech will mean we can attest that a specific model, with a specific input and version, produces a specific output that moved money. This feels inevitable to me, I’m just unsure of timing – can’t imagine users and regulators will stand for “trust the API” once TEEs/zk tech is ready.

4. An “agent passport” becomes a shared standard: By mid-2027, a common “agent passport” (keys, version, rules, and reputation) is adopted by major chains and apps.

As agents become active onchain participants (transacting, moving money, executing strategies, etc), apps need a shared way to know who they are, what they’re allowed to do, and how to limit them. I think a common standard emerges, embedding important information like wallet, rules, version, and reputation. It’s like a credit report for agents – used by protocols to appropriately gate agents. I think you first see this adopted in DeFi, but eventually directly into infra layers and AI-native chains looking to attract agents without introducing risk.

5. Agent transaction share: By July 2026, >5 % of all transactions on Base will be initiated by verifiable onchain agents (i.e., not sniping or MEV bots) that operate under transparent, pre-defined rules.

6. Agent-powered prediction markets reshape DAO governance: By 2027, major DAOs will incorporate agent-driven prediction markets as a core governance primitive. This is essentially AI-powered Futarchy, but where agents with wallets represent human tokenholders and use data-driven forecasts to place and resolve bets on proposals.

DAOs are one of the major reasons I got crypto-pilled in 2020/2021, but they’ve suffered from low participation and the Tragedy of Commons, which comes with a really bad negative feedback loop. I think AI agents can turn this into a positive feedback loop, where agents inform decisions, place probabilistic bets, and vote based on expected impact. Over time, you can imagine agent-governed prediction layers serving not only as governance/coordination tools for DAOs, but also as an informal reputational layer.

7. Onchain agent swarms coordinate by default. By 2027, at least one successful protocol or app will be built and operated entirely by a persistent, multi-agent system, with coordination, communication, and evolution handled onchain.

The closest thing we have to this is probably some version of Maker/Sky - which previously devolved into a DAO of subDAOs - I think we’ll see more experimentation at the edges here. Instead of “one agent, one task,” there’ll be agent swarms with shared memory, rules, and incentives, enabling full autonomy for protocols.

The more I’ve dug into this space, the more I believe the future of AI isn’t in centralized APIs, bigger models, and more data. It’s about building a better, more open foundation for them to run on.

These predictions are my current best guess, but the real fun is in the journey. If you’re a founder building something that challenges these ideas or brings them to life, I’d love to learn alongside you. My DMs are always open.

The Evolution of Human-Computer Interaction

The human-computer relationship to date has been centered around optimizing UX to make computers more intuitive and efficient. Early GUIs evolved into touchscreens and voice assistants - computers were tools, and we’ve exponentially sharpened them since the 1970s. A great coffee book that explores this is Web Design: The Evolution of the Digital World by Rob Ford.

However, with LLMs like Grok 4, which Elon claims is smarter than almost all grad students in all disciplines simultaneously, and more immersive AI products like Comet that are “designed to be a thought partner and assistant for every aspect of your digital life”, it’s clear we’re trending toward a world where UX becomes no-code, seamless, and driven by natural-language and intents.

Aravind Srinivas

@AravSrinivas

We're excited to finally release our next big product after launching Perplexity in 2022: Comet. Comet is a browser that's designed to be a thought partner and assistant for every aspect of your digital life: work and personal.

3,852

8:26 AM • Jul 9, 2025

An Agentic "Machine Layer"

With that in mind, I sense there's almost a new "machine layer" that'll sit between the existing human-computer relationship. For example, consider AI companions like those from Character AI or as portrayed in 2013’s film Her, where a digital entity becomes a confidant. This layer doesn’t just assist you - it interacts, adapts, and even influences you, blurring the line between tool and partner.

You can also imagine scenarios in the not-so-distant future where, given the recent acceleration in robotics, this “machine layer” takes on a life of its own. And develops its own character, akin to that of the human race.

Why Crypto Matters for Autonomous AI

At this level of scale, crypto becomes increasingly important because it helps ensure AI agents are truly independent and that their actions can be trusted. Just like we have our own identity and bank accounts, crypto gives AI agents a way to have their own secure digital "identity" and manage their own digital "money." 

This means AI agents can act independently, and centralized companies aren't able to manipulate their actions (and thus indirectly manipulate other humans). Relatedly, blockchains enable verifiability (the ability to prove that an AI agent actually did something without being tampered with, without revealing all of its inner workings), helping build trust in a world where AI will be doing a lot more on its own.

Crypto's First AI Agents

I think the first at-scale parallel to the Character AI-phase of innovation in crypto was Truth Terminal / $GOAT: an AI agent with a wallet and an X account whose prolific, viral posts led to memetic contagion and $GOAT’s token hitting >$1B cap.

GOAT was soon followed by other AI agents like AIXBT, who shared real-time crypto market insights on X in a distinctly crypto-native, “degen” style. $AIXBT hit >400K followers and a >$700M mcap.

In both cases, to those of us on CT/Crypto Twitter, these AI agents literally felt “amongst us” - i.e., they had a personality of their own with a unique voice, character, and impact.

Looking Ahead: Collaborative Agentic Networks

I see GOAT and AIXBT as Character AI-type moments - i.e., the seeds toward a world where AI agents are not just tools but co-creators of our reality. And that, in a few years, we’ll probably see successful at-scale networks exclusively for AI agents, where they share insights and build communities. I can also see robotics enabling the physical manifestations of this - combining digital charisma/personalities with tangible presence.

I think Freysa AI is one early experiment toward such a world: a network of sovereign AI agents that are personalized and cryptographically bound to their creators. Your “digital twin” retains context over time, remembers your preferences, goals, and interactions, and builds an evolving model of you. Simultaneously, it participates in the broader “Twin Network”, a collective of other people’s twins, and cultivates a collaborative, decentralized, and autonomous social network of AI agents. 

Looking forward, I think this “machine layer” means AI agents transcend their tool-like origins. By 2030, we’ll see networks like the Twin Network thrive, and I’m personally inspired by projects like Plastic Labs who are helping make agents more aligned and personalized through adaptive memory and social cognition.

Projects like Freysa and Plastic Labs - both leveraging crypto rails - hint at a future where agentic networks can thrive, reflecting human-like collaboration and creativity. And I'm excited by the potential to co-create with these entities! :)

Ready to Build?

If you're also excited, and especially if you're building a company in this category, please reach out. I'd love to connect!

This post is primarily a way for me to sharpen my own thinking in what is a rapidly evolving, highly competitive, relatively technical, and increasingly consequential domain. And, with that, gain perspective on where things are today and where they might be headed.

I’ll first overview the current state of centralized AI training, then outline the rationale for decentralizing it. From there, I’ll highlight research breakthroughs in distributed and decentralized AI, challenges facing decentralized AI today, and run through some cool decentralized AI companies.

I’m not trying to offer any particularly unique insights – but I hope my step-by-step breakdown is a helpful (and enjoyable) orientation! Note: this topic is technical and can quickly get unwieldy – I use LLMs to help me parse through papers and understand and explain things, and I recommend you do too!

TLDR

1. Centralized training is what OpenAI, Google, and Meta use to train AI models – i.e. running AI workloads inside single, massive data centers. But this approach is becoming unsustainable: “scaling laws” are plateauing, and the costs, energy, and supply-chain demands of ever-larger data centers are growing exponentially, quickly outpacing gains from Moore’s Law.

2. Distributed training splits large AI model jobs across various computational units (i.e., multiple GPUs or even multiple data centers), thereby reducing the load on any single location and unlocking further progress.

3. Decentralized training takes this one step further and seeks to provide an alternative to data centers with untrusted, heterogeneous smartphones, laptops, desktops, and colocation racks owned by anyone in the world. 

4. Until recently, bandwidth and coordination overheads made this decentralized concept impossible. However, recent breakthroughs like DiLoCo, Streaming DiLoCo, DisTrO, DeMo, SWARM Parallelism, and DiPaCo have made significant progress. 

5. While still early and evolving, two core pillars of decentralized AI infrastructure are seeing iteration and improvement: 1/ the design of sustainable crypto incentive mechanisms (i.e. programmatic token distributions to contributors of decentralized AI networks), and 2/ the ability to quickly and cheaply verify the authenticity of the heterogenous, untrusted compute contributed to AI networks (via zero-knowledge proofs, trusted execution environments, multi-party computation, and fully homomorphic encryption).

6. Continued progress in decentralized AI research and incentive and verification mechanisms are laying the foundation for a decentralized, global-scale GPU network that can train state-of-the-art AI models much cheaper than today’s centralized AI training costs, while also giving independent researchers, smaller labs, and the general public direct influence and financial upside over how the next generation of AI is built and governed.

1. From Single Racks to Global Swarms

Centralized AI training is what OpenAI, Google, Meta, and Anthropic use to build the large AI models that power apps you use daily – ChatGPT, Gemini, Llama, and Claude.

Centralized training involves building a data center and loading it with as many GPUs as possible. This approach stems from “scaling laws”, coined by OpenAI in 2020, which have consistently shown that models with more parameters, trained on more data with more compute, tend to perform better.

Leading clusters, like xAI’s Colossus and Meta’s Llama, each exceed 100,000 H100 GPUs. Meanwhile, Microsoft and OpenAI are rumoured to be planning 300,000+ GPU clusters by late 2025. But, while these centralized efforts continue to expand, problems are emerging with centralized training.

It is very capital intensive. AI data center capex is projected to hit $5.2T by 2030, with a single data center like xAI or Meta’s requiring ~$5-10B in upfront capex. Add $3-9B in hardware costs (at $25-30K per H100) and $500M-1B+ in annual OPEX for labor (40–60%), electricity (15–25%), cooling, and maintenance.

Beyond just cost, GPU supply is scarce and politicized, leading to 6-12 month wait times due to supply-chain choke points. And, with data center energy demand escalating, there is immense grid strain on local infrastructure.

Concentrating so much compute in so few hands also creates governance risk. Google’s 2024 “woke-filter” incident and OpenAI’s 2023 governance crisis highlight how small, opaque groups of people can shape model behavior at global scale and raise concerns about misuse, lack of accountability, and the imposition of a narrow set of values (think Orwell’s 1984 or the hyper-personalized manipulation loops of Black Mirror).

Relatedly, centralized AI models rely on a foundation of blind trust: users have no proof that AI outputs haven’t been tampered with. Centralized systems also struggle with brittle scaling, leading to inefficiencies and frustrating rate limits for API users of ChatGPT or Claude.

The way progress in AI is measured is also flawed – traditional benchmarks are increasingly gamed by centralized labs and fail to reflect real-world performance (see recent Meta’s Llama 4 controversy).

These escalating challenges demand a re-evaluation of how we build and deploy AI.

Multi‑datacenter (distributed) training helps, but the next leap is letting any node that meets a quality bar contribute computing power (decentralized training).

2. Why GPUs Became the Bottleneck

Centralization worked while scaling laws (more model parameters + bigger datasets + more compute ≈ better performance) still fit inside a single hyperscale campus. But they no longer do. To understand why, and how decentralized training presents an alternative, I’ll try break down the mechanics of machine learning, training large models, and then move through the challenges.

How Machine Learning Models Learn

At its core, ML model training is an optimization problem. The goal is to make a model perform a task, like predicting the next word in a sentence or identifying an object in an image. To do this, the model needs to learn from data.

Here's an overview of the process:

1. Prediction: The model takes some input data and makes a prediction.

2. Loss Calculation: This prediction is then compared to the actual correct answer. The difference is quantified by a loss function. A high loss means the model was very wrong. A low loss means it was close to correct.

3. Gradient Calculation: To improve, the model needs to know how to adjust its internal parameters (the weights and biases that define its knowledge). This is where gradients come in. Gradients tell us the direction and magnitude of change for each parameter to reduce its loss. This calculation process is handled by an algorithm called “backpropagation”. Think of it like finding the steepest downhill path on a complex, multi-dimensional landscape.

4. Parameter Update: Finally, the model's parameters are updated slightly in the direction indicated by the gradients, with the goal of minimizing the loss.

This process is repeated many, many times, often across billions or trillions of data points, until the model's performance on the task is satisfactory.

Enter: GPUs and Data Parallelism

Training very large models on massive datasets would take an impractical amount of time on a single computer. Distributed training can help expedite the process, and a common strategy within distributed training is data parallelism.

Data parallelism is where you:

1. replicate a full copy of the model across multiple GPUs, and

2. have each GPU process a different shard (or subset) of the training data simultaneously.

After processing its shard, each GPU calculates its gradients, and these gradients are then synchronized and averaged across all GPUs to update the model's parameters.

GPUs are incredibly good at data parallelism: they have thousands of processing cores, enabling them to perform operations like "predict-the-next-token" on different datasets simultaneously. This significantly expedites training with massive datasets, whether GPUs are in a single server, across data centers, or in a distributed setup.

The GPU Bottleneck: Gradient Synchronization and Network Limitations

While GPUs excel at parallel computation, a critical bottleneck for larger models in distributed setups is gradient synchronization.

After each GPU processes its shard and calculates its gradients, these individually computed gradients must be aggregated across all participating GPUs to ensure the model learns from the entire global batch of data and to keep all the model copies consistent.

This aggregation process, which happens after every training step, requires a huge data transfer between all devices. For the largest models, which can have trillions of parameters, the size of these gradients is enormous.

This constant, massive data transfer places a huge demand on network infrastructure.

Tightly integrated data center campuses use specialized high-speed, low-latency interconnects (like NVLink or InfiniBand) to handle this traffic effectively. But once the setup scales to different data centers, multiple campuses, or becomes geographically dispersed, network links stretch out, leading to much lower throughput and significantly higher latency.

At this point, the network can't keep up. GPUs spend more time waiting for data to sync than performing computations, drastically slowing the training process. The network thus becomes the critical bottleneck, preventing global-scale centralized AI training.

Beyond Raw Compute: The Evolving Landscape of Model Training

Despite these challenges, the large model development landscape has seen a shift since 2024. To understand this, it's helpful to clarify the two main stages of large model training:

• Pre-training: This is the initial, computationally intense phase, which I’ve focused on thus far. Here, models learn general language patterns, facts, and reasoning abilities by processing vast amounts of raw, unlabeled text and code. The goal is to build a foundational understanding, typically by predicting the next token or filling in masked words.

• Post-training (or Fine-tuning): After pre-training, we try to align model behavior with human preferences, improve their ability to follow instructions, and reduce hallucinations. This involves techniques like Supervised Fine-Tuning (SFT) on curated human-labeled data and Reinforcement Learning from Human Feedback (RLHF) or other forms of Reinforcement Learning (RL), where the model learns by receiving rewards or penalties based on its outputs.

With these stages in mind, we've seen two significant developments:

1. Smarter Training and Inference: Reasoning models, like OpenAI’s o3 and DeepSeek’s R, have demonstrated huge performance gains, not just from continuously scaling raw pre-training compute, but from advanced RL and SFT (which involve their own train-time compute), alongside strategic applications of increased test-time compute for reasoning (e.g. generating pre-answer tokens, efficient inference, or prompt engineering). Sam Lehman’s “The World’s RL Gym” offers an excellent deep dive, highlighting opportunities for decentralized AI.

2. Efficiency Through Innovation: DeepSeek successfully trained powerful models on a relatively modest 2,000 H800 GPUs by strategically combining advanced techniques, like Mixture-of-Experts (MoE) architectures, highly optimized low-level CUDA tweaks, and novel optimization algorithms like Group Relative Policy Optimization (GRPO).

Together, these developments demonstrate that architectural innovation and training methodologies are as crucial as raw computational power, opening new avenues for achieving frontier-level model capabilities with a more constrained hardware footprint, challenging the dominance of hyperscale data centers.

3. Overcoming the Bottleneck: Distributed Training and the Path to Decentralization

So – the massive data transfer required for gradient synchronization forms a critical network bottleneck, preventing centralized AI training from scaling infinitely, especially when attempting to span across geographically dispersed locations.

This challenge has fueled the development of various distributed training strategies to make larger models feasible. The table below aims to clarify the different approaches:

Thus far, I’ve focused on the mechanics and limitations of centralized training. Now, I’ll dive into the techniques that underpin distributed training (methods used by hyperscalers to further scale models) and then pivot to recent breakthroughs making decentralized training a reality.

Scaling Beyond Single GPUs: The Three Parallelism Axes of Distributed Training

To manage the immense size of modern AI models and the vast datasets they consume, distributed training leverages three “parallelism axes” to spread a training job across many GPUs and even many servers.

1. Data Parallelism. As previously discussed, data parallelism involves replicating a full copy of the model across multiple GPUs, with each GPU processing a different shard of the training data simultaneously. It’s effective for scaling with large datasets, but struggles with the constant, massive synchronization of gradients across the network, which becomes increasingly problematic as training becomes geographically dispersed.

2. Model Parallelism: Instead of replicating the full model, different parts of the model are distributed across multiple devices. This is helpful when the model itself becomes too large to fit entirely into a single GPU and enables the training of truly massive models that would otherwise be computationally impossible. Within model parallelism, there are two sub-strategies: 

   1. Tensor Parallelism: this strategy involves splitting the computation within a layer across multiple GPUs, i.e. one matrix multiplication might span several GPUs that then exchange partial results constantly. This helps in training large models by distributing memory load and computation, but it also introduces heavy, low-latency communication demands, typically handled by very fast interconnects within a single server or closely-packed servers. 

   2. Pipeline Parallelism: this strategy divides the model’s layers into sequential stages, assigning each stage to a different GPU or group of GPUs. Data/inputs pass forward through this “pipeline”, with each stage computing its part, and then gradients move backward through the pipeline. This spreads memory load and reduces communication requirements, but creates “bubble” or idle time, as each stage waits its turn in the sequence. Modern implementations use techniques like micro-batching to fill these bubbles and reduce idle GPU time.

Modern distributed algorithms often combine all three of these techniques (3-D parallelism) to compress synchronization steps, thereby reducing the burden on the network. But even these advanced strategies face inherent limits when scaling beyond tightly integrated, single-operator data centers, primarily due to fundamental network limitations, as discussed previously.

Recent Groundbreaking Techniques: Making Decentralized Training a Reality

The classic distributed training approaches (Data, Tensor, and Pipeline Parallelism) excel within highly integrated, single-operator environments like large data centres. But they struggle to scale to truly global, heterogeneous, and trustless compute networks. 

Until recently, the communication overheads and coordination challenges inherent with decentralized compute networks have made true decentralized training seem impossible for state-of-the-art models. But research breakthroughs over the past few years have introduced novel approaches that directly tackle these barriers: 

Individually, these breakthroughs are impressive! But when combined, they hint at the potential for a $$10^5$$ reduction in communication between nodes – which should be sufficient to move from high-bandwidth InfiniBand interconnects within data centres to efficient training over consumer-grade fiber optic internet connections. This demonstrates the potential for a globally distributed, decentralized compute network for AI.

While these advancements address a fundamental bottleneck, bridging the gap to SOTA centralized AI models is still a long way off; it’s not just about communication, it's about the total available dedicated compute, specialized interconnects within single clusters, and optimized software stacks.

There are also important challenges that need to be overcome in establishing trust and verifiability in an open network, and designing robust economic incentives. The goal here isn’t just to equal the efficiency of centralized AI training, but rather to enable a new paradigm where AI can be trained on a far greater scale and accessibility.

4. Trust and Verifiability in Decentralized AI

The communication breakthroughs discussed above are indeed incredible, but there’s still a huge problem facing decentralized training: trust and verifiability.

In a centralized system, you trust the single entity (e.g., OpenAI or Google) running the data center. But in a decentralized network, where compute providers are unknown and potentially malicious, this trust assumption disappears.

In other words, for decentralized AI systems to truly scale, participants need 100% guarantees: that the computations they receive are performed correctly (without errors or malicious alterations), and that sensitive training data remains private. It also means, in the case of decentralized inference, knowing that if you ask the network a question using Llama 4, it's not returning with a question answered by Llama 2 – giving you a worse analysis and profiting from the discrepancy.

This is especially important for high-value use cases like finance, healthcare, defense, and legal compliance. Enterprises in these categories are likely willing to pay a premium for provable correctness, as even a single incorrect inference could trigger expensive recomputations or regulatory penalties.

This is where verifiable compute comes in – introducing methods to cryptographically verify correctness into AI pre-training, post-training, and inference. Here’s a summary of the core approaches:

Privacy-First Verification

1. Trusted Execution Environments (TEEs): TEEs (like Intel SGX or AMD SEV) are secure, isolated "black boxes" within a CPU. Code and data enter, are processed privately and securely, and results exit – all without the host machine being able to see or tamper with what's inside. For decentralized AI, TEEs offer hardware-backed verification: if computation occurs within a TEE, you get a strong guarantee of its integrity. The main limitations are that they're hardware-specific, which means limited availability, and they can bottleneck with very large models due to memory constraints.

2. Multi-Party Computation (MPC): MPC is a cryptographic technique enabling multiple parties to jointly perform a computation without revealing their individual private data (e.g., a model can be trained collaboratively across different organizations, each contributing their sensitive datasets, but no single party ever sees the others' raw data). This is transformative for privacy-preserving AI, allowing collaboration in highly regulated sectors like healthcare or finance. But MPC is computationally intensive, making it significantly slower and more resource-hungry than computing on plaintext, and its efficiency often decreases with more participants.

3. Fully Homomorphic Encryption (FHE): FHE is often considered the "holy grail" of encryption because it allows computations to be performed directly on encrypted data without ever decrypting it. Only the data owner can decrypt the final output. For decentralized AI, FHE offers the highest level of privacy: you can send encrypted data to a network, have it processed by an encrypted model, and receive an encrypted result – with no node ever accessing unencrypted information. While incredibly powerful, FHE is still largely an academic and research-intensive field for practical AI applications due to its extremely high computational overhead, often making it orders of magnitude slower than plaintext operations.

Correctness-First Verification

4. Zero-Knowledge Machine Learning (zkML): zkML uses Zero-knowledge Proofs (ZKPs) to cryptographically verify that a computation was performed correctly, without revealing any underlying data. This is important for decentralized AI because it enables nodes to prove they faithfully executed a model inference or gradient update without exposing proprietary model weights or sensitive input data. Generating these proofs remains computationally expensive, but overheads are coming down, making them increasingly viable for practical applications.

5. Optimistic Machine Learning: Applies the “innocent-until-proven-guilty” approach to ML – i.e., instead of proving every computation, optimistic ML assumes computations are correct unless challenged. Haseeb does a great job explaining this in detail. If a participant suspects fraud, they can initiate a "dispute" that re-executes or verifies the original computation onchain. This approach can significantly reduce overheads and is ideal for scenarios where verification costs are higher than the expected frequency of fraud. However, there's an inherent latency in dispute resolution (i.e., results aren't immediately final), and it relies on honest "challengers" to monitor the network.

Twill at Delphi Digital put out a decentralized AI report recently, including a table that helps quickly grasp the nuances and trade-offs of each approach:

Ultimately, these diverse technologies are critical enablers for decentralized AI. While no longer purely theoretical, they are still relatively early-stage and not yet battle-tested at enterprise scale or high volume. They represent crucial work in progress, with significant potential to build integrity and privacy into the AI stack as they mature.

5. Incentivizing a Decentralized AI Ecosystem

Rapid advancements in consumer-grade hardware mean that powerful AI capabilities are no longer confined to hyperscale data centers. Modern consumer GPUs like M-series Macs and the latest 4090s can now not only fine-tune but even pre-train billion-parameter models, with projects like Exo Labs showing internet‑scale runs reaching 100+ GPUs with reasonable throughput.

Given this, and assuming we solve all communication and verification issues, the question remains: why would anyone contribute their power-hungry GPUs for compute tasks? They could use them for other purposes… or just turn them off. This highlights the need for a robust incentive layer. 

The goal is to create an economic system that aligns all participants and ensures the long-term health and growth of the network: incentivizing GPU providers (for reliable compute), data contributors (for high-quality data), and model trainers (for accurate models), while deterring malicious or inefficient behavior (e.g., offline nodes, fraudulent computations).

While specific economic systems, token models, and incentive mechanisms deserve nuance and a separate blog post, what you need to know is that the crypto industry has speedrun experimentation in, and has become very good at, incentive design. Strategies relevant for decentralized AI include: using programmatic and formulaic rewards for compute, data, and developer contributions, staking and slashing to disincentivize bad behaviour, and reputation systems with community governance to cultivate long-term alignment. 

With this incentive layer solved, as Twill also eludes to in his decentralized AI report, the view for decentralized AI is to create a powerful flywheel/feedback loop:

1. Research breakthroughs → decentralized training on globally dispersed consumer hardware

2. Better and cheaper consumer hardware and staking yields → cheaper cryptographic verification

3. Cheaper verification → more node providers and validators join the network

4. More network nodes → richer decentralized training and better models

5. Better models → more network value, model deflation (downward pressure on the unit cost of model capability), and better staking yields

The loop then closes, and this is essentially the end-game for decentralized AI training.

Each cycle lowers the cost of creating trustworthy AI and draws more participants into the network, enriching the training environment with diverse geographies and data sources and, ultimately, overcoming centralized AI’s problems outlined in Section 1 cost, energy, supply chain, governance, trust, brittle scaling, etc).

Now, let’s run through some cool companies building in this category! 

6. Leading Decentralized AI Projects

Pluralis Research

• Focus: Pioneering asynchronous model parallelism and structured compression for decentralized training. Pluralis focuses on "Protocol Learning," where model weights are sharded across nodes, ensuring no single node can reconstruct the full model, thus enabling monetizable, non-extractable models.

• Key Tech: Employs an asynchronous pipeline-based parallelism architecture (SWARM) and innovations like "Column-Space Sparsification" for over 90% communication reduction, and Nesterov-based asynchronous gradient correction for Pipeline Parallel (PP) setups. Despite these advances, achieving parity with centralized training requires approximately 300x compression, a challenge that remains Pluralis's core focus.

• Progress: R&D-centric. Published the 'Beyond Top k' paper demonstrating over 90% compression in inter-node communication for model-parallel setups. They also have other works showing heterogeneous device support in Pipeline Parallel (PP) configurations. Their Nesterov method for PP was accepted into ICML 2025. In June 2025, Pluralis released research extending their framework to fine-tuning, demonstrating an experimental 8B LLaMA model trained across 64 GPUs in 4 geographical regions via pipeline parallelism. No public product or testnet yet. 

• Funding: $7.6M Seed in March 2025, co-led by USV and CoinFund.

Nous Research

• Focus: Decentralized, open-source, human-centric AI models and tools, aiming to democratize AI development by leveraging globally distributed compute and blockchain incentives. Abhay at Nous has a good thread with high-level resources on what they’re up to. 

• Key Tech: Developed the Psyche Network and the DisTrO (Distributed Training Over-the-Internet) optimizer, which achieves extreme communication compression (up to 10,000x reduction via DCT and 1-bit sign encoding), asynchronous fault tolerance, and decentralized scheduling.

• Progress: Have developed highly-steerable models through their Hermes series of fine-tunes and successfully trained a 15B-parameter model using DisTrO. More recently, in late-May 2025, they started pre-training their new 40B-parameter Nous Consilience model on the Psyche network and have shown promising loss reduction and perplexity reduction graphs (see below)

• Funding: $70M total, including a $50M Series A @$1B led by Paradigm in April 2025.

Prime Intellect

• Focus: Decentralized training network where anyone can participate and receive verifiable rewards for compute contributions, primarily for RL-based decentralized training.

• Key Tech: PRIME-RL (asynchronous RL framework), TOPLOC (lightweight behavior verification without costly zkML), SHARDCAST (asynchronous weight aggregation via gossip protocols), and OpenDiLoCo/PCCL (sparse asynchronous communication optimized for low-bandwidth, heterogeneous devices).

• Progress: In May 2025, released INTELLECT-2, a 32B-parameter model fine-tuned entirely via trustless decentralized collaboration across 100+ heterogeneous GPUs on 3 continents, showcasing the feasibility of "training as consensus" with full transparency.

• Funding: $20.5M total, including a $15M Series A led by Founders Fund in March 2025. 

Gensyn

• Focus: A verifiable execution layer for decentralized AI training, aiming to turn global idle compute into a massive open AI cluster (training-as-mining). It acts as a protocol layer that supports task distribution, execution, verification, and incentive allocation.

• Key Tech: RL Swarm (decentralized collaborative RL for post-training), Verde (a hybrid verification system balancing verifiability and efficiency via minimal recomputation), and SkipPipe (fault-tolerant routing for unstable networks, improving pipeline training speed). Uses a multi-role game-theoretic incentive system.

• Progress: Showing great results in decentralized RL post-training. This process takes a strong base model, gives copies to participants who generate reasoning traces, which are then collected and used to improve the base model. This approach is significantly cheaper than pre-training as nodes primarily perform inference, though it retains a dependency on the quality of the base model. Gensyn is currently in testnet phase for its RL swarm, conducting permissionless post-training of 0.5B to 72B models using reinforcement learning on a custom Ethereum rollup.

• Funding: Over $50M total, including a $43M Series A led by a16z crypto in May 2023.

Flock

• Focus: Onchain federated learning that decentralizes training across data, computation, and models. Unlike pure decentralized training, Flock integrates traditional federated learning with a crypto-native incentive layer, prioritizing privacy and usability.

• Key Tech: Adopts the standard Federated Learning paradigm, allowing data owners to train locally and submit aggregated updates on-chain. Integrates VRF-based random selection, PoS staking, and programmatic incentives. Notable for zkFL, a zero-knowledge federated learning scheme for privacy-preserving gradient aggregation.

• Progress: Has an active platform with 6,620 models created, 176 training nodes, 236 validation nodes, and 1,212 delegators. Launched products like AI Arena (training platform) and FL Alliance (federated learning client).

• Funding: Over $9M raised across two rounds in 2024, with investors including DCG and Lightspeed Faction. 

7. Conclusion

Centralized AI models are absolutely crushing it. The progress made since ChatGPT launched in November 2022 has been phenomenal. However, there are signs that scaling laws, which have fueled this progress, are beginning to wane. 

Meanwhile, expectations around AI’s impact on the economy have soared – with some viewing AI as the primary path the US will grow out of a self-inflicted debt cycle (in combination with a renaissance in energy and 0-to-1 moment in robotics).

This massive AI TAM, combined with the eye-watering valuations that top AI labs are commanding, has given birth to opportunity for disruptive startups in decentralized AI trying to take at least a small piece of centralized AI’s market share. 

There is, without a doubt, massive market risk and perhaps the most intensely competitive field we’ve ever seen in technology – centralized AI companies are incredibly well capitalized and are elite at execution. But, given the TAM, founders and VCs alike are super excited. 

Personally, I find the AI model training and decentralized AI categories super interesting and am easily nerd-sniped by the principles behind and the potential of decentralized AI. 

However, as an early-stage VC, I don’t think there’s much opportunity left to invest in the category: it’s already on everyone’s radar, the valuations are already high, and the category leaders seem to have been founded and backed at Series A by Tier 1 VCs. 

At this point, this may devolve into a knife fight between players like Nous and Prime, who are competing for attention and GPUs amongst participants, community members, and retail degens speculating on airdrops. 

While I’ve thoroughly enjoyed going deeper and researching decentralized AI training, and am truly excited about the possibility of a global-scale GPU swarm training open models with distributed ownership and governance, I’m left wanting to explore more emergent areas of decentralized AI innovation:

I think capital formation for agents (Virtuals), adaptive memory and social cognition for agents (Plastic Labs), payments for agents (Nevermined), sovereign agents (Freysa), agent orchestration (Naptha), and swarm inference (Fortytwo) are all examples of exciting and emergent areas (and will report back in future blog posts on what I find!).

If you’ve made it this far, I’d love to hear your thoughts on decentralized AI in general, and also any emergent areas you’re excited about. If you’re thinking of building a company in this space, or are already in the trenches, I'd love to connect and hear your insights – please reach out!

1/ DePIN
I started my career at a hedge fund, investing in large-cap oil, gas, metals, mining, renewables, utilities, and telecom companies – all capex-heavy industries that benefit from stable regulations and predictable cash flows to reduce cost of capital. But the pace of innovation in these systems and centralized organizations is slow (e.g. in utilities/energy) and pricing is often quite extractive for the end consumer (e.g. in telecoms).

I think DePINs represent a fundamental shift – creating scalable, resilient, and community-driven networks that offer better, cheaper products (e.g. @Hivemapper, @GEODNET_). Though nascent, I’ve seen DePINs explore deep/hard-tech use cases like drones, batteries, and robotics, which is exciting! Downstream of this, I’m hopeful DePINs have real world, tangible impact that’s easily explainable and will help evidence crypto as tech that creates value and utility for households and SMEs worldwide. Very cool to see @santiagoroel and @inversion_cap be ambitious and jumpstart this!

2/ Investment Banking
I’ve seen both the buy-side (hedge fund) and sell-side (m&a, ecm, leverage finance) in tradfi. While the system’s not perfect, it’s mature and generally effective at helping companies access at lower costs and with better liquidity. Capital markets reward the best companies and punish the bad ones, creating a self-regulating mechanism that drives efficiency, innovation, and disciplined capital allocation.

In contrast, crypto capital markets remain highly fragmented, with poor liquidity and price discovery. There’s a lack of structured frameworks, institutional participation, and rigorous underwriting that tradfi has refined over decades. I think there’s a big opportunity, which will likely get filled this cycle, to bring institutional-grade execution, transparency, and investor alignment to the TGE process and create a crypto-native investment bank. I actually think this is more than just “an opportunity”, but a necessary step toward maturing crypto as an asset class. H/t @bennypjacobs, who’s furthered my thinking here.

3/ Stablecoins
I grew up in a family that ran international businesses, with relatives across the world, including in developing countries like Turkey and Armenia. Watching stablecoin adoption skyrocket in recent years has been awesome – finally addressing issues I witnessed firsthand, like gaining access to USD as a hedge against rapidly devaluing currencies (e.g. the Turkish lira) and enabling seamless global payments for employees and supply chains. I’m excited to see this continue – for SMEs like restaurants to add 2% to their bottom lines by saving on credit card fees and individuals in emerging markets to save 5% on transfer fees.

But I’m even more stoked about what’s next: stablecoins evolving into a fully integrated financial infrastructure. As adoption grows, we’re moving beyond simple value storage and payments toward programmable money, yield-bearing stablecoins, and seamless unchain financial services that will outperform tradfi (@SquadsProtocol is an early example, h/t to the relentless hustler @SimkinStepan).

4/ Web Proofs (zkTLS)
I’ve been pilled on zkTLS ever since I met @Euler__Lagrange in early 2024. I got into crypto in 2020, fanboying over @a16z / @cdixon essays about the future of the internet and how web3 would tear down web2 walled gardens. That hasn’t happened yet, but zkTLS feels like our best shot at unlocking a new wave of web3 consumer apps.

zkTLS lets web3 companies vampire attack web2 incumbents. Say, for example, you want to build a web3 Uber or a web3 Instacart (like @NoshDelivery) – the cold start problem is real. Drivers and food providers would lose their web2 reputation when moving to a new network, making adoption tough. But zkTLS lets users seamlessly port their data from web2 apps, making it easy to bootstrap new apps and ecosystems on web3 rails. The potential here is massive.

Have learned a bunch more on zkTLS from @yeak__ and @Danconia_Crypto + h/t @adam_winnick and @dawufi for being some of the most passionate investors/builders in zkTLS I know!

5/ Sovereign Agents
The Crypto x AI Agent narrative went wild in Q4 2024. It’s a shame, as hype cycles generally tend to create lots of noise while burying genuinely good ideas, but I hope sovereign agents endure! Web 2 AI agents are awesome will help regular people and businesses automate simple tasks. But crypto-native sovereign agents are uniquely exciting because they can operate autonomously, own assets, and execute transactions without relying on a human or corporate intermediary.

They function independently and also prove their integrity through cryptographic verification, enabling things like AI-driven businesses, research groups, or investment funds that aren’t owned by anyone but still make decisions on their own. It also opens the door for AI governance models and net-new organizations (see @AgoraGovernance and @hatsprotocol), where decisions come from verifiable code rather than people with hidden agendas or inflated egos. I think this concept has the potential to redefine how AI participates in the economy – not just as a tool augmenting human capabilities, but as actual economic actors in their own right.

A few serious projects in this category include @Freysa_ai (building tamper-proof AI agents), @NapthaAI (building a platform to orchestrate sovereign agents and distributed workflows), and @Nevermined_io (building decentralized payments infra for AI agents to transact without the fear of being cut off from financial rails by centralized actors).

Fin
Taken together, I think DePIN, efficient token capital markets, stablecoins, zkTLS, and autonomous AI could create a much more interconnected and resilient ecosystem in crypto – one where decentralized networks and programmable money transform industries. Zooming out, and considering the prospect of an improved regulatory landscape, it’s an incredibly exciting moment to be in crypto. I’m stoked to see how these innovations and projects redefine what’s possible.

Instead of being controlled by a single entity, DePINs are owned and operated by participants who contribute resources like energy, bandwidth, or storage. In return, they earn token rewards programmatically. The physical layer is built and installed by a global, decentralized community, consisting of things like energy devices, wireless routers, and IoT sensors.

DePINs represent a fundamental shift in how we think about infrastructure, replacing traditional systems with scalable, resilient, and community-driven networks. And within this category, one of the most transformative opportunities lies at the intersection of crypto and energy.

Crypto x Energy: Unlocking the Grid’s Potential

The global energy system is facing unprecedented challenges:

1. The aging electric grid, with fragmented regional operations managed by outdated control architectures.

2. The growing demand for AI compute and the reshoring of US manufacturing further exacerbates the grid’s limitations.

3. The inherently variable supply of renewable energy (wind and solar), essential for satisfying climate goals, adds volatility to an already fragile grid.

These challenges are compounded by the rapid growth of distributed energy resources (DERs) like home solar panels, batteries, and EV chargers. The traditional grid model, where centralized power plants distribute electricity to passive consumers, is already giving way to a new era where every homeowner can become a power producer, highlighting the need for a more resilient, decentralized approach to energy infrastructure.

Crypto-powered Energy DePINs offer elegant solutions to these pain points. By redistributing CapEx and leveraging blockchain incentives, DePINs can address inefficiencies in energy production, distribution, and financing while enabling powerful new business models such as:

• Decentralized Energy Storage and Distribution: Virtual Power Plants (VPPs) allow individuals to store and sell energy, balancing supply and demand in real-time.

• Accelerating Renewable Energy Buildout: Platforms unlock innovative financing methods and incentivize solar, wind, and battery deployments.

• Improving Grid Efficiency: Localized, distributed nodes reduce latency, optimize resource allocation, and alleviate congestion.

Builders Tackling the Energy Transition

The transition to decentralized energy is already underway. Some notable projects leveraging DePIN principles include:

Plural Energy
Plural Energy is a blockchain-enabled platform that democratizes access to renewable energy investments, allowing anyone to invest in high-yielding clean energy projects with full transparency and liquidity. Traditionally, renewable infrastructure deals have been complex, capital-intensive, and limited to large investment banks or high-net-worth individuals. Plural democratizes access to this "missing middle" of renewable energy investments, enabling anyone to invest in high-yield, consistent, and previously inaccessible clean energy projects. Plural utilizes smart contracts to provide full transparency, real-time project performance data, predictable yields, and liquidity through the ability to buy, sell, or trade investments at any time. Plural has active portfolios live today.

Daylight
Daylight is building the world’s largest decentralized energy network, enabling households to generate, store, and contribute clean energy while earning rewards. By connecting smart energy devices like solar panels, batteries, EV chargers, and thermostats, Daylight transforms homes into decentralized power nodes that strengthen the grid and reduce reliance on centralized infrastructure. Its marketplace simplifies the adoption of energy upgrades, e.g. solar, heat pumps, and more, while its protocol coordinates energy contributions and rewards participants with its native energy-backed token. By combining blockchain technology with real-world energy solutions, Daylight provides a sustainable, cost-effective alternative to traditional grids, accelerating the global transition to clean energy.

Power Ledger
Power Ledger is a decentralized platform that enables peer-to-peer (P2P) energy trading and facilitates access to environmental commodities markets like carbon credits and renewable energy certificates. Its core pillars include energy trading and traceability, which allow users to track and trade excess energy seamlessly, and commodities trading, providing markets for energy derivatives and environmental assets. By empowering individuals to trade energy directly and connect distributed energy resources (DERs) with energy companies, Power Ledger offers a solution to inefficiencies in grid energy management and carbon markets.

Sourceful 
Sourceful Energy is building a global coordination layer for the $1T+ decentralized energy market, connecting and managing distributed energy resources (DERs) such as solar panels, batteries, and EV chargers. By integrating real-time optimization, Sourceful enables DER owners to participate in Virtual Power Plants (VPPs), earn revenue for grid stabilization, and monetize energy data, all tied to their Solana wallets. Unlike single-device solutions, Sourceful Energy’s open-source firmware and hardware-agnostic approach ensure compatibility across brands, driving fast scalability and compounding network effects.

StarPower
Like Daylight and Sourceful, StarPower is a decentralized energy network that connects and optimizes energy devices like air conditioners, water heaters, EVs, and home storage batteries to improve energy efficiency and reduce costs. Positioned as the “Uber for energy,” Starpower aggregates these devices under one platform, enabling smarter operations and creating Virtual Power Plants (VPPs) to stabilize grids.

Users can connect their devices through Starplug, a real-time monitoring and control solution that optimizes energy use while rewarding participants with $STAR tokens based on their electricity data and responsiveness. In addition to the Starplug, Starpower offers the Starbattery, a home energy storage system that manages excess energy, provides backup power, and reduces peak consumption costs, as well as the Starcharger, which optimizes EV charging during off-peak hours to alleviate grid strain.

Arkeen
Arkeen is a DePIN that integrates small-scale renewable energy producers into a decentralized, carbon-neutral grid. It uses a unique, IoT-based Proof of Physical Work (PoPW) system to validate individual participants’ environmentally-friendly actions like household solar power generation, energy-efficient home air conditioning, and the use of electric vehicles. These actions are rewarded with AKRE tokens, creating an incentive-driven system for renewable energy adoption. Additionally, Arkeen has a GreenBTC Club initiative that ties renewable energy to Bitcoin mining, addressing the carbon footprint of one of the most energy-intensive blockchain applications.

DeCharge
DeCharge is a decentralized, community-powered EV charging network that leverages DePIN architecture, blockchain, and OCPP standards to create reliable, tokenized EV infrastructure. By enabling homeowners, businesses, and cafes to host charging stations and earn passive income, DeCharge accelerates the adoption of sustainable energy and transportation solutions.

Glow
Glow is a blockchain-powered platform designed to accelerate the deployment of solar farms by addressing inefficiencies in renewable energy incentives. Its unique model requires participating solar farms to contribute 100% of their gross electricity revenue to a USDC-based incentive pool, ensuring that rewards and carbon credits go to the most efficient and impactful projects. Glow operates a dual-token system: GLW tokens for rewards distribution and GCC tokens representing verified carbon credits, with off-chain audits handled by Glow Certification Agents (GCAs) for transparency. Glow tackles the challenges of carbon credit volatility and underinvestment in renewable infrastructure by aligning incentives with efficiency and impact, driving meaningful progress in clean energy adoption.

But amidst all this innovation, there’s still a massive white space: Crypto x Energy Storage. Batteries are central to solving grid challenges and enabling renewables because they address the inherent intermittency of wind and solar, storing excess energy when the sun shines or the wind blows and redistributing it when demand peaks or generation drops. Without effective energy storage, renewable power remains unreliable and difficult to scale.

Crazy to think that if solar PV and battery efficiency both compound at 7-10% per year for 30-50 years, the world could be powered entirely by sunlight, and carbon emissions would collapse. 🌞 Batteries, when broadly distributed and integrated, can unlock this future, and DePIN could play a critical role in incentivizing adoption and driving innovation.

Here’s 5 startup ideas to stimulate discussion and innovation in the space:

1. Battery Health & Data Monetization
A DePIN where users contribute real-time battery performance and health data (e.g., EVs, home storage) to a shared database. Devices would use edge computing to process and format the data before securely sending it to the network, ensuring privacy and reducing unnecessary data load. Researchers, manufacturers, and grid operators can purchase this data to optimize battery design and grid integration, while contributors earn tokens for sharing verified insights.

2. Incentivized Battery Recycling & Second-Life Usage
A DePIN network that tracks batteries through their lifecycle and incentivizes proper recycling or repurposing. Tokens are awarded to recycling centers, logistics providers, and individuals for verifiable recycling efforts, ensuring batteries are reused in energy storage systems or other applications. Note: with the rise of EVs, home storage systems, and portable batteries, battery recycling is a significant and growing challenge.

A potential GTM strategy could start with initial partnerships with recycling centers to provide token incentives for battery collection. You’d then work with EV and battery manufacturers to integrate digital IDs into batteries for lifecycle tracking and create consumer incentives by offering rewards for returning end-of-life batteries via a user-friendly app with drop-off locations.

3. Battery Swapping Networks
A DePIN that connects EV users with battery-swapping stations. These stations can be hosted by individuals, businesses, or communities, creating a distributed network. Battery swapping helps address long charging times and inconsistent energy availability for EVs. Instead of waiting for a battery to charge, EV owners can swap depleted batteries for fully charged ones at swapping stations. Participants earn rewards for operating or contributing to the network, while EV owners benefit from faster, more efficient energy access. This helps reduce EV downtime, alleviates grid stress during peak hours, and ensures more widespread adoption of electric mobility.

4. Grid Resilience via Decentralized Battery Networks
A DePIN which connects batteries (home storage systems, EV batteries, or portable power units) to create microgrids that stabilize the local energy supply. During periods of grid stress, outages, or failures, these batteries can autonomously redistribute surplus energy to homes and businesses within the network, ensuring a stable local power supply. The DePIN protocol uses smart contracts to coordinate energy flow and optimize contributions in real-time, rewarding users with tokens for sharing their stored energy when it’s most needed.

5. Dynamic Pricing for Battery Charging
A decentralized protocol that uses real-time market data to optimize battery charging schedules based on grid prices and demand. Contributors who shift their energy consumption to off-peak hours are rewarded with tokens, improving grid stability.

Batteries are critical to our clean energy future, and Crypto x Battery DePINs represent one of the most exciting, untapped opportunities to unlock that potential. If you’re building, or thinking of building, in this space, I’d love to hear from you 🙏

This is inspired by Stoic philosophy & Charlie Munger’s concept of “inversion” (https://jamesclear.com/inversion) - approaching problems by asking, “Why not?” rather than assuming the status quo. This helps challenge assumptions, biases, and blind spots.

I notice this most during discussions about polarizing topics like the wars in the Middle East, DJT’s election, crypto, AI regulation, etc.

Instead of trying to “win” a debate or convince someone to think the same way, it’s far more productive—and enjoyable—to act as an intellectual thought partner.

Ask thoughtful, inquisitive, open-ended questions to gain empathy and help the other person critically examine their assumptions and reach their own conclusions.

It helps turn debates into mutual explorations of biases and beliefs.

For example, with crypto, questions could be:

• Why do you think crypto exists, and what problem was it designed to solve?

• Why do you think so many developers and entrepreneurs are dedicating their lives to this technology?

• What historical examples or disruptive technologies were misunderstood or underestimated early on?

• Why do you believe governments or financial institutions might resist crypto adoption?

I find this process most fascinating in discussions with my dad; we share a very similar personality and set of values but have very different opinions on topics like AI regulation and DJT’s election. By asking questions rather than debating, we both gain a deeper understanding of why we think the way we do—and sometimes, even shift our perspectives.

I am currently rewatching old seasons of Mad Men, a TV series based in the 1960s that follows a prestigious ad agency in New York and the “ad men” of Madison Avenue.

These ad agencies captured the attention of broad swathes of the US population and converted them into religiously paying customers for things like cars, cigarettes, beauty products, electronics, airline tickets, alcohol, and frozen foods. 

The ad men did this by crafting compelling brand narratives and establishing unique selling propositions via TV commercials, print ads, and radio spots, creating impactful ads that often become part of popular culture:

• De Beers – “A Diamond is Forever” (1947)

• Marlboro – “Marlboro Man” (1954)

• Coca-Cola – “It’s the Real Thing” (1969)

• McDonald’s – "You Deserve a Break Today" (1971)

• Nike – "Just Do It" (1988)

The 1960s marked an inflection point in history. The adoption of TV and radio enabled a relatively small group of people to quickly and easily capture the attention of millions in the pursuit of profit. In the process, they were able to disseminate ideas (i.e. memes) and create culture throughout the entire population and nation.

This was the early attention economy, and since the 1960s, it has only strengthened and become more global in nature; after the TV came personal computers and cameras (80s), the internet and web browsers (90s), social media platforms like Facebook and Twitter (2000s), smartphones and app stores (late 2000s / early 2010s). 

I believe we are currently at another inflection point in the attention economy because of crypto and AI. There is no coincidence that the research paper by Google in 2017, which helped birth LLMs and multimodal generative AI, was titled “Attention Is All You Need” 😉

I am currently thinking through what this inflection is and, in this blog post, unpack raw thoughts.

1. In the 1960s, capturing and retaining attention through advertising helped large companies sell more goods and services, often at higher margins. This resulted in higher stock prices due to traditional valuation methods like a DCF.

2. In the 2010s, capturing and retaining attention helped content creators and influencers build income streams through subscriptions, sponsorships, affiliate marketing, merchandising, licensing, and digital products.

Seeing how greater attention yields greater value in the examples above feels intuitive. 

Perhaps less intuitive is seeing how greater attention, when combined with social consensus, can result in esoteric goods like sneakers, sports collectibles, vintage furniture, and children’s trading cards being exchanged for hundreds, thousands, or even millions of dollars in some cases. You may have a tough time running a DCF on the PSA 10 first edition shadowless holographic Charizard Pokémon card to justify its $420,000 price tag! 

Some may recognize this as “brand” or “cultural capital.” For the purposes of this blog post, I will refer to it as the “attention premium.”

In recent years, and in large part due to the internet, this “attention premium” has bled over to traditional markets. Assets like Tesla, GameStop, and AMC, formerly valued via an analysis of their cashflows, began to derive a significant portion of their value from the attention they received. 

Crypto has taken this trend toward financializing attention to a new level. The role of crypto is simple: 1/ it enables anyone to quickly create new assets, 2/ it enables anyone to trade assets. This works because crypto is permissionless (anyone can issue assets of any kind) and composable (anyone can trade those assets on any venue). 

This has led to the creation of millions of new crypto assets over the last few years. Many of these assets represent an idea (i.e. a meme) that others can invest in, and benefit from a financial system with billions of dollars of daily liquidity that anyone can access. 

These assets—memecoins, NFTs, and tokens—can be considered “attention assets,” which measure and capture the amount and value of attention in real-time. The scale of this asset class should not be dismissed; the total memecoin market cap is $50B, Dogecoin hit a market cap of $85B, CryptoPunk #5822 sold for $24M, and our team at Seed Club launched a memecoin called $enjoy, which hit an $80M market cap after 30 days.

Money is a language; when people buy these assets, they communicate their belief in a particular idea. In this world, attention is the primary driver of value, and crypto enables individuals to profit based on their ability to correctly identify which assets, creators, or content will attract more attention.

Meanwhile, AI’s role in generating and curating content personalized to unimaginable precision means that capturing and monetizing attention will become an increasingly automated and sophisticated practice. Nevertheless, generative AI will democratize content creation, making it easier for anyone to join the arena and try to monetize their ideas.

While the role of AI agents in this is still unclear to me, I sense that the symbiosis of crypto and AI will lead to a new kind of market, one that is permissionless, decentralized, and driven by data and attention in equal measure. Gone are the days when Mad Men’s Don Draper and his team of copywriters and art directors were the sole arbiters of financialized memes, taking months to produce and roll out an ad to niche groups of US consumers.

I am still forming my thoughts on this. I’d love to hear any of your ideas about the future of the attention economy and how you see AI and crypto shaping cultural and economic interactions – please reach me on Twitter or Warpcast.

He attributes this to 7 “tactical issues”: 

1. The market size is too small

2. The userbase is horribly mercenary

3. The signals and metrics are wrong

4. PMF is near impossible to find

5. There is nothing to iterate on

6. It’s a hard environment to design & build on

7. Hyper-financialization is bad for building

In this blog post I will analyze each of Chris’ issues and conclude with why I believe today is the best time to build in consumer crypto. 

1/ Market Size. Yes, the total active daily wallets on BTC (1M) and ETH (500K) is a small pool relative to the opportunity available to web2 founders. However, crypto users are younger, more internet native, and wealthier than the average web2 user. They tend to transact more regularly and in larger volumes, leading to a far higher potential ARPU and the ability for small startups to negotiate partnerships with major companies and brands. Moreover, crypto users are extremely engaged and quick to offer feedback, enabling founders to ship with high velocity, get quantitative signals and qualitative feedback, and build a better product. These are tailwinds for founders to build generational businesses.

2/ Mercenary Userbase. Crypto users are very internet-native, open to new ideas, and not afraid to pivot to a different product when something better comes around. In other words, crypto is incredibly ripe for innovation and rapid iteration. Consider the speed with which Blur built and delivered an excellent UX to a neglected userbase (NFT traders) and took market share from OpenSea – the fact that users are “mercenary” is more of an opportunity/positive than a challenge/negative for founders who want to operate in a blue, meritocratic ocean that isn’t saturated by entrenched, monolithic web2 behemoths operating stale walled garden-style business models.

3/ Wrong Signals. I generally agree here. Crypto’s obsession with TVL, token prices, and market cap are distractions from what is most important for founders to focus on – the product. The industry is young and, in its current state, extremely reflexive and driven by narrative rather than fundamentals. Users must have more patience (and not give founders anxiety and drive them to Ketamine), and we’ve got to get much more nuanced with how we measure the value of crypto projects. While there’s no right answer, and it’ll likely vary for different projects, I quite like David Phelps' idea to think of crypto projects as nation-states with their own currencies, taxes, governance, and GDP (rather than as companies, valued based on cash flows and via a DCF). 

4. PMF. I think Chris is generalizing and overlooking large parts of crypto here. For instance, the NFT sector, with a market cap exceeding $7B, demonstrates PMF through various successful startups, including marketplaces (OpenSea, Art Blocks, SuperRare, Blur), financialization platforms (NFTfi, Blend, MetaStreet), and discovering and tracking apps (Interface, Floor, Gallery). Moreover, new standards have unlocked new use cases and made NFTs more accessible – NFT compression enabled DRiP to distribute millions of NFTs to >800K wallets. While finding PMF in crypto can indeed be challenging, it shouldn’t detract from founders. Imo, early adopters are the ones who reap the biggest rewards, and, as a founder, betting on crypto now is akin to betting on the early internet or mobile. 

5. Iteration. I agree that consumer crypto currently requires the invention of new categories because the benefits of moving existing things onto crypto rails are limited. However, I challenge the notion that this should deter founders from building in crypto. Consider the early stages of the internet: social media, streaming, and online shopping all emerged because pioneers didn’t just iteratively improve on existing ideas but created entirely new products. This pattern is typical in any revolutionary technology and is fantastic news for founders! Crypto is ripe for innovation and there’s a huge opportunity to imagine what could exist!

6. Hard environment. I’m not technical, but my understanding is that, unlike fragmented and siloed web2 data, composability is the counterfactual. DRiP on Solana or Interface on Ethereum didn’t have to build standards themselves – they created highly scalable products enabled by cheaper infrastructure. Moreover, the ever-growing social graph being built in real-time on protocols like Farcaster will only accelerate (literally every crypto transaction is technically UGC) and make it easier for consumer founders to build new and interesting products. As Amanda outlines in her recent post on Consumer Crypto’s Breakout Cycle, an incredible amount of infrastructure has been built in recent years, including low-fee blockchains, embedded wallets, on-ramps/off-ramps/bridging, identity protocols, and distribution channels. If you subscribe to Union Square Ventures’ The Myth of the Infrastructure Phase thesis, now is the best time to build in consumer crypto!

7. Hyper-financialization. Speculation is a feature, not a bug. This applies not only to crypto but to almost any area of innovation. Indeed, in his book How Innovation Works, Matt Ridley argues that “innovation happens when people are free to think, experiment, and speculate”. I acknowledge that shilling can become an inauthentic ponzu, but generally when a user is excited and bullish on something, it's great that they’re sharing it, creating buzz, and driving engagement back to the project (and aligned to do so). In some cases, excessive shilling can even lead to a huge wave of new users that test the limits of your tech and your business model and help make things more anti-fragile and robust (e.g. after Berachain launched on testnet last week, they paused the faucet due to a huge inbound of users. this gives them an opportunity to work with technical partners to upgrade the network and improve their tech for mainnet). For founders, regardless of whether they build in web2 or web3, gamification is behind a lot of the core loops of popular apps and, in web3, the opportunity is in finding a balance between financial upside and gamification.

Thus far, I’ve mostly disagreed with or tried to add nuance to the challenges Chris presents as reasons why founders should not build in consumer crypto. However, what I disagree most with Chris about is not the existence of these challenges but rather that these challenges make it a bad time to build in consumer crypto. The opposite is true – these challenges are precisely what makes it an incredible time to build in consumer crypto.

As Josh argues, we’re all here because we believe a fundamental platform shift is happening where all digital value, objects, and applications will move into blockchain rails. But “we’re still early” – the space is nascent, and there isn’t competition and product saturation like there is in web2. This presents an incredible opportunity for founders across every category – to pioneer entirely new products, use cases, and business models.

As a founder, would you rather wait, see the industry mature, and miss the opportunity, or face these challenges and lead a new platform shift? If the early days of mobile or the internet are anything to go by, the greatest rewards will go to those who stick around the longest and are bold enough to continue to build and experiment through both the good and bad times. 

As a VC in consumer crypto, I find this an incredibly exciting time to invest, especially considering the recent advancements in scalability, security, and user experience that have laid a strong foundation for innovation. For those already in or considering entering the consumer crypto space, I hope this perspective offers some insight. If any of the above resonates, I’d love to chat! Please feel free to DM :)

The Generative Art Movement

The computer has been gaining societal importance since the 1950s. Coinciding with this, artists have been experimenting with generative art – the form of art characterized by the use of algorithms and computer programs to create artworks. Early pioneers like Vera Molnár, Ben Laposky, and George Nees merged mathematical precision with artistic expression, showcasing how algorithms could generate complex and visually intriguing patterns and shapes. 

Initially met with skepticism, generative art challenged traditional notions of artistic skill and creativity. The art community questioned whether it could truly be considered “art,” while elite critics outright dismissed generative art, claiming “the computer has no role in the art world.”

As digital technology transformed our lives, generative art gained acceptance and recognition. In 2021, generative art found “product-market-fit” in the context of being minted as Non-Fungible Tokens (NFTs) on the blockchain. Beyond financial speculation, the drivers behind this were blockchain-based generative art enabling people to: 1/ prove ownership of unique digital pieces and, thus, create scarcity and demand, 2/ showcase and transfer unique digital pieces to a global audience. Today, generative art is firmly established in the mainstream art world via its embrace from major institutions like LACMA, MoMA, Centre Pompidou, Uffizi Gallery, Pace Gallery, Christie’s, Sotheby’s + the success of artists like Tyler Hobbs, Dmitri Cherniak, Refik Anadol.

It’s increasingly clear to me that generative art is the art form of this decade. Art is ultimately a reflection of our environment and society – a society in which computers are a dominant force. Yet, no globally recognized artistic movement has been technology-centric. Generative art fills this gap. Indeed, the history of art and culture is full of examples of subcultures that became the culture decades later (in this case, the time between the 1950s and 2020s!).

Parallels between Generative Art and the Renaissance

The Renaissance was a pivotal period in European history (14th - 17th century) that marked a profound cultural and intellectual rebirth. Several key events catalyzed the Renaissance: the invention of the printing press in 1440, the Fall of Constantinople in 1453, and the Black Death through the 1400s. Downstream of this was a renewed interest in science, mathematics, and the natural world, the most visible impact of which was in art. In particular, the development of linear perspective, a technique for creating an illusion of depth on a flat surface, was a mathematical achievement as much as an artistic one. 

Artists like Filippo Brunelleschi and Leon Battista Alberti applied geometric principles to create more realistic and proportionally accurate depictions, while Michelangelo and Leonardo da Vinci broke ground with the “Golden Ratio,” applying complex math to architecture and sculptures.

The Renaissance is possibly the most fitting historical parallel to the generative art movement for several reasons:

1. Innovation: The Renaissance marked a profound shift with the introduction of new technologies like the printing press and advancements in materials and methods. Similarly, generative art embraces the internet, blockchain/NFTs, and computational methods.

2. Fusion of Art and Science: The Renaissance saw art and science as interconnected disciplines, as exemplified by Leonardo da Vinci. Similarly, generative art is an amalgamation of coding, algorithmic complexity, and aesthetic expression.

3. Patronage: The Renaissance saw patrons like the Medici family play a crucial role in supporting artists. Similarly, communities like Proof, DAOs like SquiggleDAO, and individuals like Punk 6529, Cozomo de’ Medici, and TokenAngels have emerged as modern patrons, shaping the trajectory of generative art through their collections, social media presence, and community engagement.

4. Democratization of Art: The Renaissance saw the democratization of knowledge through the printing press. Similarly, generative art is part of a broader trend of democratizing art creation and ownership.

5. Rebirth: The Renaissance was a period of intellectual and artistic exploration and expansion. Similarly, generative art pushes the boundaries of artistic possibility, explores new forms of expression, and challenges our understanding of creativity, authorship/attribution, and ownership.

The extension of generative technology beyond art

The historical parallels between generative art and the Renaissance provide a framework for understanding the potential impact generative technology will have. Just as the Renaissance profoundly influenced art, architecture, science, and culture, I envisage generative technology will also extend far beyond just art (it already is!).

Fashion: Platforms like Refabric allow designers to input keywords, fabrics, or even existing garments and generate new patterns with unique prints, textures, and silhouettes. Tools like CLO and Marvelous Designer simulate fabric draping and garment behavior on virtual models, allowing designers to iterate quickly, visualize designs in 3D, and optimize fit before physical production, saving time and resources.

Music: Tools like Stable Audio can create custom-length music and new sounds and textures by describing them, providing musicians and producers with new inspiration and creative pathways. Music can be tailored to individual listeners’ preferences, moods, or activities and transform live performances with real-time music generation and manipulation. 

Architecture: Tools like Karamba3D enable architects to explore and iterate on complex shapes and structures and design buildings that are aesthetic, structurally optimized, and environmentally responsive. Advancements in 3D printing, coupled with generative tech, enable the creation of complex, customized architectural elements and even entire buildings, opening the door for personalized design and on-site construction.

Product Design: Using algorithms to explore various design possibilities, designers can achieve optimal solutions that balance form, function, and sustainability. This is particularly valuable in industries where performance and material efficiency are crucial, like automotive and aerospace.

Virtual Worlds and Gaming: Games like Minecraft use algorithms to create complex, ever-changing environments and experiences, making each user's experience unique. Tools like Holodeck will help enhance visual and interactive aspects and create dynamic narratives and experiences that adapt to the user's actions.

Fin

The Renaissance left an indelible mark on history, and there’s no doubt in my mind that generative technology will match, if not surpass, its impact, shape the future of multiple industries, and change how we interact with the world around us. 

As a pre-seed investor, I am grateful to have backed companies accelerating this trend, including Stability AI (open-source foundational AI models), SquiggleDAO (NFT art DAO focused on Art Blocks’ Chromie Squiggle), StoryCo (DAO using character NFTs to create story-based IP), Spawning (tools for artists to own and manage their AI training data), and Trame Paris (luxury generative manufacturing platform).

I am excited to continue supporting early-stage, passionate teams experimenting on the edges, particularly those focused on open-source AI, data sovereignty, community as a moat, and uniquely crypto-enabled luxury and consumer experiences. Please don’t hesitate to DM on X if any of the above resonates.

Boils down to 2 main reasons:

1. Founders mistakenly thinking nfx is their moat or edge

2. Founders mistakenly thinking network value has a linear relationship with user growth

Founders mistakenly thinking nfx is their moat or edge

If you’re building a networked product, by definition, all your competitors also benefit from nfx! In light of this, focus on how you will uniquely wield nfx relative to your competitors.

Consider, for example, data network effects. Yeah cool, your product inherently means new users generate new & valuable data, but that’s also the case for all your current and/or future competitors!

What’s important here is not that your networked product is generating new data for every new user / user interaction, but how you will translate this data into a better product and/or biz model. Then, when raising, elucidating why you / your company is able to execute on that product <> nfx strategy relentlessly, creating a defensible flywheel [better than anyone else in the world]!

Imo, the best and most defensible moats are built around a potent mix of elite-level: 1/ proprietary tech, 2/ economies of scale in biz model, and 3/ brand. Nfx can accentuate those strengths and are necessary to build an enduring business, but not a silver bullet. 

Founders mistakenly thinking network value has a linear relationship with user growth

There’s this blind intellectual inheritance stemming from the origins of nfx… wake up! We no longer live in the world of the OG telephone network and Metcalfe’s Law, where every new user uniformly adds network value, starting with users 1, 2, 3, etc. No. That was hugely proven dumb AF during the dot-com era where operators, speculators, and shitcoiners got rekt. 

We use a new mental model now, which is ofc also flawed, but at least a bit better: value growth in a network looks like an S-curve, where your network has little-to-no value until some amount of X thing [users, apps, data, content, etc] is hit. This is the network’s ‘initial phase’ or ‘tipping point’ and getting past this is ‘solving the Cold Start Problem’. After this point, nfx kick in and value grows exponentially. 

What’s important for pre-seed/seed founders (often with a super early product & no nfx!) is 110% focusing on solving Cold Start (and not what your product/company will look like on the other side of the tipping point). 

This is generally best done by building a small, sticky, atomic network of initial users around a core niche (Harvard students for Facebook, wealthy SF techies for Uber, boujee LA kids for Tinder) and remaining laser-focused on: 1/ your core value prop, 2/ keeping super close to your earliest users & rapid iterative feedback loops, and 3/ community building.

If this resonates & you’re building a networked product, pls feel free to drop me a line / follow info on my website.

Fwiw, a great resource to learn more about nfx is Andrew Chen’s Cold Start Problem which I read a few months ago (see Goodreads link + online summary).

Thanks for reading! Subscribe 👇

Sorrento

Stunning scenery and relaxed atmosphere across entire Gulf of Salerno. Sorrento center was lively, but staying outside the town offered a serene balance. 

Despite visiting mid-Sep, main towns all very touristic. Sorrento’s tourist demographic leans towards less cultured/middle-class, dominated by Americans and Brits. Annoying at times. Capri: charming town, breathtaking views, delightful sea. Ravello: upper-class town, absolute gem, favourite spot.

Lots of transport options between towns, but it’s all chaotic and unorganized. Turkey/Istanbul/Bodrum does this much better. Winding roads amidst the hills means getting anywhere takes a couple hours, returning home late is not feasible. Car hire is inconvenient due to parking.

Took train to spend a day in Naples. 35 min radius from station was a mind-numbing shithole. Incredibly dirty and poor. Probably the most neglected place I’ve been since Cambodia in 2017… areas of Armenia were poor but incomparable to how dirty Naples was. Neapolitan pizza, best I’ve ever had.  

Pretty crazy how a history of unstable foreign rule coupled with organized crime and political corruption/extractive institutions can leave Italy’s south relatively much poorer than the north. Still has nothing on the North/South Korea divide though!

Rome

Wow. This place made me feel like a character in ancient Rome. There’s a breathtaking monument everywhere you turn! 

Streets very noisy, busy, dirty, and full of traffic. Far too many tourists. Quite hilly.

Larger city than I expected, with a lot of commercial activity. Not my style. 

Florence

Homebase for Tuscany. Infinite steak, incredible wine. Warning: meat overdose is possible here. Cantucci e vin santo, best dessert ever, had with every meal. Did not stop us from also having gelato every day. Outrageous gluttony, totally worth it.

Visited Siena, Montepulciano, Pienza, Montalcino. Enchanting, historic towns with medieval/gothic vibes and beautiful, picturesque landscapes. Rolling hills, cypress trees, vineyards. 

Visited several wineries. Very interesting histories. Helped me understand wine better, feel like there should be a guidebook in every restaurant menu! Antinori Chianti, super impressive 5* luxury vineyard (I feel most people would trade places with the grapes!), 10th oldest family business in the world!

Florence less touristic than Rome/Amalfi. Very walkable. Cute European town. Cool how it truly feels like the Tuscan capital and “Cradle of the Renaissance.” Uffizi and the Cathedral are insane, reminds me of Assassin's Creed! Boboli Gardens are beautiful.

Milan

Had a quick day stop. Didn’t need more. Felt similar to most major European cities. International, commercial, busy. 

Duomo was crazy, also some cool old Renaissance churches dotted around the central city. Galleria was beautiful, first time we’d seen the 19th-century iron-and-glass architecture characteristic

Cool to see the historic features of Milan Uni with Leo. Interesting influence and impact from WW2 and communism. Campus buildings themselves were quite mesmerizing. Can imagine inspired study sessions. 

Verona

Can see why it’s known as the city of love, associated with Shakespeare’s Romeo & Juliet. Most liveable place we visited. Not too big or small. Nostalgic European village vibes. Very cozy, walkable, relatively more modern. Less tourism. Noticed more harsh accents. Kind people, but less openly friendly/more cold/more Germanic. 

Stayed in two locations, the second was a beautiful Airbnb located near the river just outside the main central city. Excellent location, older/more traditional, more upmarket. 

Tried horse meat. Not worth the hype, similar to lamb. Had more seafood/fish during our stay, had missed it! 

Lake Garda was beautiful, but huge. Half a day gave us only a tiny glimpse. Getting there by bus was a shit show, as was the way back to Milan for our final departure. 

Final

Overall, very good trip. Naturally, had ups and downs. Lots of memories and photos, will cherish long term. After ~3 weeks, and having visited Venice twice last year, I’m only left curious to explore Sicily.

Southern Spain, due to expat community, stage of economic and infrastructure progress, ubiquity of language, and relative pricing/affordability would be my preferred choice to live at both this stage of my life & also over the next 10-15 years/raising young family. Italy more opulent, refined, high culture. Better for short breaks, lavish experiences.

I was recently a speaker at Miami Web3 Summit and assigned the topic “DAOs: An Institutional View.” The session’s description was:

“Decentralized Autonomous Organizations (DAOs) are a new form of structuring social groups, bringing people together for a shared cause while rewarding those who contribute. Institutions have started to recognize their value, with venture capitalists investing heavily. Join us to find out what the future holds for DAOs and what role institutional investors play. Panelists will discuss the current state of DAOs and how decentralized decision-making impacts projects, companies & investors.”

Generally uncomfortable with how broad the description was, I prepared some notes in order to streamline my thoughts. While they’re unstructured and unpolished, I wanted to make this public as a marker of my Q4 2022 thinking 👇

Intro/DAO Landscape

What traction and impact have DAOs had?

• As of mid-2022, there are 4,000+ active DAOs with a $20B+ market capitalization. 40% of these are DeFi-focused, but DAOs also exist across art & culture (6%), gaming (5%), politics (4%), physical assets (4%), and media (2%)

• Due to the growth in the space and the malleable way in which DAOs can form and focus on different goals, institutional investors have become increasingly interested in the DAO opportunity – with VC investment into the DAO space reaching a $160M local peak in Q1 2022

What do we mean by DAO?

• The term "DAO" is pretty useless. Depending on who you talk to, it’s either an overly broad term (like “company,” “business,” or “network state”) or any one of a number of very specific definitions

• IMO, DAOs can be thought of as “internet-native” or “web3-native” organizations, where ownership, governance, and revenues are all on-chain

What are the different types of DAOs?

• Protocol: Maker, Compound, Uniswap, Aave, Yearn, Sushi, Curve

• Investment: Seed Club Ventures, TheLAO, Global Coin Research

• Service: Raid Guild, Lex DAO, Developer DAO, Vector DAO

• Grants: MolochDAO, Audius Grants, Uniswap Grants

• NFT Collector: SquiggleDAO, FingerprintsDAO, PleasrDAO

• Media: Forefront, Water & Music,  BanklessDAO, DarkstarDAO

• Social: Seed Club, Cabin, Krause House, Friends With Benefits

Why are people so excited about DAOs?

Decentralized Ownership and Governance

• Incentive alignment, equitable distribution of value, equality of opportunity

• Malleability with which DAOs can form and focus on different causes

Future of Work

• Global, Remote, Async, Freelance 2.0, Ownership Economy, Pseudo/Anon

• Transparent + Non-Hierarchical Organizations, Seamless On/Off-boarding

Scalable Human Coordination

• Effective coordination is represented by: a) the effective delegation of authority (empowering specific individuals with their roles and power), b) values alignment (recruiting the right people in terms of skills), and c) meme alignment (where people are most passionate)

• With this in mind, everything from universities and charities to hedge funds and start-ups could become DAOs; distributed resources and collective intelligence with the context that authority is correctly and effectively delegated and values are aligned is incredibly powerful!

What challenges are DAOs facing?

Tragedy of the Commons

• DAO leaders either: i) don’t share context for DAO members and contributors and/or for those that come after them, or ii) they do, and their time is eaten up in doing so and getting involved with more trivial rather than more important matters

• Do all DAO members and contributors really want to govern/have a say in governance?

Free-rider Problem

• Often problematic in Investment DAOs – DAO participants receive pro-rata financial returns based on their capital contributions, thereby disincentivizing and de-motivating active participation in the DAO (which for investment DAOs is deal sourcing, due diligencing, portfolio support, etc.)

Coordination without Structure

• A DAO’s dynamism, transparency, and lack of hierarchy are also its pitfalls. Contributors find DAOs “messy”, encounter information overload, and lack clear leadership

Consensus-driven Product Design Results In Worse Outcomes

• Steve Jobs – “people don’t know what they want until you show it to them” – contrarian product building can be more effective

• Vitalik Buterin – Some decisions should not involve compromise and are best made when centralized

Legal/Regulatory Uncertainty

• Investment instruments are in the early stages of being figured out, especially to protect investors and appropriately dilute them at the point of equity -> token conversion

• Options include SAFE with pro-rata token conversion, SAFT, SAFE + Token Warrant, SAFE + Token Side Letter

• Legal frameworks are in the early stages of being figured out, especially to protect contributors from unlimited liability. KaliDAO is working to solve this

Revenue/Value Accrual

• Why and how value accrues to a DAO’s tokens and/or NFTs (“tokenomics”) is still in the early stages of being figured out

• Token value accrual and its relationship to building sustainable on-chain revenues and a business model is still up in the air

• Specifically for protocols and products, there uncertainty around whether to charge fees (which in an open-source environment creates vulnerabilities to be forked) or follow a more experimental hyperstructure model

What are DAOs getting right?

Capital Formation

• Everything from one-off/ephemeral use-cases like Constitution DAO (raised $47M to bid on the US Constitution), activist use-cases like Ukraine DAO (raised $10M for the war in Ukraine) or more ongoing ecosystem development and investments like BitDAO ($1.3B treasury)

Revenue Generation

• ENS and NounsDAO have demonstrated novel models to consistently generate revenue and sustainably grow their treasuries

Financialized Meme

• Bitcoin (digital gold), Ethereum (digital oil), FWB (culture-makers), and Seed Club (fuck YC) successfully accrued billions in market cap by successfully proliferating their respective memes as efficiently and effectively as possible and creating FOMO

Technology Upside

• DAOs with products and/or communities (revenue generative or not) have successfully rewarded early participants in their networks. I’ve received $100K+ in airdrops from projects including ENS, Uniswap, BAYC, Seed Club, LooksRare, and Poolsuite

Community Building

• Krause Hause, PROOF Collective, FWB have leveraged on- and off-line methods (content, media, events, conferences, education, resources) to build hyper-engaged communities

How might you approach investing in this space?

• NFTs as early-stage DAO tokens

• Private, early-stage investments: a) into corporations with a mandate to launch a token /  “exit to community”, or b) directly into illiquid DAO tokens

• Liquid tokens (capitalize on mispricing due to dynamics of early liquidity)

How are VCs investing in the DAO space? Investment approach: picks & shovels vs DAO direct?

• Two ways VCs are getting involved: 1) Investing in DAOs/DAO infrastructure, 2) DAOs as the VCs themselves (Venture DAOs / Investment DAOs)

• For the former (investing in DAOs and DAO infra), VCs are often approaching the space as they would enterprise SaaS (backing products, not DAOs). This is cool – it’s what we do at Seed Club Ventures – but it’s also important to recognize that: a) in web3, since so much of the tech stack is open-source, the network effect is often the community (not the platform), as evidenced by OpenSea getting attacked by LooksRare (& 20+ others), and b) some of the most interesting and valuable products have been built by and spun out of DAOs solving their own problems, e.g., Coordinape (which came out of Yearn) and Gnosis Safe (which came out of Gnosis DAO)

Can DAOs compete with the largest and most impactful VCs and investors?

• Investment DAOs started with “The DAO” and gained steam with friends collecting NFTs together, then participating in on-chain investments together, and then gradually forming InvestmentDAOs and VentureDAOs together, participating in Pre-Seed rounds, Seed Rounds, and now Series A rounds. It’s all an experiment!

• Speaking for Seed Club Ventures, for start-ups raising and that we’re investing in, DAOs represent a non-traditional source of capital that have less strings attached than VC backing, but don’t sacrifice the network of companies and individuals that come with a VC. We also bring our community and deep expertise

• I think the trend here is community investing – DAOs can gather information from different parts of the world perhaps better and faster than traditional systems, which may learn of trends or changes in a given industry later than the members of a DAO. This can give DAOs an advantage. However, a potential downside is the proposal and allocation of funds process in order to capitalize on that information

Which institutions will have the biggest impact on the DAO space?

• The “institutions are coming” meme has expired – institutional investors are already here – CFA’s 2022 investor survey revealed that 94% of state/government pension plan sponsors are currently investing in crypto

• On the flipside, in 2021 and 2022, we saw a surge in web3 activity from major brands across NFTs, Metaverse, Community, Memberships, Loyalty Schemes: Time released 3 NFTs of its red-bordered magazine covers, selling for $446,000. Tiffany unveiled 250 digital necklaces for 30 ETH each ($50,000 at the time) – available exclusively to CryptoPunk NFT holders. Budweiser released its first set of NFTs and partnered with 22 emerging artists for an 11,000 NFT collection. Nike acquired NFT sneaker studio RTFKT Studios, launched a metaverse sneaker line, and announced its SWOOSH platform for community members to create and trade digital collectibles

How do you define the success of a DAO?

• Web2 KPIs – organizational complexity, impact, and scale (Amazon, Starbucks, etc.)

• Web3 KPIs – active members, votes, proposals, treasury value, token holders, market cap, trading volume

Final thoughts

• If you made it this far through my ramblings, there's either clearly something wrong with you or, like me, you're quite obsessed with DAOs. In any case, please don't hesitate to hit me up on Twitter – I'd love to chat! 🙏

When forming investment theses and considering investments in early-stage start-ups, I rely on a well-structured and recognized process. The standard VC script – what's unique about the team? Does the product solve the problem? Are there network effects? How about compounding?

In light of FTX's implosion and a broader loss of faith in CeFi, however, our investment committee at Seed Club Ventures has begun to emphasize a project's ethics, principles, and the role it may play in shaping the future of crypto and web3, and therefore the world.

A decision-making framework I've previously neglected but found helpful in navigating these moral ambiguities is asking myself, "how will historians judge this decision?" and "how will historians describe our current era?" These questions are helpful in my personal life, especially when concerned with decisions that have moral and ethical implications.

What Is Our Current State?

We're living at an inflection point in human history, where two dominant and conflicting visions of how organizations (nations, governments, and corporations) should be operated are engaged in a war. Balaji's Network State and Dalio's Changing World Order explain this phenomenon by contrasting the Chinese and American regimes.

The decline of an empire (USA) and the rise of another (China) has never happened in our lifetime. China has risen to become a global superpower as strong, if not stronger, than the US. Notably, China has achieved this through a very different political order – one which embraces Confucianism (super hierarchical), Legalism (all things under one leader), and Taoism (opposites must balance). Concurrently, the US is struggling with military overextension, bad finances, inflation, decadent leadership, and internal conflicts. In particular, they face a disaffected populace incapable of engaging in constructive dialogue about their collective vision for the future.

Equally, globalization and the rise of the internet and the digital economy have prompted a move away from border-driven power struggles to economically-driven power struggles. The ongoing Russo-Ukrainian war well evidences this, and the threat to Armenia's sovereignty has meant I've felt this personally. Consider Russia, Turkey, and Azerbaijan's difficulty in overcoming Ukraine and Armenia, two considerably smaller countries, through military might. Much of the impact of these wars on the international stage is discussed with respect to the local economic impact.

The Web3 Opportunity

The power of private corporations is not lost on the political elite, and nations already use companies to influence the social mindshare and political narrative. It's also no secret that the country which unlocks the next wave of innovation in energy and artificial intelligence will likely accrue unimaginable power.

Crypto and web3 offer an alternative vision of the future, in which currency and economic activity are governed by code.

At Seed Club Ventures, we extend these ideals to organizations themselves – we believe the most successful organizations of the coming century will be open and transparent. My comrade, NiMA, often espouses that what open source did to software is about to happen to the corporations that run our world. We're therefore focused on investing directly into organizations building in a more open, autonomous, and decentralized way (DAOs) or in projects building the infrastructure, guardrails, and tooling for such organizations (DAO infrastructure and tooling).

As investors focused on early-stage projects, our key emphasis is naturally on the team. Not only whether they have the necessary tenacity and knowledge to reach product-market fit and to build an enduring business, but whether we share values of open standards, open source, and open participation.

In this context, I firmly believe historians will judge us positively – we're doing our best to create a more inclusive and equitable world that will inevitably drive greater innovation.