ZK Proof Generation Infrastructure: Overhyped or Underutilized?

Transcript

[Music] okay welcome everybody like this will be if you don't know this will be the longest slot we have at e pra so I guess the organizer think this is very interesting topic as we are the only one with the 55 minutes yeah my name is Tibor I'm the CEO and co-founder of Maya ZK where we focus on Hardware acceleration how to make proving faster cheaper also like we call it custom firmare for gpus and yeah I will be joined today with the great lineup of the expert in their own work and yeah I have Martin

from zot toas from rindra and Pedro from astec and I will let our panelist to introduce themselves awesome thank you Tibor so uh as Tibor said my name is Martin I'm from zerit uh zerit is a zero knowledge rollup so uh you know for the uh purposes of this panel we are actually producers but also very much consumers of zero knowledge proofs uh about me I'm a technical co-founder at zerit uh I have a PhD in algorithms and complexity so not really you know deep background in cryptography but now I work in zero knowledge circuits uh I have a rich background in blockchain industry I've been uh in the space uh since 2017 or 2018 starting as a security researcher

uh eventually I became game uh rollup developer so yeah nice to be here thank you hello Thomas from Rockway X um currently head of the infra division in our company we are running the most efficient ethereum validator according to rated Network we are also running some Solana validators and now we are like exploring the space of ZK basically providing the hardware in the most efficient way hello I'm zedro I'm a Deval engineer at Aztec Labs we are uh encrypting ethereum basically that's uh that's what we do it's an L2 uh private L2 on ethereum you

can have both private and public transactions both private and public State and you can do really really cool stuff for that and yeah obviously it involves a lot of proving so that this panel is quite it's very interesting for us nice to be here okay so the topic of this panel is actually like ZK proof infrastructure overhyped or underutilized it's kind of clickbait topic I would say so hopefully at the end we will be able to tell if it's overhyped or underutilized but okay I will start with basically Martin if you could explain like the ZK proof uh chain and also like what does it actually mean like ZK infrastructure in your own View and also from the Z

side all right so uh the Z knowledge proof supply chain is essentially the entire pipeline everything that happens uh between some kind of an application knowing that it requires a zero knowledge proof and that zero knowledge proof being submitted to the verifying entity that is consuming that zero knowledge proof so in the setting of a zero knowledge rollup we use zero knowledge rollups for scaling and for convincing the verifier which in our case is a smart contract on the L1 that we have correctly executed the uh transition function stipulated by the evm machine so zerit has a sequencer

sequencer organizes transactions into the block we are waiting for a whole bunch of blocks to show up eventually we take the state route and the ZK proof for us um certifies that that state roote was obtained properly and that state root is submitted to the verifier on chain uh what it specifically means for circuit is that we have some kind of zero knowledge circuits that uh specify the logic of the evm that we are you know running on our blockchain and those zero knowledge circuit operate under the hood we are running the machines that or we are running the the software that generates uh those provs and uh we are currently doing it uh internally but I'm

sure that we will sort of you know get to that topic yeah okay so like Martin was basically describing like ZK proof chain for like the zk4 scaling as the ZK L2 but Pedro you are from the ZK ZK rollup like uh you aaging ZK for privacy so how does like proof supply chain looks like for the a um It's Complicated yeah basically it's different from from usual um in the sense that we some are made by the client so when you make a private transaction that transaction is actually run by you as a user on your phone

mobile or like desktop whatever and that proof will be rolled up um and will be sent will be verified and together with the rest of the public the public proving so there was two sides on one hand we want proofs that can be very easily recursive um and fast on a phone on the other hand we we want proofs that can do this scaling and can actually uh roll up as many as many transactions as possible into a root rollup that will go on an L1 and that's the biggest difference uh and that's that's the thing that we care about uh it's the kind of ZK that we do

is actually zero knowledge so we are actually hiding uh some some properties we are hiding some we're keeping retaining the this privacy um and uh and at the same time we also want the the scalability while with other proving systems that are mostly aimed at at scaling uh they don't necessarily uh keep the the zero knowledge part so I would say it's mostly like the succinctness part of the of the snark or the Stark more than actually the the Privacy pring Parts um so yeah that's uh that's basically where we are with with the proving uh both on the client and and on the server and definitely this kind of

proving marketplaces those are uh um and just the the way that the sequencer can actually offload the proving into other parties this is in in our proposal for the what we call the side car uh which involves this kind of marketplace yeah yeah perfect we will get to the ZK marketplaces because this is my favorite topic When I Was preparing for this panel I counted like six different ZK marketplaces but we currently have like just five big ZK rollups on deployed on mainnet so it's interesting to see we have more ZK marketplaces than rollups but the question for you to much you are basically on the other side of the ZK

supply chain those two guys are representing like the consumers like they are in need of the ZK LS but you are the one who can provide right as you are operating the infrastructure so how does the ZK supply chain looks like from your view from my view basically it's all starts with electricity you need to have it electricity and then you need to have like a hardware which can like start counting whatever nonsense these guys prepared for us to count I'm just joking so yeah we basically for us it means to be able to be fast and also like cheap because like as of today the costs are pretty high so what is my mission and my team is to basically to build the most efficient prover for for the for Market

more uh efficient Pro like setup let's say for for the marketplace itself so we can be like pretty competitive and also like the users in the end will pay less for the whole stack so basically what we are doing because uh we are like bare metal operators so we own the hardware and we can touch touch the hardware and play with it so what we have basically achieved we have gained maximum performance out of the hardware so it's a everything is running pretty smoothly and now we are waiting for those marketplaces to go live and get truction to to show uh how cheap it can be okay so we talk a lot about like we mentioned ZK Marketplace but actually

some projects they are recognized as like ZK as a service and maybe it's not clear so like question for everybody do you make like difference from your application if there is a z you to much as an infrastructure provider or from the aex side is there any difference between ZK Marketplace and approve as a service the there is for for us the marketplace that means like we have to compete and there's like multiple competitors there are like multiple things you you have to solve when it's like a proof of service we basically own the whole stack so like we just provide a service but the marketplace is also like a service but there are like multiple things which has to be like um

um prepared or taken care of so uh when I look at uh ZK as a service or a Marketplace essentially in uh my position or in the position of zerit we have some input for the prover and we require a zero knowledge proof to be generated right so in this sense uh it is uh actually almost equivalent whether we request this proof to be put on a Marketplace and then multiple entities you know somehow compete for who produces the proof cheapest fastest whatever and then we consume the given proof or whether we actually request it from one service that is capable of delivering this proof right uh there are

definitely going to be some differences in terms of uh you know the qualitative delivery of of the proof will it be cheap will it be delivered fast will it be delivered reliably right uh so those would be the differences but again for us as a consumer of that proof that we need to submit to our verifier the bottom line is that we require the proof to be present somehow that proof needs to come to us and we want it as quickly as possible and as cheaply as possible it can be achieved both via the marketplaces and you know uh ZK as a service type thing one thing we care about uh talking about this like it's about the the speed

and and like what's the impact of uh actually having faster and faster appr uh either by Hardware Hardware acceleration um and whether that could actually pose a problem of of centralization and this is something that we that we care about like I I remember when I when I I remember being in a Jordy Bina talk and he was starting to talk about transaction in terms of how much they cost in in AWS in instance I was like oh this cost 7 cents in a I was like wait we're talking usually about gas prices not about like the amount of dollar you need to to prove something but that's actually the cost the actual cost of uh running a machine it's again the actual electricity you need the

actual manufacturing of of the chip if we go to a to a fbga or an asix um thing you know and so this is this is kind of we want that to be uh obviously it's going to be competitive it's going to be uh provs will compete to prove the most and to prove the cheapest but we also want to avoid de centralization and to avoid that big uh big companies can actually like envidia or something just starts proving and starts their own Pro and just eats up all the all the all the proving um space on Aztec and that's actually poses a problem because they could uh halt blocks they could just

wait for their L1 sequencer to stop uh and not accept any L2 blocks until it's there there's just so many things that can go wrong and can collude uh that we care about is decentralization so we are a bit aware about the we all want speed we obviously all want Hardware acceleration but this is a concern that we do have I don't know my fellow panelists if do if this is something that crosses your mind and how how do you think these kind of things will go uh and actually pose a problem in the future yeah okay so I will most likely skip couple question because you already touch on like interesting topic or like topic it's discussed but it's not clear the uh what's the answer so the question

is simple is there any risk associated with centralized provs as soon as it's like ZK so it's sound and complete right so you cannot prove something uh which is not true and if you are able to prove it you are able to prove the whole statement so from this view I mean like centralized Pro as soon as there I mean let's say there will be no like sequence of censorship and live issue is it totally okay or not to have centralized entity generating all proofs of one chain or for multiple chains so uh this question actually has multiple layers right um the first question is we are generating a zero

knowledge proof the user always needs to be asking is this the zero knowledge proof that was supposed to be generated and that actually proves the given claim right in our particular case the claim that we require to prove is that transition function stipulated by the evm was executed correctly and honestly if I Implement zero knowledge circuits that prove something completely different you know data was seen there was non zero amount of bytes that came as an input that can also produce zero knowledge proof but it's not that statement that we really cared about right so the first point is if you are running a centralized prover then you

probably want to open- Source the circuits that actually uh show the logic uh that you're proving and then you are relying on the community to sort of uh be able to verify for themselves that these circuits prove uh what they need to prove now if you wanted to run a zero knowledge rollup uh with uh completely centralized prover this is um this is uh essentially still not enough right because if I unplug that machine from electricity as I said that's where everything starts uh if I unplug it from electricity it will clearly not be producing zero knowledge proofs right

but the zero knowledge rollups the ability to record its uh State Roots uh on L1 it's fully tied to the capability of providing this zero knowledge proof right so if I lock it down if I turn it into a c calized kind of operation there is a good chance or there is a risk that that service will at some point go down how this would impact the users is that the rollup at that point it would still be producing new blocks on L2 but if inside of those blocks there were some exits there were some withdrawals of assets from that L2 back onto L1 those would not make it onto L1 because there

would be no record on L1 that these exits were H that these exits happen happened that these exits happened and were actually derived uh correctly in the state and therefore you would not be able to prove the presence of your uh withdrawal transactions in the state of L2 to the bridge to the canonical Bridge provided by the ROP and therefore that bridge would not release assets so uh if you want to have uh centralized proving service there is one more thing that you need to implement uh in your zero knowledge rollup and that is called an escape hatch you essentially have to give your users some kind of opportunity uh to exit with their assets even though the zero knowledge proofs

are currently not being produced right you can think about it in multiple ways this actually already has nothing really to do with the zero knowledge technology you can either give the uh the users the capability of generating that proof themselves right you can give them the inputs and you can give them the software and they can run it on a very beefy machine and so on but that's no longer centralized right uh or you can imagine some other mechanics saying hey if we haven't seen zero knowledge proof in I don't know 7 hours that means that the service is currently unavailable right so some other smart contract mechanism would have to kick in that enables the users to actually say hey I kind of want to withdraw everything that I had on that chain so this is what we

call an escape hatch and that would allow you to have uh zero knowledge rollup that operates a centralized provs and is still trustless in this sense but again once you adopt this path you again have to engineer for the trustless really really interesting what you what you said about the escape hatch because before Aztec came in we had Aztec connect some of you may know and and Aztec connect was sens but we will still run a sequencer for about a year and now we don't run it anymore but if you if you have funds logged in there you can actually run your own sequencer

so you can use this escape hatch and and we have instructions for that and we have the code and everything so you can actually go and run it's going to be very expensive because you're going to pay for the W block for you but I mean if you have't that amount of funds you didn't draw in the past year then you can do it on your own that's really cool like because it's completely decentralized like we don't have a sequencer but you can go and withdraw your funds if you have them in there so so and another another point I we're talking just to answer your question about the about the the fact that the proof is a proof and you can actually change that that first of all it's not really true I mean you can you can actually prove something that's not correct as long as there's social

consensus on this right and so you can always Fork but then you don't have no prover so there was there was this uh dance between having some kind of social consensus and and also having a aover for that so you don't want a centralized Pro because you don't you don't you don't want the the the chain to go in direction that everyone everything is fine and in in in you can have for example uh sensor transactions in there and so this is the special the censorship resistance resistance is the is the biggest part of not having a centralized prover but that said I mean usually people ask me are the zkm safe I say I mean if the circuits are then approve is a pro and that's right it's

it's a bit it's a bit like that uh I don't know if circuits are safe but uh yeah the proof is the proof but like let's say the circuits are enormous great like I mean the traces the witness generation it's so great like you need like this crazy 96 CPU 768 gigs of RAM uh I don't know nmv you know h00 then like the proof is a proof but not everybody can generate the proof do you see also like a hardware requirements problem for ZK proving like in a way of decentralizing the pr if not everybody can uh let's say it's kind of like Solana versus if right validator like not everybody can like like I know but I will not say it uh run the proofs for

most of the ZK l2s so what's your take on this like does a Aztec actually do something so you know every Joe can actually run the PRS yes yes we care about that a lot you know there's seems like every few years there's this ethereum killer project that just does a different iteration of increasing the block size so like why don't we just increase the block size yes that's the project I'm going to be and and and that doesn't work like we've seen Bitcoin cash T try that didn't work and and there's many projects that are just like oh I mean if you run a sequencer that has a lot of resources maybe you can have more transactions that's not the point that's just not the

point but with Prov is different because again approve approve is approve still we care about that we care and and so our proposals for for for proving systems are mostly in the sense that you can prove part of the tree because you know how the tree looks like like the TRS are there you made the transactions you know how the tree looks like you just need to prove it and so you could for example commit to a SP to just to part of that uh like a sub tree of that Merkel tree and then commit to that and have backup have fallbacks have whatever and then there's a bigger proof that commits to a bigger sub tree and so that's perfectly possible to have one one proof one root proof that's going to

be verified on L1 and have multiple provs proving that proof because proving is not exec it's not executing you can you know how it should look like you're just proving that the the the computation was correct so that's perfectly possible to do on on Aztec but we don't care we don't care like our puzzles is mostly the sequencer chooses the sequencer can do an auction the sequencer can decide proving themselves the sequencer can choose and this is very interesting because we we have no opinion really about uh how how will the proving system work yeah and we will get to your like request for proposal

because this is very interesting how Aztec is building the ZK St but a question for you toas like uh you are done the infrastructure provider you are the one who says like okay this is how the bare metal will look like I need to order like one 1,000 I don't know age 100 so basically what challenges is associated with this for you because we have like I mean like each rollup has its different proving system in a sense like at least tweaked then we have like this Z KVM this is like totally different landscape and I assume you want to serve the majority of the market so like how do you actually serve how do you choose the hardware like which is the right for the ZK proving and maybe if you our audience would like to you know participate in a proving are they

even capable of like buying such a hardware and operate it on their own yeah first it's complicated because like even uh like all the projects are a little bit different and all the projects are evolving so like uh at the beginning it was all CPU based and like people were thinking okay let's do it on a CPU and then like there was a journey with fbga oh let's do fpga so everything will be on fpga and it turn out fpga is s of they are not the best way possible then then was there was a hype of okay let's do gpus okay like we have found like a good GPU which is like a good uh

performance perir um electricity spent which is like Nvidia L4 which is really nice card and there are there are also like some projects uh using uh or telling us hey let's buy Asic it's super cool like we have uh created like a special processor which is optimized for ZK I'm like okay cool this sounds interesting and like me U tinkering with gpus it doesn't make any sense but then researchers came and said okay we will use like a different fields and like uh different mathematics and that Asic is right now not usable so like for me it's as of today how I see

it it's basically like uh tinkering with CPUs and gpus and finding right balance and basically like every project needs a different ratio and it's also changing in time so like the configuration I buy today might be useless in a uh in a year or maybe in a week so like nobody knows and basically what we are doing is working with each projects and trying to fine tune the configuration for them to work uh with the current code but also like looking at the future so they say okay we are the ones we are able to offload everything on GPU which everybody claims but not everybody has

uh really done it so we are like going slightly T to more like optimizing the the gpus itself and as you ask for if it's possible to like buy a machine which can generate the proofs at at your home you can buy it you will pay like a lot of money for electricity and your wife will hate you that happened to me because like I started with e mining I bought a rig and had it in my home and like she was hating me this has to go away so don't do it if you like your wife okay so you're basically saying like we don't know how the hardware stick should look like for to S majority

of the market uh ZK market so basically then maybe the question is like uh isn't it a bit too early for this like ZK as a service ZK marketplaces if we don't know what the underlying infrastructure and I'm now mean a hardware will look like so it's just easier for projects now I'm talking about the zkl tools just you know stick with the proving inside the AWS or like a Google cloud and you know like it will be less risky I would say they can if they have a lot of money and good backing from the VCS but we have been running some marks uh comparing our our setup and AWS and I think because of AWS is pretty generic

for mostly like AI workloads they cannot the AWS cannot deliver the performance so like I think it's better to find like a provider which can work with you and give you like better efficiency per dollar I would say because that's pretty bad on AWS yeah so this is mostly because AWS sucks I cannot tell it it like this but like for the ZK workloads it kind of sucks because uh the gpus are pretty good on AWS but the CPUs which uh still like most of the algorithms rely on suck okay thank you uh we are in the

half time basically so I would like to maybe give a chance to our audience if you have some question one or two questions and yeah go ahead thank you hi um I was interested in because you talked earlier about um centralization of the of the pro of proving Market or so to say um for rollups it's basically not that much of a problem if it's and because we usually do not care about this zero knowledge aspect of a zero knowledge proof but more about the SU synness of the proof so I think it's not that much of a problem but would you if you if we go one step um ahead of that and say we also care about the Z knowledge aspect

um how do you think would be a pro provider um handle this kind of problem so because I have to disclose my witness to you if I want to um if I want you as my proof provider and if I care about the Z knowledge aspect I usually don't want to give you my witness so um would be interested in how you see this problem or is there any way to solve this and the problem was so basically if I want to compute a z knowledge proof and I care about the Z knowledge aspect I need to give you my witness so that you can perform the witness extension or or maybe I have to do the witness extension on my local device which is for dkml for example would be kind of

expensive and then you can compute the proof but for that you have to learn my secret data sure you are absolutely correct uh so when you care about privacy uh with zero knowledge Roll-Ups uh by the way we do not do privacy as their kid but if we wanted to we would have to Blind the inputs right uh the fact that we would be using external party to actually produce these zero knowledge proofs you're absolutely correct again we would have to disclose those inputs to that external party that is producing the zero knowledge proofs so that yeah so uh you know the alternate option there would be actually having

the user to generate the zero knowledge proof themselves right they they know the input they do not want to you know show that input to anybody so they can right away come with the zero knowledge proof and somehow provide that proof If the rollup is ready for that to the rollup and say hey I've done something in here this is the proof that I did and the rollup might say oh okay well I I'm going to submit it to the verifier and if it is really the case then then there you go uh it it actually goes through but it it it is a valid problem when you talk uh to the about the marketplaces and about external parties but I think that Pedro has a comment I just need to say this is kind of sneaky question because those guys are from Tao so they

want to heard like you will do NPC and everything will be okay yeah thank you there some contributions to Noir it's great great project anyway H yeah but as as you were saying like I I was I was like jolted but didn't makes perfect sense like you always want the user to have to to make that proof the user could choose to hand those private inputs to to the pro and then act that's actually fine I remember some time ago we were in a in a hackathon and in the same hackathon were syry and like most people were just doing all like putting all their proofs on Syer so they could hunt both bounties I said why not I mean it's it's just like it's up to you to decide on whether

do you want to give your private inputs to someone else to prove or not but the fact that you have this option that's the most important thing maybe it's not a big problem for you uh if you want to do that but most of all regarding regarding that about the fact that you have different ways kind of different types of ZK in a way I don't like to think about that but there is but there there was also different proofing systems you for example for scalability people people use a lot of starks but for for ZK for Z knowledge for privacy people use tend to use snarks and this this means like also different hardware and one of the big problems of Hardware is like it's kind of slow development like I was like you know it's very it's

a slow development you it you you can take like two years just to design a circuit then it actually needs to be like physically manufacturer before you before you can test and the reality of proving systems is that they change every month so there is little it's very difficult to make like a proving system like something that magically works and like you can just fit an input and it will come with a proof outside um and I think that's not where where Hardware development usually usually goes they usually go on small like ffts and and small parts of the the proof that actually they are used everywhere and so it's kind of okay to have a specific circuit or Asic or whatever to do that

um but yeah I guess that's this is the the the biggest difference is like maybe maybe we need to reach this point where proving systems are standard are standardized we can reliably trust that uh this snark or this this way of um making the proof that we are settling for some checks or something that that will take that will settle down for a year and then we can maybe think about uh having hardware for that but that's really cool it's really cool question really cool development as well yeah thank you for a question uh do we have one more question okay thank you um how does economics look like as infrastructure provider

like how do you cover your costs like electricity and etc etc is it is it project per Project based or like can you can you give some rough numbers rough numbers so it's pretty capex intensive you have to buy a lot of of Hardware uh you need to find a good space where the electricity is cheap or you can build your own place where you can like have the data center and yeah then you can get like much cheaper than a aw us if you if you find tune all those like bits and pieces you can be like much better

in that that I know clouds are more expensive than the than in in house like infrastructure but uh how do you get paid [Laughter] like how you get the as of today there are like not many paying customers as of today because we are like still in the face where we are like talking to the projects and trying to optimize it and like they like there is like one deal we only have which we are like currently working with and right now now we are like paid as like an IT service in dollars okay just like usual service provider yeah okay

thanks actually this was great question I have like prepared uh couple question on this topic and I think I will skip to it right now so I will ask maybe the audience do you have rough idea or any idea how much the zero knowledge actually costs how much do we have to pay to generate like you know 500 swaps on I don't know Unis swap V2 on ZK L2 up around $100 $100 to prove zk2 block on zkl to block 500 transaction of Unis swap swaps plus the gas sorry plus the gas for verification uhh okay uh so

we're getting there no so okay the question is how how much does it cost to prove the block we don't count the verification or da $2 $2 okay some other guesses okay so it's uh it's a really interesting question and I don't know uh but I will take what I know so basically the guys from the polygon they presented in Denver before IEP okay so before blobs so to transfer like uh 500 transaction of Unis swap swap it cost like $1,000 for da this is like data availability cost and this counts for 96% then we have like sequencing fee

which is like around like 0.03 which is like free percentage % and then we have a appr fee and this is like 0.0013 so this is how much they pay for like running their appr actually to generate the ZK proof so I will ask similar question do you think is actually providing like proving Services you know viable business like can we make money similar to staking Services if it's like the proving is just 0.13% of the zkl to block costs yeah we believe so but definitely it will be

like a much smaller Market in my opinion in the crypto space than like staking or like Bitcoin mining it will be definitely uh smaller market and the profit margins will be like definitely lower so that's why we also like optimize for efficiency but I believe like the the ZK proving will not be only about like scaling uh SC scaling for l2s and so on but also like other use cases which can like I don't know like ZK pass ZK email and more like General proving for also like non crypto use cases in my opinion because that will be much bigger because it's like vitalic said like everything

will be in the end aggregated into one proof so that there is basically like race race to bottom yeah yeah I mean z is the end game of course uh Pedro one thing is like you mentioned a th000 th000 uh k for gas for da that's the biggest cost that's ridiculous like like it's you know it's 1,000k for posting it's not even execution it's just data and so that's the reason why people are talking about all the these DDA Solutions because it's actually something that's it cheap and it's like artificially or because when eum was born it wasn't something that we cared about and that's the part that that's just the price is wrong the price was

wrong from the beginning because no one know knew we were going to get get there but if we talk about blobs if you talk about all all these da solutions that will not be 1,000k so it more together with the proving time with proving size and with proving cost and so from then on if you see that there are more like if you spend like 30 30% in each one of these proving sequencing and and da then then yes that's definitely that just just as much of a market of proving proving marketplaces than is for da uh it's just a matter of like having more people using the network are there we don't know I I think without privacy

there aren't yeah maybe Martin you are the one who actually you don't have to disclose the numbers but do you think like the as it is right now like ZK proving can be like sufficient business like compared to how much you maybe pay for approving it's complicated so uh again the question is very loaded right if uh I want to be producing these proofs myself myself I have to be running these provs and that doesn't come for free right I need to have some people who make sure that these provs are running the machines at some point if I run on bare metal I mean they become obsolete they burn whatever I need to replace hardware and the

question is do I want to be dealing with that or do I want to be developing my rollup right so uh I think that there is space for ZK proving uh you know once these marketplaces come online once we have such Services widely available it will actually become much more simpler for people to develop zero knowledge rollups because you know if you right now imagine the situation of a rollup operator uh that wants to run the proving centrally you develop your circuits and then you know as we discussed earlier I think tore mentioned the specs of the machines then you have to you know obtain those High spec machines with uh what 400 gigs 800 gigs of RAM you know very expensive uh gpus

and so on once you start using the marketplaces you actually completely offload that problem to the services that are hooked up to the marketplaces or to the services themselves right so it really depends how much of the troubles you want to deal with uh yourself as the rollup or as the consumer of the proofs and uh you know less troubles shorter time to market right so I I think that there is definitely some some opportunity uh how large the opportunity is uh that really depends I know that uh circuit produces blocks every 2 seconds right so in some sense the amount of proving that we have to do it's capped the blog has specific capacity one block comes every two

seconds well you know this is actually not going to go up unless we reconfigure the chain right so we have uh let's say a capped amount of business that we can give to these marketplaces and if you know that then the amount of business and the size of the market really depends on the number of projects that uh require such a service maybe a question for you guys do you need like a redundancy in generating those proofs or you are just okay with like one provider yes uh so um again I mentioned it earlier uh if we do not have a zero knowledge proof available we are not able to roll the state up right so it

really depends on how you architect the rollup if you do not have your escape hatch and you know even if you have your escape hatch and you do not want people to start exiting using your escape hatch then you better have that redundancy that actually you know gives you a little bit more reassurance and a little bit higher availability and lower chances that your proofs will not be available at some point so it is a valid argument and yes it comes into the account we actually you know we we are proving on AWS despite they're sucking um that's not my statement we are pring on albs and and we do not run a single Pro we do run multiple provs because of the redundancy problems so uh yeah it is a problem that needs to be

thought about uh on our side like we are still in the process of uh developing the proofing system because it's yeah it's kind of very specific to astec so we're not there yet with actually having the Prov uh ready for user user consumer but yeah we still use AWS for our for our stuff uh basic basically for our proving but uh again this is we want that to be accessible for both consumer level and Enterprise level uh given this kind of uh sidecar solution where the sequencer

will choose maybe it would auction for example uh their proving the proving of the block they have sequenced uh and even the payment will go uh will go maybe for from the sequencer to the prover and not necessarily from the network being to theover and so it's up to the sequencer to decide on who is going to proove that block and he will not be posting that block and if there's no one posting that block maybe there will be some kind of racing racing um system where like the the the most the fastest whatever block comes in after a certain period of time will be as long as it's valid will be included um and so yeah definitely there there's going to be this uh but that also presents some other problems uh but yeah it's we

wouldn't care too much and the truth is like most most sequencers nowadays most validators those these days are are on the cloud as well so like if you chut down AWS uh what's going to happen to the human Network that will be interesting so it looks like everybody is running in AWS so this maybe answer our question if the uh this infrastructure is overhyped or underutilized but okay like both of you guys like Pedro and Martin uh you are on the main net right like mayet sorry devet mayet soon so basically could you share maybe like do you have some plan for decentralized proving or this is not like a top priority for you because with

the main at Lou is coming much bigger problem than I guess Pro so what's your take on like decentralizing Pro it's a first priority for us because because we're going to launch decentralized so there has to the proving has to be decentralized we we rather wait for like next year to have something that it's actually reliable and then launch with a fully decentralized Network then then like launching now and like oh we sometime we will decentralize the pro not shaming just saying that that's that's our goal we also are building something that I think no one other no not other any other project is building so that's also so uh I guess to our advantage we do

have the time to develop things slowly and be sure that what we're doing but yeah this is um like top top priority yeah so I actually have the other stance uh for us is actually not top priority as tior mentioned uh there is a lot more than just proving to you know running a zero knowledge R up or an L2 in general so when it comes to these type of questions do you want to decentralize it do you want to offload it right it always for us at least comes to the cost because uh as an operator I have two options it is definitely going to cost me something and I can either project that cost onto the users meaning that

the gas fees on the network will be more expensive or I can absorb that cost saying hey I am a company right so I have probably some sources of revenue and uh I can actually choose to fully or partially subsidize operations of of these Pro provs so our stance currently is that uh we are actually fully comfortable uh operating the provs ourselves uh and fully subsidizing the cost hence we actually do not require that decentralization to be such a burning topic that we need to address it immediately uh with decentralizations of proving uh it you know you also kind of want a little bit of maturity in your circuits Pedro mentioned it earlier circuits you know and the provs they

actually change every month everybody is uh upgrading their circuits upgrading their circuits making sure that they are more and more efficient because you know we we've said this magical word zero knowledge proof here but it's not true that a general statement or single statement has one and only one zero knowledge proof that proves that statement you can actually prove it you know usually from multiple angles right and depending on what angle you choose and how exactly you do that uh that's going to influence the computational cost that you are going to have with that so uh you will want to be at some point where uh or at the point where you are fairly aware of of your cost you are fairly comfortable with that and you are

saying fine this is fine to offload you know on on the other entities when when you come with decentralization in our particular case we are running ZK evm again we do require those machines that have very high spec uh we do not think that uh there is you know so many of them that are available out there for the general community so we are not putting this as top priority once you know our main net is out our engineering resources are a little bit more relieved then this is definitely a problem that we would like to tackle I also kind of want to address one misconception that very often comes in the space so very often when people say something is decentralized uh you know the community

sort of imagines that it means that it's always up and it's always running and you know somebody's going to do that that actually is not the case the fact that you decentralize something is that you provide everybody with opportunity to perform that action right but if nobody's watching if nobody's interested well nobody will be proving for you right so uh in the end even once we decentralize and if we decentralize or once we decentralize um uh it will still be a future where we have to make sure that there is at least one entity that is providing those proofs otherwise we will have those proofs right so in the sense there is actually not so and so much uh you know immediate benefit from

decentralization for us and that's why it's not on the top of the road map to much so I asked Al about the about their future but what about you like uh what's your take on like actually zik Hardware acceleration and will the ASC hit the market and how do you prepare for this because we have this like you know no company who is working or like many companies working on this and I think like you just bought a new hardware it was pretty expensive and then next day boom we have a ZK Asic so what do you do actually not going to happen that's a great question there were there were times like when I was

pretty worried like hey we are like buying those gpus and then like the companies come with the A6 but I'm not worried of today because like like building the ASC and like then manufacturing the ASC it takes if you are like pretty good year or two but like look at the history like uh what kind of achievements we made in two years so like they are basically not able to cope with it and it will take some time and I don't know how long when the like the ecosystem and like the all the AR algorithm are stable enough that it will make sense to have an Asic like

it happened with Bitcoin that it's pretty standard s Sha but like in the in the CK space maybe I'm totally wrong but I'm not seeing that in a future like two or three years when this will happen so uh toas is absolutely wrong here he should be extremely worried about the A6 coming out uh the reason is that once they do he's going to buy one or 10 of them he's going to put them in the kitchen and his electricity bill will go through the roof and your wife will hate you again okay uh we have just three four minutes left so let's close us with some fun questions so basically question to all of you what do you think will happen

first ZK like actually Nvidia will start working on like ZK GPU bit main will shift from manufacturing A6 for Bitcoin and start manufacturing a for ZK or actually AWS will launch ZK as a service in my opinion I'm like bit main my opinion is AWS will launch ZK as a service yeah I'd like to be of the other opinion but I don't it's I think I is going to consume the whole thing um Nvidia doesn't care right they they they are with AI already they they have already have all the all the market for

that and bit main that's okay for Bitcoin I mean they are the only ones that always win with Bitcoin so and for a it's it should be pretty easy for them to just have um specific Hardware um because they have a lot of it so so I absolutely agree with that right when you look at Nvidia uh they are manufacturing gpus we already compute at gpus they are Absol absolutely fine right on the other hand that's uh why I answered the way I did uh all these Cloud providers they have purchased these gpus already right so they need some load on them and uh running a ZK as a service uh that's sort of a logical

use for them so it's a viable option in my opinion I totally disagree with you because they don't because they don't need a load on their gpus they have they have enough demand from Ai and in my opinion it's too Niche for them to like optimize for for ZK but like I agree with you like it's uh it makes sense for them to do it but it's too too too too small and they're like occupied with other workloads okay so we can meet here in one year and see if the Nvidia is working on actually AWS zik as a service thank you all for coming and thank you our panelist for joining me thank you everybody

[Music]