Provers Networks from First Principle I Uma Roy at Pragma Denver 2025

[Music] and now I'm going to very shamelessly switch over to our next talk talking about Pro networks from scratch so please give a massive Round of Applause to Uma versus Sy Labs um hello uh I'm Uma the co-founder and CEO of succinct uh this talk is going to be a lot less controversial and uh back to some of the Wonders that we're building in this space so at cinct we're building the cinct pro Network and as background I wanted to start with kind of a history of other compute build outs that we've seen throughout various different use cases so so the first big compute buildout that we've seen is one of the biggest projects in crypto or the biggest project in crypto Bitcoin usually when you think about Bitcoin you think of it as decentralized money it has a cap Supply this haveing schedule it's a store of value but another way to view Bitcoin is actually as the world's biggest compute coordination mechanism the proof of work mechanism incentivizes a global scale infrastructure buildout of miners that are mining the network and increasing the hash rate and Bitcoin improves its properties when the hash rate is increased and it incentivizes this hash rate to come online and the compute makes the network more resilient and have better properties and the interesting property about this Global decentralized buildout is that bitcoin's hash rate exceeds that of any one company or even any one nation state and it's kind of just gone exponential over the years another example of very large scale compute buildout it's also centralized although it's not coordinated through a decentralized network is an AI so I'm sure we've all seen the headlines recently of things like the Stargate project which is a plan to invest $500 billion in AI infrastructure in building out data centers and Chip Fabs across the country and in general there's actually a lot of different companies who are working on this next generation of GPU data center like cor weave and Lambda that have tens of billions of dollars in market cap and lots of Revenue and making these GPU Data Center Specialized for AI workloads built out throughout the world okay so now with that background let's kind of go back to ZK which is another very computationally expensive workload that we think will also necessitate a similar sort of buildout uh so let's start with kind of what we're building at CN which is a ZK VM called sp1 that makes ZK really easy so the previous generation of ZK what I like to call kind of ZK 1.0 you had all these DK rollup teams who built their own proof systems uh their own rollup node and had to build their own stock vertically integrated to deliver on ZK end to end with succinct or ZK VM it makes ZK really easy so you can just write normal code you just write rust you stick it into sp1 and then anyone can now become a ZK rollup or for that matter prove any computation not just ZK EVMS because sp1 makes ZK general purpose and super easy it becomes as broad platform for generalizable ZK computation and so that's kind of why our company's motto is prove the world software because we've made it so easy to experiment with ZK that now we're going to see all these new apps including all types of rollups built on top of this new substrate and already we actually see a lot of adoption with our zkm uh mostly in rollups uh some of them are evm some of them are other types and we working with a lot of teams on kind of deploying ZK at scale and broadly um now that we have ZK VMS and this is kind of the clear next Paradigm of ZK what I like to call kind of ZK 2.0 where ZK is easy it's cheap it's fast and you can build and prove anything with it I think ZK will become really really critical to actually scaling blockchains and fundamentally like in blockchain architectures ZK makes a lot of sense because if you have a decentralized system uh historically the only way that you can could have decentralization but verifiability was that you'd have every node in the system reexecute all the transactions and so that's why blockchains are expensive and slow there's a lot of re redundant re-execution that imposes fundamental overhead but with ZK you can actually flip this on its head you have one person generated a proof that the computation is correct and then everyone in the world just verifies that proof very cheaply and doesn't have to redo it so this is why verifiable compute is going to be the fundamental building block behind scaling technology which is of blockchains which is something that ethereum knows very well and now other ecosystems are also latching on so for example like Celestia and Bitcoin l2s are becoming very popular and even chains like salana recently have started talking about this concept with their Network extensions as they kind of realize this Paradigm of scaling and kind of getting to global scale and a big question people often ask about ZK and I think they should is kind of like what is the market for ZK and so I think that's in any kind of like compute buildout it's really important to understand what the demand side will look like and so there's a few different ways of evaluating this and the first lens you can look at is just assuming that you know every rollup transaction get ZK proven so rough today roughly rollups are 200 TPS in the future if you have rollup scale ethereum da scales or maybe other ecosystem scales you have maybe 4,000 TPS to 4700 TPS and each transaction pays a tenth of a cent for proving cost and you multiply it all the numbers you get that in general there will be $150 million of demand uh for proving and this is just for the rollup use case and that's already like a pretty big number another lens to kind of evaluate the quote unquote Market size on the demand side for ZK is looking at what it replaces which is economic security so today if you're trying to build a blockchain and you're not doing a rollup uh you have to build an L1 and you have to get all these validators to reexecute all the transactions and come to consensus on the ordering of stuff on your chain and if you look at what chains pay today for Economic Security per year it's actually hundreds of millions of dollars each and in total it's probably a couple billion dollars in emissions of their token and that's actually a lot of money and in the future you can imagine that some of these chains it's just strictly a more economical decision to pay for ZK verifiability instead of this economic security verifiability and so one thing that's really interesting is that if ZK proving gets 10x cheaper than uh paying for this Economic Security well already that's a really big Market but actually this could lead to even more increased Demand right as costs come down we've always seen that demand for things often arises and this is a very popular gons Paradox that people have talked about in Ai and you can imagine a similar thing will hold true for ZK so hopefully all of this is kind of con convinced you that in ZK we're kind of at this turning point where ZK is finally easy thanks to ZK VMS and sp1 there will be a lot of demand and even already today with all the rollups there is a lot of demand and so now we can finally start talking about the supply side so given all this demand and given that ZK is a very computationally expensive workload it's similar to llms or it's you know similar to proof of work uh to generate a proof um we're going to need a lot more improving capacity to kind of serve this workload and so that is why we're building the cinc network uh the CNC network is uh the mission is to dramatically expand the world's proving capacity to help prove all these use cases of ZK including but not limited to chains and rollups so more concretely what is the cising network uh it's a decentralized protocol on ethereum that coordinates uh on one side the demand side that I just talked about out all the users of ZK proofs who are requesting proofs and provs who are on the supply side who are actually generating these ZK proofs uh for the end application and it uses cryptoeconomic incentives and an auction mechanism to help price the proofs and make sure that there's a really healthy competitive Marketplace uh for all this computation and there's actually several reasons why doing this as a decentralized Marketplace makes a lot of sense so so um there's this popular topic more broadly in crypto of deepin which is a decentralized physical infrastructure Network and the Prov Network falls into that category it's physical infrastructure it's a network and it's decentralized um there's a lot of reasons why ZK proving is uniquely suited to a decentralized network uh the primary one is ZK proofs are self-verifying so if you have a decentralized network where random people across the world are doing tasks and uh you know generating proof uh it's critical that their work can be verified for very cheap to ensure that they actually did the work so there's a bunch of companies who are trying to do decentralized AI inference and that stuff like that is very difficult because AI workloads are not inherently verifiable so it's hard to know if someone did the right thing or if they're just like giving you some random answer of like random English words and so there's a lot of protocols that are trying to come up with kind of clever mechanisms to ensure this verifiability but with ZK because the proofs are self verif in and that's definitionally true they can be checked by anyone for correctness so we actually don't care who's generating the ZK proof the proof is the proof If we verify it we know it's valid and we know we did the work so it's in some sense a more complex expressive version of bitcoin's proof of work another thing that's interesting about you know this decentralized physical infrastructure is that anyone in the world can actually contribute their own capacity and so there are GPU clusters all across the world today um that are not in data centers that can help be coordinated through the network to provide this proven capacity and make it more efficient and then over time we imagine that ZK proving workloads will actually uh go down the route of Bitcoin proving or sorry Bitcoin mining where you have more specialized Hardware like fpgas and as6 that extract more efficiency and especially once you have specialized Hardware doing it coordinating uh existing players like AWS and gcp they're probably not going to stock all these like special ZK proving as6 and so it makes even more sense to have this like globally incentivized decentralized buildout of this infrastructure similar to Bitcoin now one of the points I talked about previously is that a lot of the world's compute even gpus aren't necessarily in places like AWS and uh gcp and these you know tier one clouds you actually see that even in AI a lot of the GPU clusters are in like random places and actually coordinated through decentralized protocols as well so for example there's companies like vast AI or hyperbolic or a bunch of GPU marketplaces that have emerged in AI to coordinate you know this decentralized buildout of you know AI Pro AI clusters and I think you can imagine a really similar Trend happens in ZK proving as well because today it's literally using the similar sort of gpus and it's a very similar workload okay so now I've kind of talked about the motivation for why we're building this network kind of the actors we're coordinating why it makes sense to build it in a decentralized way and why it's actually fine that it's decentralized because verifiability means that anyone in the world can generate a proof and we're happy with that so in our Network um there's this kind of third actor that we introduce called the prover uh so the prover is basically the person who's actually generating the ZK proofs and uh that is a critical part of the network on the supply side so this is a great diagram to kind of illustrate the flow of things in the network on one side you have the demand which is things like rollups Bridges oracles but then even web two applications like privacy and identity and then they all get proven with our zkm sp1 by just writing normal code and then those sp1 programs uh you request proofs for sp1 programs on our network and on the supply side the network kind of coordinates people who are running their own GPU CL ERS in random places maybe people who are running stuff on AWS and have spare capacity and a bunch of other actors all across the world and our hope is that the network kickstarts this virtuous flywheel where basically applications request proofs because the network makes the demand very legible and transparent and pricing legible and transparent you have provs building out more capacity and then the provs compete on costs as the proving costs come down then more applications can be enabled by cheaper and faster approving and then that flywheel kicks and Spins to get more and more ZK adoption the particular mechanism in the network that we coordinate um is this proof contest auction for provs to compete on price so normally um in proof of work there's a lot of redundant work so if you just had a bunch of provs across the world racing to generate appr proof that's really wasteful because too many people are doing the work work and it's redundant and in ZK cost is super super important so that's kind of like a non-starter in proof of stake you have this allocation mechanism on like who the person is according to their stake who gets the next proof uh but that's also not a great fit for ZK because we want people to compete on the price they're able to provide uh for the proofs not just according to like how much Capital they have at stake with an auction model uh we can kind of get the best of all the properties we want or if you run an auction between all the provs you can have a single prover who wins according to the most competitive price and because of that you the ux of the network improves more and more over time and in particular our auction is special because it has this parameter where it's not always the case that the highest uh or the best prover wins uh so initially when in the bootstrapping phase you can imagine that you kind of uh give smaller provs a chance uh to win so it's not just super concentrated market and then over time as the market structure evolves and you have many parties competing uh then you can make your auction more uh concentrated so that you're providing the most aggressive pricing and more details about that can be found in our Network white paper that we wrote so recently we launched our test net over approver Network which allows uh anyone to kind of submit a request for appr proof and um you know generate appr proof through our initial pH of our Network and in this case we're actually generating all the proofs uh and there was a lot of proofs that were generated um a lot of people signed up all across the world they were taking actions like proving ethereum blocks doing other fun stuff like proving ZK video games um and yeah a lot of people kind of enjoyed playing around on our test net and seeing what's now possible with CK and so yeah we proved over I think at this point it's maybe over 200k proofs trillions of risk five Cycles or zkm um and yeah people are kind of being exposed to ZK and what it can do uh for the first time which is pretty cool and overall our goal is that with the network it's kind of the center of a bunch of ZK applications and because of what we're building at cync and the fact that it finally makes ZK easy as just writing normal code and normal software you can imagine that you have a bunch of applications that are uniquely enabled for the first time they can finally use ZK so our goal is the application layer will be really rich and expressive because ZK is easy the network will be very fast and provide proofs for fast and cheap and coordinate all these provs across the world who are going to compete and you know help with ZK adoption so the next phase is we kind of want to open up our network uh to the pr side so make it possible for anyone participating in in the proof contest and test out our mechanism make sure that it works end to end and make sure all the provs across the world can start participating and then finally we want to have a main net with kind of the battle tested mechanism and both the demand side and the supply side uh fully open and I think one thing that's also really exciting Beyond justar network is kind of zk's exponential progress throughout the years um and so this is just even looking at SP 1's performance um from when we started around a year ago to now and since then we've taken the performance and accelerated by you know almost 20 to 50x depending on how you count it and so we kind of imagin that ZK performance is going to continue on this exponential curve and can even be exponentially increased further with specialized hardware and specialized as6 which is hopefully something that our prover network will incentivize the buildout buildout of over time very similar to how Bitcoin mining started with you know people and hobbyists Mining at home on their computers moved to gpus and then progressively fpgs and as6 and we saw the hash rate go exponential over time we hope that this will also happen in ZK through our Network so to kind of summarize I think that uh we're obviously really big Believers in ZK we think that ZK is at this point where it's going to permet the entire blockchain stack you can imagine a future in which every single transaction action gets ZK proven like every single transaction I roll up ends up in some sort of ZK proof of a block or you know range of blocks and in this world Prov networks will be a critical part of the blockchain stack so every transaction today it pays a little bit for execution it pays a little bit for da and in the future it'll pay a little bit for proving and that is kind of like what we see as the biggest use case of the demand side of ZK over the next one to two years and we hope that you know with the what we're building these networks will make be resilient they'll be decentralized and most importantly they'll be competitive so that ZK pricing can continue to go down and enable a bunch of new use cases so yeah um that's kind of the entire summary of what we're building with the cinc network why we're so excited about how it'll accelerate ZK and yeah you can check out this QR code for our paper um and yeah come prove with us thank you [Music]