State of the Proving Infrastructure Landscape - 2024Q2

Sep 06, 2024

As zero-knowledge is becoming mainstream, and more and more projects are building provable compute into their tech, the speed of development is changing gears. To address the increasing demand for ZKPs, several teams have announced building prover infrastructure and launching their proving solutions. While these projects can all be grouped around one core service—ZK-proving—, their scope, offering, and protocol design often differ significantly.

With the growing demand for ZKPs it’s time to take a broader look at the proving landscape, map the different proving services, and see how the overall ZK-proving infrastructure has evolved, and where it is headed.

Introduction

While researching the proving infrastructure landscape, we have identified around 20 projects offering or planning to offer ZK-proving in some way.

These projects are (in alphabetical order): Cysic, Fermah Layer, Gevulot, Ingonyama, Irreducible, Lagrange, Lita, Lumoz, Marlin, Nexus, =nil;, NovaNet, RiscZero, Sindri, Snarkify, Strobe Network, Succinct, Taralli Labs, Zero Computing and ZKPool.

Some of these are building prover networks, while others proof marketplaces, be it auction-based or order book-based. Certain teams are offering proving as a centralized provider, but there are also those focusing on building decentralized infrastructure. The devil is in the details, and one needs a deep dive into the architecture and code (often challenging due to the lack of publicly available resources), to better understand their unique positions, and how these solutions differ. But before we get to some of these details, let’s take a quick historical outlook.

General outlook

Last summer, a lot of discussion was around sequencer-prover separation, and whether L2s should build their in-protocol (enshrined) prover networks or outsource proving to third-party providers when aiming to decentralize their proof generation. Third-party proving saves protocols significant complexity, but depending on the execution guarantees offered, it may introduce some external dependency, increasing risks related to liveness or censorship. On the other hand, decentralized in-protocol prover networks come with less efficient hardware utilization due to the idle time between workload allocations, create fragmentation in proving compute resources, and add complexity to running prover nodes if operators need to manage several nodes in distinct networks.

The vast majority of proof demand is currently coming from ZK- or validity L2s, coprocessors, zkVMs, and ZK-applications. For coprocessors, zkVMs, applications, and chains built on ZK-based rollup stacks it might be very natural to source ZKPs through third-party proving because it would allow going to the market faster, with lower costs and less complexity. But for large Ethereum L2s (such as Starknet, zkSync, Linea, or Scroll) that have been proving blocks internally in a centralized way, the question of decentralization may be more complex.

After a year, with the strengthening of the modular thesis and as more sophisticated proving infrastructure is developed, the industry is mostly leaning towards outsourcing the task of proof generation. At this point, however, a new type of fragmentation started to appear with the launch of several different proof markets and prover networks. They seemingly fall into the same bucket, but in reality, there are significant differences among them both in their scope and architecture, as well as their role in the proof supply chain.

Proof demand and supply

To understand the characteristics of the existing prover networks and proof markets, we need to consider what they are focusing on within the proof supply chain. We identified two groups of projects that are complementary to each other: demand-facing and supply-facing protocols.

Channels of proof demand

Looking at the current landscape, a large portion of the proving services were launched by teams that have been building ZK-related core products in-house, for instance, zkVMs or coprocessors, which create demand for proof generation. Their proof markets and prover networks are essentially complementary services to the in-house products. Below is a collection of these:

Ingonyama’s ZaKi is a proving service for users of the ICICLE library with accelerated hardware.
Irreducible’s zkProver initially supports proof generation for Binius.
Lagrange Network initially supports proofs for Lagrange’s ZK-coprocessor and state committees.
Lita’s prover service is generating proofs for the Valida zkVM.
Lumoz Network is doing proof generation for their zkRaas clients.
Marlin’s Kalypso is a proof marketplace integrated with Oyster.
Nexus Network is generating proofs for the Nexus zkVM.
=nil;’s Proof Market supports circuits generated using the zkLLVM compiler.
NovaNet initially supports proving for their WASM-based zkEngine.
RiscZero’s Bonsai supports the Risc0 zkVM to generate proofs for arbitrary computation.
Succinct Network is a prover network initially generating proofs for SP1 and Telepathy.
ZKPool is an exception as they have no in-house product similar to the above, however, the team’s initial focus has been on proving Taiko blocks.

The above-listed products are naturally creating proof demand through their users, therefore these projects are acting as channels of proof demand, whether it be through one or more supported prover programs or proof systems.

The same applies to projects such as Sindri or Snarkify which do not have an in-house zkVM or other products but provide proving services for certain domain-specific languages (DSLs) or proof systems they have integrated. In their case, the supported proof systems are the drivers of proof demand.

All these channels of demand have one thing in common though: they need to source compute capacity to be able to service their users.

Aggregators of proof demand

The channels of proof demand have multiple options to source compute for proof generation. While different alternatives exist, such as in-house hardware instances or rental in data centers, most of the industry currently relies on traditional, centralized cloud providers, like AWS and GCP. But aggregators of proof demand— supporting all proof systems across the board— also started to emerge.

Gevulot has a unique position in the market as the first supply-side aggregator: it is building a permissionless, decentralized ZkCloud that is also proof system-agnostic.

Any prover program can be self-deployed on Gevulot, making the platform capable of supplying proofs to any prover network and proof market. Because of the flexibility introduced by self-service prover deployment, the protocol can also serve any ZKP demand directly. Prover node operators will be able to join permissionlessly and generate proofs for any on-chain prover and verifier program deployed on the ZkCloud.

The aggregators of proof demand offer a legitimate alternative to centralized cloud compute platforms. Not only can they decrease the costs of proving significantly, but by being decentralized, they can also provide increased liveness guarantees and censorship resistance, without compromising on performance.

Viewed from two different angles—demand and supply side—, the proving infrastructure landscape could be summarized as follows:

Support and integration

Let’s take a look at the different design choices that proof providers have made.

Prover support

As discussed earlier, several proof markets and prover networks are supporting the proof system related to their in-house products, but some projects in the proving landscape provide support for multiple proof systems or plan to do so in the future.

Out of all the projects analyzed, we identified some protocols that can currently support multiple proof systems or prover programs: Gevulot, Lumoz Network, Sindri, and Zero Computing.

Lumoz Network can provide proof generation for different L2s within their RaaS offering. The chains supported by their prover network are mostly built on the Polygon zkEVM, with some exceptions, a testnet built on zkSync and another on Scroll’s zkEVM.

As a centralized entity, Sindri supports multiple domain-specific languages (DSLs) such as Noir, Halo2, or Gnark, as well as different proving schemes, including Groth16, and Plonk.

Zero Computing, a centralized proof provider with customized instances in data centers, can also generate proofs for multiple provers, including Scroll’s prover and RiscZero’s zkVM.

Gevulot offers proving for any proof system. The project has been building a permissionless, decentralized ZkCloud, that allows users to self-deploy any arbitrary proof system as an on-chain prover program. Programs can already be deployed and proofs generated and verified free of charge on Gevulot Devnet.

We have also found that a number of projects plan to support multiple proof systems, such as Cysic Network, Irreducible’s zkProver, Marlin’s Kalypso, NovaNet, Snarkify, and Succinct Network.

Integration method

Another key characteristic of proof markets and prover networks relates to the flexibility in integrating new provers and proof systems.

Only two protocols allow such self-service prover deployment already or plan to do so. Marlin Network is building to offer the users of Kalypso to self-deploy their provers by creating so-called “Markets”.

Gevulot has been offering self-deployment of any arbitrary prover program on its Devnet since March 2024. This approach allows for greater flexibility and enables user-side optimization.

All other protocols are supporting provers that have been integrated and deployed by their teams.

State of development

While several protocols are developing ZK-proving infrastructure, they are at different stages of their development.

Ready to roll

Based on the information available, the following projects are already generating ZK-proofs for their users: Gevulot, Lagrange Network, Lumoz Network, =nil; Proof Market, RiscZero’s Bonsai, Sindri, Succinct Network, and Zero Computing.

Who are these users? As mentioned earlier, Lagrange Network initially supports proof generation for their ZK-coprocessor and state committees, =nil; offers proving through their Proof Market for circuits generated using the zkLLVM compiler. RiscZero’s Bonsai offers proof generation for arbitrary computation with the Risc0 zkVM. Lumoz Network delivers ZKPs for the validity rollups launched through their zkRaaS platform. Succinct Network is a permissioned, private beta network that mostly generates proofs for SP1 and Telepathy. Sindri and Zero Computing offer proving services for the DSLs and proof systems they integrated already, and their clients are the users of those proving schemes.

Gevulot offers proof generation through its permissioned Devnet for any prover. It includes the entire proving pipeline: users can deploy any arbitrary proof system as a prover program, submit proof requests, receive the actual proofs, and have them verified, all free of charge. Its scalable, and production-ready prover network, Firestarter is launching in September.

Soon to launch

Several new proof providers are launching. Marlin has built the Kalypso SDK for users to interact with their proof marketplace, however, Marlin Hub shows Kalypso is still under construction. The Nexus Network is under active development, currently in a prototype stage as per their docs. NovaNet and Cysic Network recently launched their initial testnets. The waitlist is open to register for Ingonyama’s proving service, ZaKi, and Lita Foundation is also adding a prover service based on AWS Lambda for Valida zkVM programs.

Early stage projects

There are also some early-stage or stealth projects to take note of. Among these are Fermah Layer, Strobe Network, and Taralli Labs. All of them are building parts of the proof supply chain.

Future perspectives

Cooperation vs fragmentation

While competition among proof markets and prover networks can contribute to driving down the cost of proving for users, it also creates significant fragmentation. On the other hand, many of the protocols being developed offer complementary services because they focus on some aspects of the proof supply chain more than others. This allows for cooperation in various ways that are still to be mined. This is part of the reason why in our report we started with refining existing categories and differentiating between channels and aggregators of proof demand within the proving landscape. This approach immediately reveals new ways for cooperation among stakeholders of the proof supply chain, and in future editions of our report, we will also explore the ongoing cooperative efforts in the landscape.

Abstraction for outsourced proving

What is the user experience of L2s and ZK-applications when it comes to outsourcing proof generation? Until now little focus has been put on the UX related to sourcing proving compute, even though it is likely to become a key factor in the dynamics of prover decentralization. As the industry matures, and more and more proving services are launched, fragmentation could be the biggest bottleneck on the path to great user experience.

Gevulot recently announced establishing a generic abstraction layer for outsourced proving to which any L2 or ZK-application can easily connect. Prover networks and proof marketplaces with different architecture designs and job allocation mechanisms could all integrate with it, providing end users the option to select the optimal proving service as per their priorities. The goal is to develop the software as a neutral, open-source public good that can abstract away all complexity and allow very easy and simple procurement of proving services.

For further information on how stakeholders of the proof supply chain can participate in this collaborative effort and development, visit the article ZkBoost: Proof Supply Chain Abstraction.

Conclusion

Looking at the current state of the proving infrastructure landscape, it's clear that the industry is in a dynamic phase of growth and innovation.

While fragmentation presents challenges, it also drives competition and innovation. We believe that exploring ways for cooperation among different stakeholders and integrating complementary solutions is the path forward, where shared goals and collaborative efforts could significantly enhance the efficiency and accessibility of ZK-proving infrastructure.

Looking ahead, the focus may shift towards improving user experience and achieving greater abstraction in outsourced proving. Initiatives like Gevulot’s generic abstraction layer point towards a future where sourcing proving compute becomes seamless and user-friendly, promoting broader adoption and faster integration of decentralized ZK-proving.

Disclaimer:

The proving landscape is evolving day by day. If you feel we missed out on some important protocols or developments, or find any inaccuracies, please get in touch with Norbert from the Gevulot protocol research team, and we can make the necessary changes.

———

About Us:

Gevulot Blog

Discussion about this post