ZK technology is rapidly commercialising. Every 3 months, we see a new proposed zk proving scheme that improves on either proof generation times, proof sizes or memory usage. In the context of blockchain, zkApps require execution environments (zkVMs or zkEVMs) and settlements (DA, Ethereum, etc). zkVMs and zkEVMs like Starknet, ZkSync fight to provide the most performant execution environments. And settlement layers are becoming cheaper thanks to upgrades like Dencun, DA layers like Celestia & Eigen DA and coordination layers like Avail Nexus and Polygon AggLayer. 

We’re also seeing more zkApps developing in the market like zk coprocessors, zkml, zk oracles, zk bridges, zkdids, zkTLS etc. 

Decentralising proving is the next step 

One of the key pieces of tooling for zkApps that continues to be TBD is the prover market layer. This layer is akin to cloud computation in Web2 where zkApps send their workload and node operators aid in proof generation & verification. We’re optimistic that zkVM / zkEVMs will recruit services of “proof markets” to reduce their in-house costs and possibly use coordination layers to reduce settlement costs too. Before this happens, there is still much work to be done to optimise prover circuits. For example, running a ZkSync prover is not profitable for anyone except for MatterLabs and optimising it to run on consumer-grade devices is not a priority. Scroll & Taiko however have taken a different direction and decentralising their provers is top of the mind for these teams. 

Breaking down market participants & GTMs 

1. Hardware accelerators like Ingo, Cysic, Ulvetanna etc who own the GPU / FPGA / ASICs and will run provers for dApps themselves.  zkApps however will have to set up SLAs with the teams for their services.

2. Proof Markets are general-purpose  prover networks that run multiple types of proofs and accumulate “demand” from multiple applications. 

3. Proof Networks  are  node networks which run application specific proofs similar to an app-specific dePin network. This may also include proof aggregators like SHARP prover or Nebra however proof aggregation is possible on proof markets and on coordination layers. 

4. Coordination Layers are newer ideas and the best way to think of them is as zk-DA layers that help apps achieve soft fast finality. They will usually come with a bootstrapped proof market (ex: Avail Nexus) or will work with proof markets (ex: aligned x gevulot?). 

Value capture across participants 

Applications & zkEVM/Vms capture value through dApp interface fees and gas fees on transacted volume; but it’s not 100% clear which market participant captures the computational market for zk infra. 

Coordination Layers will compete with the next fast & cheap chain 

At first glance, coordination layers like Aligned Layer and Avail Nexus are the obvious choice. They’re akin to DA layers like Celestia that run Rust-based verifiers and enable significant cost reductions to zkApps allowing them to submit tx batches with higher frequency and ensuring soft fast finality. They will work with all provers and will have some level of “vendor lock-in” as they become the composable settlement layer. However, they are exposed to competition from DA layers like Celestia and faster alt L1s like Monad and are exposed to mercenary grant-hunter developers. This layer also don’t amortise the proving cost for any apps, only the settlement costs. 

Hardware accelerators have bigger bills to pay

The next one are the hardware accelerators that maintain their own hardware and develop optimisation algorithms. While they’re an interesting long term bet, they have high capex. To have any significant performance compared to peers; they will have to build their own FPGA / ASICs. This however is too early for the industry as ZkEVMs/VMs are constantly iterating on their provers. It also means that with each “upgrade” to the network; the hardware accelerators must reiterate on their chip design & manufacturing to compete on performance, which continues to be a high capex business model. Their hardware assets are also susceptible to high idle times and degradation as newer GPU models are released. It’s likely all the companies in this category will participate in general purpose proof markets to bootstrap demand for their GPUs.

Ulvetanna (now: Irreducible) is the most interesting one however - they have built their own proving scheme Binius which is accelerated & optimised for their hardware. They aim to build an ecosystem around their proving scheme and become the go-to prover for it. 

Proof networks attract mercenary compute 

L2s with steady proving requirements can easily bootstrap their own prover network and incentivise through their L2 token. But it also opens the space for design questions like should they prioritise cost vs speed of proof generation and how to do parallel proof generation to minimise idle time. Some projects like Aleo prioritise PoW-like fastest proof generation but the majority of the market will settle for cheaper proof generation. We see this time & again with projects like Pokt Network (dRPC), Helium and Render Network. 

Decentralized Proof Markets are basically DePin Networks 

We already have successful examples of decentralised networks getting mainstream adoption like Render Network being used in Apple’s latest iPads as a preloaded app (OctaneX). We can expect similar dynamics in the decentralised proof market space too. The next 6-12 months are the most crucial for these teams to acquire zkApps’ mindshare and bootstrap computational supply. They have virtually 0 hardware costs and will be the easiest provers for zk apps to integrate. This is already happening with Gevulot’s devnet where Taiko & Starknet stone proofs are being generated. The proof markets are permissionless and can also participate in the network & spend in-house R&D time on building FPGAs & ASICs in the future if that becomes a more profitable strategy without compromising on cash flow in the short term.  

ZK technology is rapidly commercialising. Every 3 months, we see a new proposed zk proving scheme that improves on either proof generation times, proof sizes or memory usage. In the context of blockchain, zkApps require execution environments (zkVMs or zkEVMs) and settlements (DA, Ethereum, etc). zkVMs and zkEVMs like Starknet, ZkSync fight to provide the most performant execution environments. And settlement layers are becoming cheaper thanks to upgrades like Dencun, DA layers like Celestia & Eigen DA and coordination layers like Avail Nexus and Polygon AggLayer. 

We’re also seeing more zkApps developing in the market like zk coprocessors, zkml, zk oracles, zk bridges, zkdids, zkTLS etc. 

Decentralising proving is the next step 

One of the key pieces of tooling for zkApps that continues to be TBD is the prover market layer. This layer is akin to cloud computation in Web2 where zkApps send their workload and node operators aid in proof generation & verification. We’re optimistic that zkVM / zkEVMs will recruit services of “proof markets” to reduce their in-house costs and possibly use coordination layers to reduce settlement costs too. Before this happens, there is still much work to be done to optimise prover circuits. For example, running a ZkSync prover is not profitable for anyone except for MatterLabs and optimising it to run on consumer-grade devices is not a priority. Scroll & Taiko however have taken a different direction and decentralising their provers is top of the mind for these teams. 

Breaking down market participants & GTMs 

1. Hardware accelerators like Ingo, Cysic, Ulvetanna etc who own the GPU / FPGA / ASICs and will run provers for dApps themselves.  zkApps however will have to set up SLAs with the teams for their services.

2. Proof Markets are general-purpose  prover networks that run multiple types of proofs and accumulate “demand” from multiple applications. 

3. Proof Networks  are  node networks which run application specific proofs similar to an app-specific dePin network. This may also include proof aggregators like SHARP prover or Nebra however proof aggregation is possible on proof markets and on coordination layers. 

4. Coordination Layers are newer ideas and the best way to think of them is as zk-DA layers that help apps achieve soft fast finality. They will usually come with a bootstrapped proof market (ex: Avail Nexus) or will work with proof markets (ex: aligned x gevulot?). 

Value capture across participants 

Applications & zkEVM/Vms capture value through dApp interface fees and gas fees on transacted volume; but it’s not 100% clear which market participant captures the computational market for zk infra. 

Coordination Layers will compete with the next fast & cheap chain 

At first glance, coordination layers like Aligned Layer and Avail Nexus are the obvious choice. They’re akin to DA layers like Celestia that run Rust-based verifiers and enable significant cost reductions to zkApps allowing them to submit tx batches with higher frequency and ensuring soft fast finality. They will work with all provers and will have some level of “vendor lock-in” as they become the composable settlement layer. However, they are exposed to competition from DA layers like Celestia and faster alt L1s like Monad and are exposed to mercenary grant-hunter developers. This layer also don’t amortise the proving cost for any apps, only the settlement costs. 

Hardware accelerators have bigger bills to pay

The next one are the hardware accelerators that maintain their own hardware and develop optimisation algorithms. While they’re an interesting long term bet, they have high capex. To have any significant performance compared to peers; they will have to build their own FPGA / ASICs. This however is too early for the industry as ZkEVMs/VMs are constantly iterating on their provers. It also means that with each “upgrade” to the network; the hardware accelerators must reiterate on their chip design & manufacturing to compete on performance, which continues to be a high capex business model. Their hardware assets are also susceptible to high idle times and degradation as newer GPU models are released. It’s likely all the companies in this category will participate in general purpose proof markets to bootstrap demand for their GPUs.

Ulvetanna (now: Irreducible) is the most interesting one however - they have built their own proving scheme Binius which is accelerated & optimised for their hardware. They aim to build an ecosystem around their proving scheme and become the go-to prover for it. 

Proof networks attract mercenary compute 

L2s with steady proving requirements can easily bootstrap their own prover network and incentivise through their L2 token. But it also opens the space for design questions like should they prioritise cost vs speed of proof generation and how to do parallel proof generation to minimise idle time. Some projects like Aleo prioritise PoW-like fastest proof generation but the majority of the market will settle for cheaper proof generation. We see this time & again with projects like Pokt Network (dRPC), Helium and Render Network. 

Decentralized Proof Markets are basically DePin Networks 

We already have successful examples of decentralised networks getting mainstream adoption like Render Network being used in Apple’s latest iPads as a preloaded app (OctaneX). We can expect similar dynamics in the decentralised proof market space too. The next 6-12 months are the most crucial for these teams to acquire zkApps’ mindshare and bootstrap computational supply. They have virtually 0 hardware costs and will be the easiest provers for zk apps to integrate. This is already happening with Gevulot’s devnet where Taiko & Starknet stone proofs are being generated. The proof markets are permissionless and can also participate in the network & spend in-house R&D time on building FPGAs & ASICs in the future if that becomes a more profitable strategy without compromising on cash flow in the short term.