Boundless by RISC Zero I Jeremy Bruestle & Kartik Talwar at Pragma Denver 2025

[Music] this is going to be a fireside chat with Jeremy Jeremy is the CEO of RIS zero and we're going to talk about all the impressive things that they've been working on especially with the launch of boundless so without further Ado please welcome Jeremy all right hello everyone hi um I want to jump right in because uh there's so many questions that I have so I want to see if I can get through most of these but kind of the high Lev question we'll kind of make this uh for for the audience first is uh tell us about what boundless is and uh why did you decide to focus all of your efforts on on this update yeah so you know we were the creators of the original sort of general purpose zkm uh but the reality is is that if you want to do verifiable Computing if you want to do Computing off chain using ZK there's a lot more to it that's required than just being able to run a ZK VM because you have this question of how who's doing the proving how is it getting verified how is the payment processing working um and so basically boundless is you know what we're calling sort of a universal ZK protocol and the idea is to bring the power of verifiable computing to all of the existing chains right so we're not making a new L1 we're not making a new L2 we're actually making a new protocol that uses um you know smart contracts on existing chains and um these New cryptographic Primitives to basically make it possible for people on any chain say etherum say salana to request execution offchain get that execution done by a decentralized set of provs and get the results back on chain um and verified uh so the idea here really is that if you look at you know onchain execution there's only so much room in a given block and you end up in this world of scarcity where when you want to do something computationally intensive you have to pay massive gas costs and there's a fixed Supply right there's only so much in a block so as the demand for computing goes up the supply is fixed the prices Spike uh whereas ZK fundamentally changes that um so you in the sense that as the demand goes up more provs can come online the cost should never be more than the physical cost of doing the computation which is actually now cheaper than the cost of verifiable comput through consensus so the goal of boundless this has become true fairly recently as we've seen major performance improvements in ZK VMS um but it still isn't fully realized because of the complexity of doing all the sort of onchain integration bound's goal is to kind of make that seamless and bring that to everyone um that's actually awesome because I have a bunch of follow-ups there already but maybe to kind of set the stage can you also compare how other ZK VMS and other ZK EVMS differ and kind of what the pros and cons are just how they contrast to each other yeah I mean I think that um you know we when we came to the ZK space one of the questions we looked at was how do people actually use this in practice and how and one of the problems that DK had in addition to Performance was was also just this usability issue right people don't like writing circuits it's very development intensive and so we had this idea well what if we built a zkm that allowed you to run ordinary rust code inside of ZK so that the development process was easy and actually now with the stuff on boundless you can actually run solidity even inside you know the zbm um and I think that we actually sort of set a new standard right we've seen a number of other uh folks follow with this idea of a risk 5 zkm and so in some sense I think by moving toward existing standards um we've also sort of moved the industry that direction and have had seen a lot of other people do the similar things and you know in some sense the Z AVM uh itself is commoditizing um because of this right which we view as actually a positive thing for the industry and certainly for users uh because you know the technology uh becomes more similar across these different systems um yeah um how does this compare to the last version of zkm that you've done and kind of in addition to these um custom I guess runtime times which I think you talked about you can run solidity code offchain are there any other efficiency gains and enhancements on just the runtime and performance yes so I think one thing to noce is really in some sense um boundless is really like solving the sort of next set of problems once you have a zkm how do you do onchain settlement how do you buy and receive proofs how do you do aggregation in some sense though the you know the boundless still uses the sort of um existing risk 5 Z KVM although we are in addition working on new versions of the risk 5 Z KVM I think that um you know and that we are for example announced a um partnership with irreducible to build a a version that runs with that we're working with some people on the hardware acceleration side um the the short version is that the zkm itself continues to get more performant all the time uh you know I think we haven't gone a quarter without at least like a Forex Improvement in performance and that's just because people are finding different ways to make compilers faster or other like what's kind of the general Trend that you get this big of an improvement every quarter uh I mean it's really a lot of it's just engineering I mean the truth of the matter is is that there's been a lot of theoretical breakthroughs over the last couple of years and it just takes a while to get those implemented um and there's also just a lot of room you know making Fast GPU code there's lots of places for tuning um and there's just continual Innovation and I I think that what's interesting is is that you know about 6 months ago we had this major sort of threshold as an industry where the cost of proving something in ZK and verifying it on chain is actually just cheaper than gas cost even for very low gas cost chains so actually that that that gas cost change happened quickly for Ethel one but now even for very low gas cost Chains It's actually just cheaper to run your computation offchain improv it onchain and I think that we're going to see over this next year possibly another 10x of reduction in costs to the point where Computing using ZK will become just it yeah exactly the default we we actually think that most execution in general is actually going to move to offchain execution um and and and not only will we see existing protocols moving you know to that model but we'll also see new things coming about that couldn't be done otherwise right so for example Ian layer in part of their slashing mechanism is using our system in part because the amount of computation they were doing was in some cases like a billion gas right which doesn't even you can't it's a little bit more than the limit I think a little bit more than the yeah the gas limit and and and also would be completely you know financially in feasible right now and and suddenly you open up these new capabilities where you know running something that takes a billion gas is not a big deal right and so we anticipate that we'll actually see more uh ex verifiable execution and more complex protocols uh becoming possible because of this that's awesome um so I want to talk about how mless lets you run solidity code off chain you Rec call it steel uh tell us about how steel works and just exactly what's happening when somebody's trying to run an evm contract yeah absolutely so um you know with boundless we sort of have these core things like the the mechanisms of the provs and how we do settlement and then we have these additional things that we you know call extensions and that's something that we are building a number of extensions we also have a lot of Partners building extensions and so steel is sort of one of these extensions the idea with these extensions is a way to make the process even easier or to wrap even the the the rest of the thing so the way that it works with steel is that you can basically take any contract that exists already on chain say let's use ethereum L1 as the example and you can say hey I just want this contract executed offchain and I want to know what result it gives me so say it's a view view method um you can simply request that execution and then steel and boundless will do the rest of the work necessary to you know figure out who's going to do the proving get the proving done um get the resulting proof uh potentially aggregated verified on chain and then once that's done you basically effectively get a call back saying hey you know this is this result of executing the smart contract so in some ways it's almost like sort of like an AWS Lambda where really you're just saying hey go go execute the soft chain and tell me what it says um and you can run arbitrary solidity contract and and the you can access all of the existing onchain data um and so for lots of use cases where you're for example doing computations on data that's already onchain it's just like a very straight forward mechanism for people to use to be able to do large computations does it require the developer to change anything on how they're doing the contracts no so you just get to access the state of any other protocol that you're connected with yes okay um yeah awesome spefic Al the the there is some complexities around the fact that sort of there's an asynchronous answer but for anything that's like a view contract it's really it's literally that simple amazing um I think there's another kind of thing there with with kyua and um you're kind of also making this easier with rollups um walk us through what that is and uh then have a few other questions about uh some possible concerns yeah yeah absolutely so so um you know similarly one of our goals we sort of as a as a company have this approach of sort of um you know we call it sort of vertically opinionated but horizontally modular where what we want to do is we want to have an idea of some technology for each different portion of the stack and and provide that but also allow people to do their own thing too right so we you know kyou is sort of our answer to the question of how do you do uh you know a rollup stack uh although we also of course very much are happy to support people using ZK to do other rollup Stacks we have like Tao for example uh who's working with us uh or the wonderful solve Labs um but so our our view on Kalua we we actually are supporting both uh ZK fault proofs uh so that you could do an optimistic rollup using ZK uh and that the the use of ZK just massively reduces the complexity of the you know sort of fra you know the the the the fraud game so that you can make it much faster and get faster finality um and then the other we're also going to support validity rollups using kyua as well uh so the idea would be that you could run a full ZK validity chain um using Kyo so this becomes the destination instead of you're saying interoperability with other ZK chains is that right yeah I mean one could build a your you could directly use kyua to build sort of a ZK to um so yeah um so the kind of thing I'm curious about is obviously I think anybody who's trying to build something scalable would be very excited about these things because now they can codee just the way they code now uh in two uh or even if they're web 3 only developers you now have this what like a 90 plus 99% efficiency gain on gas transactions um and just instant sort of Confirmation and just a lot more simplification but it does mean that the data and all the information except that the verification proof lives on chain or lives off chain just the proof is on chain and you now kind of create a separate set of things you have to worry about in terms of having data availability be be robust and and there and accessible um how do you think about these trade-offs how do you kind of think about what does this mean for somebody kind of delegating trust here to the system and uh what they should be thinking or not thinking about yeah so I think one of the key things that we consider in the design of boundless is we are very much interested in uh very strong decentralization and so from the very you know get-go um boundless is fully decentralized and fully censorship resistant in the sense that the um sort of boundless contracts that run the marketplaces exist on the existing chains and they inherit all of the censorship resistance and um you decentralization properties of the the chains that they serve so if I want to order a you know proof on Ethel one um you know there I'm inheriting those properties from eel one now to your question about data availability um with regard to the what the ZK proof proves the ZK proof proves that some computation was run correctly on some data and it depending upon the use case how you verify the correctness of that data differs so in the case of for example um you know uh something that's being done on L1 so for example let's say uh you're using a steel example right in the case of Steel the data that you're Computing on was data that was already on Ethel one and part of the proof contains the sort of root hash of an eum block that was used to sort of see the data and so in that regard that proof in fact implies the availability of all the data the proof operated over um other use cases for example uh might be if you're doing a privacy use case then in fact the proof is Computing on data that in fact purposefully isn't on any chain and in fact may only be known to some you know user um other use cases that we're seeing are for certain rollup use cases where they want to prove a state transition function for the rollup but they don't want to necess necessarily put all of the full rollup data on chain directly they may for example use a you know some other data availability provider say Celestia and in addition to proving that the state transition was correctly executed We additionally can use sort of a Celestia like client mechanism to prove that the state of Celestia does in fact hold that data right so I think that um you know from it it does depending upon your use case there is some addition complexity regarding how you verify that the data is available but the idea is is that the underlying mechanism supports multiple ways to do that got it um who does the proving in this example and and how can is that structured yeah so the proving in boundless is done um be by a fully decentralized network of provs right so any it's an open network anyone can run aover um and there's basically mechanisms by which when a proof is requested uh provs can bid on being the prover that does a particular unit of work um and they also have to stake uh a fair amount of of of a bond to verify that they correctly produce the proof in a timely manner and if they don't they they lose their bond the end user actually gets to select sort of what level of bond is required uh in order for approver to lock in additionally approvers if they don't want to do that can also just race to produce the proof and the first one to produce it can so you're saying they can either just be available to do whatever comes in or selectively prove just a couple things that they are able to yeah exactly so so the idea here is is that we really want the the proving Marketplace to be uh you know fully decentralized and to allow anyone in there's what's really interesting about boundless is there is there's no centralized sequencer there's not even a new consensus domain produced really it consists of these Marketplace contra tracks on all the existing chains um some cryptography and a completely open network of improving infrastructure that's run by multiple parties um and there's no sort of there's no Central Authority there's no machine you can turn off to stop the availability as long as there are people who want proofs and there approvers willing to service those requests the entire thing functions and inherits the uh sort of security guarantees of the chains in which the proofs are requested and uh verified on um all this sounds kind of too good to be true um so maybe I'm going to press you on there's got to be some trade-offs um what is not as anywhere from efficient to nice or easy with this approach because uh sure this is only Pros here what are some of the cons I mean I think the one so the the really the the biggest I think uh additional complexity of doing offchain execution um it has to do with sort of composability atomic composability of transactions right so um right now in the ethereum world because you sort of are able to lock the state of the entire system you're able to you know call from one contract to another contract to another contract by multiple parties and if there's some sort of failure in the process the whole thing sort of automically rolls back right so you have this so which is you know for example that's a a core capability um that's required by the sort of mechanism for like flash loans right um uh and so with K uh proving the idea is is that while a single proof can be Atomic um you know it add does add this sort of uh sort of asynchronicity in the sense that there's the request of the proof and then later there's the the result of the proof and between those points in time state may have changed so when the proof comes in you have to verify that the state is still what it needs to be for updates to happen right so you still can do the atomic application of it so that adds us small amount of additional complexity um obviously there's you know some you know small latency involved in the process as well um but generally you know what's interesting is is that from both a latency and a cost perspective we have this very wonderful feature that as the volume of offchain execution goes up uh the latency and costs go down because what happens is um you know aggregation allows reduction of verification costs and the more things you're aggregating at once actually the more cost reduction you get so is this verification or generation of proofs uh this and this is on the verification side right um and because there really there's two two pieces of two things that have to be done once you decide you want something proven you have to actually generate the proof and then you have to verify the proof on chain um and you know what's interesting is is that in terms of the volume of proving capability this is really I think think the key differentiating factor and I I talked to this a little bit earlier with regard to this IDE of scarcity versus abundance of execution in the ZK world what's fascinating is there isn't a real limit to how much offchain Computing can be done if there's a huge set of demand for proving anyone can spin up a new node and do more proving and so what happens is as demand goes up and up the cost never rises above the cost of running a ZK prover on an AWS on demand instance because if it ever was if it was ever more expensive to prove something than it cost to run it on AWS there's an instant Arbitrage open for anyone who wants to spin up an AWS machine right and in an open permissionless system Marketplace people will uh take advantage of that exactly exactly and and by taking advantage of it they actually bring the cost back down right and so what's interesting is is that we move to this world where the amount of execution that you can do will become effectively boundless right and the uh the the amount of cost that you pay for execution will be Fairly reliable now we actually expect the cost to be lower than the cost of abds on demand because people want to build a data center to do this and they can get much Reserve pricing for yeah just buying more exactly that's awesome um well so I guess my my last question is um where is boundless now in terms of what's possible Now versus will be soon and kind of couple years down the road what do you believe uh would be possible or you would like to see more on absolutely so so um right now we are um you know know in at in a test net phase um we have many provs and uh protocols uh working you know already on the network um we are moving rapidly toward mainnet um you know which will be sometime this year um and then you know in terms of the longer Direction um I I I think we anticipate the a massive growth of volume improving and just a extreme reduction in costs right like we're talking with um you know as I mentioned earlier uh people working on custom ZK hardware and we would anticipate that the cost of ZK proving will drop lower and lower and lower such that we really believe not only will most existing execution move to this model but that really uh whole new types of projects will become possible because the execution will no longer be a limit uh for really web through applications I mean that's going to be exciting and I can't wait to uh to see that world so thank you so much for taking the time and please give him a big round of applause [Music]