Sreeram Kannan - EigenCloud: The Verifiable Cloud for the Agentic Era - TOKEN2049 Singapore 2025

Good morning everybody. Really excited to be here. I'm Shri Ram, founder of Iden Cloud. So what I'm what I'm going to talk about today is our new uh progress into the agentic era. So the talk is titled verifiable cloud for the agentic era.

Really excited to come speak about this year token. So before we go into the new things that we're doing at a cloud, I'm going to start with some numbers right. So we raised you know I cloud raised like 250 million total 170 million from Andre um $20 billion plus of each stake the number one US protocol by TVL. So you know hundreds of developers building really interesting new services on top um the fastest growing developer ecosystem when tried for number one DA protocol a huge unthriving ecosystem you know $500 million plus raised by the ecosystem really valuable okay so this is what we've done till now what are we doing next and what's our vision and why are we doing this so this is our uh recent data When you think about crypto, crypto is about building unstoppable verifiable applications. So you think of Bitcoin, what did Bitcoin do?

Bitcoin is an unstoppable money, right? It runs forever. You know, there's only 21 million bitcoins and you own your coin. So it created this $2 trillion asset. And when you think about what has happened beyond that, Ethereum created this unstoppable verifiable computer and a computer in which you can deploy programs and these programs run continuously.

So next era I think what happened is these things got scaled more right you know Salana for example took this and scaled it up to say you can run sinance at the speed of light. Okay, so this is how we see the evolution. But underpinning it is the scale of compute. So you look at Bitcoin. Bitcoin can run on a $5 compute.

Ethereum sits in a $500 laptop. Salana runs on a $10,000 server. Same thing with other L2s. But to actually realize the full potential of crypto, we need much much more. We need to be able to fit massively multiplayer games on chain.

We need to be able to set AI agents that you cannot run on any single chain today. We need to be able to set a Uber, the Airbnb, OpenAI, and AWS into these chains. How do we do it? That's our vision at Cloud. You should be able to run billions of dollars of compute to run any application imaginable on the verifiable unstoppable rails.

So that's our goal at Cloud. Millions of AI agents holding tens of trillions of dollars. That's what we're targeting. Okay. How do we get there?

Okay. So this is the vision. How do we actually get there? So we think about it. Uh you know I get cloud is not an L1 or an L2 or or an app.

What is it? The third thing it's basically allows infinite scale computation offchain but with smart contract scale verifiability. So you run the application logic offchain and you can actually bring those results back on chain securely. Um and when you think about when you run a computation offchain, how do you know the computation run correctly? If you're just going to run it on Google and then bring the results onchain, that's not interesting.

It's neither unstoppable nor verifiable. So when you think about the grade of verifiability, we have many different ways. You can just write the application fully on chain and that's going to make it verifiable. that writing it for you upchain onchain constrains you massively. You're constrained by the virtual machine, you're constrained by the programming environment, you're constrained by the hardware, you have lots of constraints.

So you have basically this set of trade-off between what you can do on the cloud fully programmable, highly expressive to what you can do on chain and there are all kinds of trade-offs in the middle and what we do at a cloud is let you choose your own trade-off. you can actually find uh you want to run like a hardware security or cryptoeconomic security or a ZK based security you can actually choose which point in the trade-offs you want to operate on okay so computation runs offchain and then the results are brought in via onchain so that's the basic premise of cloud so how do we actually arrive at this how do we get these properties how do we get to cloud scale where you can scale computation infinitely while maintaining crypto grade trust because if you lose the second aspect then we don't really get the properties. So the way it works is you think was you know on Ethereum on any chain you can have the staking contracts and layer lets anybody stake and run computation like I pointed out you have tens of billions of dollars of stake. So you stake on layer and then you can run compute compute is a container standard where you can just like take a container deploy to compute it runs where family and brings the results. So the node is running compute and what it also does it archives the snapshots into IGA.

IGNA is a data availability network. You think of it as a storage network. So you where you can probably retrieve data from it. So you throw the snapshots into DA decentralized network. The node running compute may be centralized.

There's a single node running it. Okay. So the node runs it and the node brings state back on chain where you can actually take actions on chain. Okay. So a single node stakes running the computation writing the data to DA and then brings these results back on chain.

Now you can actually uh take lots of interesting application patterns where you uh run the computation offchain and then take actions on chain. I'll give many examples later. Uh but what happens if this single node there's a single point of failure if it goes out okay this node is gone now what happens is first on air you have this thing called slashing slashing means if a node misbehaves then it's going to lose its stake so that's why we have tens of billions of dollars of stake on layer okay so it's going to lose its stake so it it gets slashed but just slashing it imposes penalties for misbehavior but doesn't make the system unstoppable able it is stoppable. You just need to pay your slashing penalty and you can stop the system. But what we do is any other node can come in and stake and pick up the computation from where it was sos.

So what you do is you go and retrieve the data from DA and then you can actually continue on and run compute from Danol. So the key thing to observe here if it's all too technical is to abstract out you have layer for staking on DA for DA availability compute for computation these three interlock to produce unstoppable verifiable compute at cloud scale. Why do I say cloud scale? It keeps stretching the amount of computation available keep stretching because anybody can come and stake and provide compute and keep running. We already have hundreds of operators billions of dollars of stake.

Okay. So that's the goal of IGN cloud. So in this uh slide I'm comparing cloud to an existing web two cloud. How are these two things related? Then how do they interact?

So at the base layer of a regular cloud is a data center. Right? So you have Amazon or Google at the base layer is a data center and the data center is where everything else is built on top of a data center is centralized and nonverifiable. At the base layer of cloud is layer layer is decentralized and verifiable. You have nodes that come in anybody can come in and stake and run operators and keep the system going.

So that's the first thing the analogy to a data center is there. Then you go to a cloud you don't just get like dark metal data centers you also get a set of very useful services storage computation AI being the top three services you want to store data you want to do computation you want to have intelligence and so those are the three services that are available the most popular services on top of an Amazon web services that's exactly the three things that we are building on top of layer IDA for data availability, IGEN compute for computation and IGEN AI for AI inference. So building you know I mentioned earlier about how compute works uh in terms of running offchain compute and bringing bringing the results back on chain. The same thing is true for AI. You want to run AI inference offchain and bring the results back on chain verifiably.

And so when you think about running AI offchain, one of the biggest constraints, biggest problems is AI is non-deterministic. If you run the same, you know, if you ask Chad GBD the same question couple of times, you get a different answer. And the reason it's non-deterministic, the surface level, there's a thing called the random seed. You know, there's a randomizer that actually is applied at the beginning. But even if you control the random seed, you use the same seed again and again, you will not get the same answer.

It's because there's all kinds of non-determinism in how you know a GPU works, how it's scheduled, how the key value cache works and all these things. What we have done at IGEN is to control this end to end. So you get a fully deterministic AI stack. So that's what AI is fully deterministic AI made by fireable because you have the stake and operator backing it. Soon cloud comprised of layer decentralized web fable network at the base services like IGEN comput AI built on top which are first party services that we've built and there is a thriving ecosystem of AVSS built on top what is it similar to in the cloud you have this uh whole set of applications they called software as a service right software as a service solutions that actually provide you know useful and interesting services whether it's running a database whether it's running an authorization whether it's payments via you know stripe or PayPal and so you have these uh software as a service solutions built on top the same thing on IG cloud is these alsutonomous verifiable services and we have a whole suite of them each of which specialize in some kind of verifiable task so if you look at in the cloud era the software as a service was one of the most profitable sectors We expect in a similar way there will be highly specialized purposeful aliass built on top of and finally applications can mix and match these services and produce highly expressive very useful applications at the end.

Today in crypto people talk a lot about are all the applications going to completely internalize the infrastructure. So if you went rewinded back to 1998 or 2000 on the web applications you will see a similar debate. People will say, "Oh, you know, should every application should run its own servers?" And it turns out that's absolutely not what happened. Netflix, the largest internet consumer, still run on the cloud.

Netflix, the largest internet consumer, still runs on the cloud. It's because it's super hard to build and scale these things correctly. We believe the same thing will be true for crypto. So, we are building for a third thesis of crypto other than the app chain pieces or the big chain pieces. the cloud thesis of crypto.

Okay. So Ian is already the category leader in restaking category leader in DA. We created the category of AVSS and also category there. What we're doing next is launching compute and AI. You know last night uh Singapore time comput and AI launched on mainet alpha which is a first version of our mainet.

And both of these services are now available to be used. So really excited to see what developers are going to cook up. Okay, what is our bet? What are developers actually going to cook up? So if the if the the talk was technical and it's zoned out, it's time to zone back in.

I'm going to talk about applications. What can you do with this? What are the new and exciting things that can be built on top of so first one agentic finance right? So if you look at you know crypto today one of the biggest unlock was the automated market maker unis swap launched with you know where you can just go and provide liquidity passively. We believe the same thing is going to happen but at a much bigger scale because of AI trading agents where you can give money to the agent and the agent actually goes and does actions on your behalf.

So we don't think this is a purely crypto trend. It's going to be a big deal even outside crypto. Why is it if you think of a you know a machine learning model or a chatbot like drop or chat GPD they don't want to necessarily integrate agentic trading desk because this is a huge liability imagine drop goes and buys dogecoin on your behalf and then you may be agrieved that you know doge went down so you have a big problem there like when you're running these agents you need to give different kinds of guarantees our vision is we will guarantee that this particular agent code can only trade operate your money. This is a very different kind of guarantee, but what you're guaranteeing is that this code, this agent code runs and trades your money and so this can actually move liability out of like a you know us or anybody who is building this to say oh you know you know the code you decide whether to use it exactly what Ethereum does for smart contracts. Okay.

That's what we envision and anybody can come provide liquidity or like give the money into a vault and the agent can go and do the trades but end to end the agent is verifiable and unstoppable runs on its own runs forever. So we did a recent partnership with Google cloud to actually accelerate their agentic payments protocol. You know you tell an agent, hey go do these things and there's a manifest and a rule that you can actually communicate to it. But how do you hold it accountable? How do you hold it trustworthy?

That's what you can do on top of a cloud. Okay. So that's agents, right? The other thing, so the easy thing to think about when you think about cloud is if you had infinite scale computation, what could you do, you know, in a verifiable unstoppable way, you know, kitplon gaming was supposed to be a big, you know, um a direction and I don't think it has happened yet, you know, what what was originally envisioned. uh the the one thing that we found actually as a really interesting use case is running gaming tournaments.

So imagine you are a non-crypto game right is basically you know we have this partner called openfront which is a a battle royale game which is not at all built for crypto rails it's actually you know massively multiplayer game 400 players can play this simultaneously and you want to run tournaments on top of this game anybody can come in and put like $100 and say hey I'm participating in this game tournament who wins the tournament is going to get $50,000 and so you want to build these kinds of tournaments You can build the tournament like financial logic on chain but the game itself runs offchain but verifiably. So you get the superpower. In fact it's a new business model for games you know instead of you know just selling subscription or selling game assets you can actually build games with inbuilt markets. So that's another thing we're very excited about you know on on that. Okay the next application social media right?

So you know one of the things uh so this is an altercation between Elon Musk and Sam Alman right they're both you know uh saying that hey you know Twitter is rigged you know it's favoring Elon or you know the app store is rigged it's favoring chat GPT so what do you do can you actually solve this problem by having verifiable feeds verifiable ranking you can actually do this on top of ien cloud any anytime you want to do a computation you want to know that the computation happen exactly as stated you can go launch it on a plug and so we have a launch partner called open rank which is actually building you know verifiable seeds when you look at your seed on you know the app is called cura when you look at your seed on cura you know that this recommendation algorithm ran and produced my results the results were not manipulated there is no reordering there is no you know attention hijacking going on so what else you know agent authentic governance. So when you think about like what's going on in crypto, how do you actually solve governance or you know if somebody comes and contributes to the iggon protocol with new codebase, how do we value and you know give them you know tokens and rewards. How do you do this in you know you can think of it in a big brain open source funding way or you can think of it on on a dgen memecoin way. You have a memecoin and a memecoin wants to say whoever is helping spread the memecoin they should get more tokens. So what you can do is you can actually give write an AI agent and hook the AI agent to have mint rights for your token which means it can create new tokens out of you know your token contract.

It's a highly trusts sensitive position to say, "Oh, this particular agent can mint tokens. It better be verifiable. It better be trustworthy." That's exactly what you can do on ironcloud. You can build extremely interesting new applications which is simply impossible before.

Have an AI agent crawl your Twitter, crawl the GitHub, uh see what contributions you made and then reward you accordingly. Um we had a demo app called VIP score which shows you know how how w you are and in a verifiable manner. Okay. Um finally I think this is the biggest one sovereign agents said you know uh Satya the Microsoft CEO was asked on a podcast do you think AI is a new kind of life is is it in your species and I said no you know it's it doesn't have property cannot own property an AI agent an AI agent cannot take liabilities cannot uh enter into contracts cannot start companies they can't do any of these things because our legal system doesn't allow it to do But this is exactly what blockchains were designed for. Any program can actually own property on top of blockchains.

And with IGN cloud, you can actually make AI agents just like smart contracts. What this does is suddenly you can build highly fully sovereign, highly expressive AI agents that the agent is its own person. It's not working for you or working for me. It's working for itself. So this is a shift in the paradigm and we're very excited to push this frontier ahead.

Very excited to be here. Thank you all for attending my talk.