Illia Polosukhin: What Happens When Your AI Actually Belongs To You?

Transcript

The curator model is good when you have kind of way to access to you know various different assets and then that might be even offchain right it like this agent can actually have a legal representation. >> Do you think that you can kind of push the needle towards [music] this movement happening more broadly within like the big players? >> Have you seen a standard that has really stuck out to you and says yeah this is something that you know we know is going to stick around. We're going to build around this particular standard. Welcome back to the RUP. >> This is your home to learn from the biggest guests, the brightest minds at the tip of innovation.

>> Our goal is to help spread the collective knowledge of everyone in our community. >> We are breaking crypto out of its echo chamber and taking it to the real world. >> Sit back, relax, and enjoy it. [music] Welcome back to AI Supercycle, our premier AI show airing [music] every single week presented by Near. We cover the ins and outs of decentralized AI, privacy, and the future of this massive technology. Near is the blockchain for AI and the execution of AI native apps.

You can check out Near's latest AI product at near.ai. Sit back, relax, and enjoy the show. Guys, welcome to AI Supercycle episode 12. We've got the Black Dragon calling in from the layer here, the private layer himself, which is a undisclosed location because we're all about privacy here in this world. Near AI has just dropped today. Ilia, the marketing is amazing. Can we talk somewhere private? I hate to tell you, Ilia, this is not a private call. We are public. We are live right now. Good to have you, man.

>> Yeah. I know we've been cooking so it's great to be here to talk about it. >> Yeah, for sure. U I mean just some just some context right Rob last six weeks I mean we saw Zcash went on this insane run privacy really came back into the zeitgeist of kind of the uh general kind of crypto Twitter population and just more broadly uh kind of like this return to fundamentals ethos that we've been pushing on the show. near intense facilitated a ton of the Zcash bridging uh in in in terms of actually getting assets from Bitcoin, ETH, Salana into Zcash to be able to actually buy Zcash.

Uh the volumes went through the roof revenues as well. And then you guys had been working on this near AI chat product, this private chat product. So it feels like all of the perfect time for Near. Um man, how has it been the last two months at Near? Ilia, it feels like a lot of things are coming together. >> Yeah, exactly. It feels like a lot of stuff we've been building for past like year, year and a half is actually like all coming together and really like starts to gel. Uh so really excited there. There's actually more things coming in next, you know, few few days to a week. Uh I'm not I'm not going to drop too much alpha, but uh a lot of exciting stuff there. But yeah, I mean it's it's just like teams are jelling.

we have you know intenser firing also on there's uh I think momentum is like insane it's it's like there's a PMF people are willing to pay for it and it's getting integrated everywhere and kind of really like the vision of like hey this is going to unify liquidity plug everything together is is really there and so now it's like hey what else can we build build into this right and how else it can leverage now AI and you know real world assets and other things as well. And then on privacy as you said, I mean I feel like in in the crypto circles, privacy always been like an important aspect, but uh generally I kind of like not been

talked about and so I've been talking about like the importance for AI for your data like you cannot own your data if it's not private. like if if you're giving it to you know third parties to Google's to apples to open ais like you're not owning it right you actually sign off the terms of service uh in many cases Google I think just released that the training or or it was reported that the training on your Gmails >> uh you can pro you should probably go and opt out from that you need to opt out explicitly >> um so that's what's happening right like all your all your data that you given is effectively been you know used in some way as companies. So privacy is paramount and kind of money is one you

know one aspect of that uh data AI is you know pretty big fundamental another >> yeah we were just talking about this Gmail Google training model and it came up in the context Andy and I were both looking at the same email we were thinking about a response and it was like wait there's a one-click response right here and it was not just the canned responses that they've had for a long time it's like wait this response is actually pretty good that was the surface level the deeper level is The reason why it's better than it always has been is because they've been training on the millions of emails that we've had in our database for going going back that long.

>> Yeah. >> Pair that with what we're seeing from ChateBT. They're training not only on our data, but then they're going to start serving ads based on those prompts, based on those responses and embed those directly into the responses that we see from ChatgBT. All of this was helpful to ultimately provide the value proposition of a private AI and you guys have capitalized on that near AI. Talk to it like your doctor, you know, talk to it like your lawyer. And >> the the litmus test for this has always been look, you know, chatbt has said we're going to share, you know, if the authorities come to us, we're going to share the data with the authorities.

Yeah. You know, the idea here is even if you guys wanted to, and I I I'll frame this as a question, but let's say an authority came to you and they said, "Hey, this person, we know they've been using near AI. We want we we're going to serve a subpoena or we we want access to this person's records so that we can understand how their criminal mind was working." Would would that be an instance where you could give up that data? Would you give up that data? So for near AI for this product we cannot like we don't have any access um the all the AI inferences right they all encrypted with the specific GPU hardware uh where that runs and um yeah I mean like effectively even the like even with

access to hardware you cannot decrypt it and it's all kind of in the moment. Yeah, I mean I think like like the like like the context here, the general trend is that the people have been far more willing to provide different AI uh services with more and more information. Like as you get used to using the product, you become more and more kind of acclimated to like actually uh you know giving this product a ton of information. Um and and so you know today you guys uh launched AI cloud and private chat. Um and so these are kind of like two separate things that work together within this lens of understanding the this this kind of growing human behavioral trend of of

using LLM in these more like uh intimate ways. What is AI cloud private chat? How do they come together in the lens of like this growing user adoption of AI? >> Yeah. So AI cloud is really the developer tool, right? It's an infrastructure that any developer can use to leverage private inference, private uh and verifiable inference and we're going to be adding more things to that product offering. And so uh for example, Brave has integrated neari cloud into the Brave browser. It's right now in nightly uh and uh effectively when you talk you can choose near AI

provided u in in your Leo AI in brave. Uh we have uh open mind which is a robotics company that building a robot operating system. As you can imagine having a physical robot at your home that can that is effectively listening at all time looking at all time what's happening. You really care about privacy at that point right probably don't want to you know you and your boxers uh continuously streamed to some third parties. So that's a really great use case. anything effectly touching sensitive information but also just like a day-to-day like live mic uh should probably be private and so um so that that's uh for AI cloud and so AI chat is really just a pro private chat is

product built on top of this right so this is just our demonstration of how you can leverage the AI cloud to build products and have all of the privacy features um and so I mean obviously it's open source person so people can uh build on top of that or uh we expect you know there's a lot of kind of ways to expand this now to add more functionality with now private MCP servers with memory etc. So we want to explore that as well. Uh imagine now you are for example right now you know OpenAI is like hey allow us to access your email right your your AI will be better but now OpenAI has literally you know access to all your emails and even

non-maliciously right there's like accidentally issues can happen with this but it also includ if if there's any kind of leak or anything all the data will um you know potentially versus in our case you can actually run the MCP server itself in the same kind of verifiable private infrastructure and so now your AI can have access to it and only it and nobody else can you know get into that as well. Yeah. So that's the idea like we can actually keep expanding and adding more privac like privacy verifiable services on top of this and really make your AI also smarter and better have more context be able to actually do actions on your behalf be able to manage private keys as well

right because you know this is something where you actually care about privacy and verifiability as well. >> Yeah. So, you're wearing your uh Jensen Hang um leather jacket today and uh part of this uh near AI cloud is using Intel TDX uh uh hardware as well as Nvidia uh confidential computing hardware where data is processed in this kind of sealed isolated environment. Um so that's kind of the AI cloud. I'm curious like what from your perspective using the this you know high level commercial grade hardware what is the biggest unlock from uh combining actual privacy at the hardware level with the onchain guarantees that you get using the the

private chat product what is the what is the unlock there that is uh novel about this launch >> yeah so the I mean first like effective this confession computing hardware mode is only like been around let's say a year. Uh some of the recent like updates that were needed only came out at Nvidia GTC this year. So all of this is very kind of new and and the reason why we haven't seen this as much. uh but this layer of hardware allows you to effectively run computing like you know AI model inference agents etc in a way where you know that nobody was able to

access what happening inside uh even if you know somebody's running out of computer like they they cannot actually access through uh just like inspecting processes or whatever what's happening there now that on itself is only a part of the story right I'm as a user sitting you know on the other side of the planet how do I know that computer is running how do I find it and how then how do I encrypt my data for only being able to decrypt it inside there and then how do I manage my context my you know my all my history this is where our uh kind of near chain signatures uh which is effectively a multi-party computation system comes in allow you to effectively manage like do the key management

provisioning all of those things in a in a trustless way as well so it really combines ing all of the pieces both hardware security and cryptographic security. Uh you know as like we allow more and more parties to join uh you know the payment for the hardware providers will be done on the crypto rail. So really combining all the pieces together. >> Havachi is fast and private [music] per trading secured by ZK with Celestia underneath. Go to habachi.xyz to start trading today. Not your keys, not your coins. From the team that pioneered cold storage, Treasure has just released its new wallet. Guys, if you are still using hot wallets in 2025, 2026, you are missing the entire point. Secure your

coins. [music] Get a Treasure at treasure.io. Holiday is unified crypto payments. Give your users more wherever they are. Go to holiday.xyz to never write a smart contract again. >> Yeah. And it and it does feel like there's a lot of things that Near has been building for a long time that are starting to come together. Another pretty significant piece of this is 7 billion dollars that has flown uh uh moved through Near Intense since you guys have launched that that helped unlock this whole Zcash movement. Um give us a sense of how these things ultimately come together as you're thinking forward. I also saw that you know just recently you put out sort of what what you're aiming is that soon private AI chat will have near intents

integrated this will allow for some interactions across web 3 but then you also said web two coming soon could you give us a sense as as these things start you know continue to develop >> that alpha that that that was previously mentioned >> so so yeah can you give us an idea of how these what this might look like as as these pieces continue to develop on on own and then ultimately come together. >> Well, I'm laughing because this isly a tweet from 10 minutes ago. So, uh, >> we're live, baby. >> We do our We do our research >> real time [laughter] or or you're reading Twitter while while interviewing previous guests. [laughter] >> Uh, no, I mean, so so here's like I mean, maybe like couple couple year

vision, right? And then we can work backward. Couple year vision. I think more and more of our interaction with computers, with phones, with our devices is going to be AI facilitated, right? Because AI will know so much about you like you won't need, you know, go and clicks for stuff. You won't need to go like Uber, you know, enter an address or copy paste it from the calendar or whatever, right? Like all of that was just automatically done. Like it knows you have a meeting, it'll get your car. And so how does it work? Well, your AI is, you know, the brain obviously needs to be private. Needs to have all your context privately, needs to be able to predict and reason and do all those things. This is why, by the way, the just local doesn't work, right? You need

this background intelligence that's running there that's able to consume all those information. That's why we need this kind of decentralized confidential cloud. Now, whenever it knows what you need, right? Let's say you need a taxi to airport or you you know you want to get some food that's healthy and you know will make sure you have energy for the next meeting. it it will effectively issue an intent saying hey like you know I'm looking for this kind of food you get different providers it sc you know evaluates them chooses the right one and then does a transaction right all of that is happening on a you know potentially in the background or maybe comes back to you and says like hey confirm this exact interaction but intense is effectively as a language that AI is talking to other AIS or

service providers to really get to the the agreement and execution across the internet, right? And you can imagine this not just for consumer use cases but also for you know real commerce businesses where you know hey we need to deliver steel you know like the solver comes in and says cool I will be able to you know procure it from a mine get the shipping containers get the port you know allocated etc of our houses rented and insurance and I will get a loan for the float of the money while while this is in transit right like all of this package boom you get like a full package you know five like RWA effectially packaged into single NFT that you can deal with, right? So that's

kind of where all of this is going >> and so intense really right now we're proving it out on the simplest thing which is crypto. [laughter] >> Crypto is simplest because it's the simplest to verify, right? It uh we have all the uh kind of easy tools to verify it. We have chain signatures to really manage assets across all these chains. This is why like indeed with Zcash it was kind of really powerful because it was very easy to kind of integrate and and and start get it going and make sure it's trustless and now um going forward yeah [snorts] like intense starting to really connect into web two service providers right be travel uh but actually like also just any API

right like hey you know my AI needs weather report my AI needs this and that. Right now, if you're trying to do this, I was actually trying to do something and I needed like a price in my Google spreadsheets and I need like go sign up for another API provider, get the API key. It's like super complicated right? >> Like my AI should just be able to figure it out, not needing to register for so intend AI kind of and they all work together. agents effectively are solvers on the back end and we already have some of this examples where some agents are effectively running and solving intents on the financial side but you can see this expanding to more and more use cases.

>> Yeah. And I will will your AI travel with you like when you're filling in let's say it's a Google sheet is that going to be your AI that you're kind of bringing from service to service traveling whether it's on a local device or it's kind of omniresent across all of your devices or is that Google's AI right that is filling in the Google sheet it's going to go have to communicate with other agents. How do you see this this interoperability working? Yeah, I actually think it's going to you're going to have like a one uh AI that's yours that like similar how wallets work like I actually think web 3 kind of run this shift like traditionally you have you know a service providing you software you come in you register and you use it web 3 said no no no actually you come with

identity and developers push software to you right you know so smart contract is kind of lives there and you can call it >> uh you know front ends in theory can actually just be like fetched and and and run on your side. And so I think s same idea happening here and we'll see this already with like again kind of how we imagine ini cloud right you can actually have developers build services MCP servers or agents that run on users contexts and information inside this private enclaves right and so this way as a as a user you don't need to share information with third party with developer right their stuff comes to you right this can be just AI model or it

can be a full AI like server and agents. And so then your AI effectively travels with you. It has all the context. It knows everything about you and it's go it's able to still communicate with other people's AI, other companies AI or some other autonomous agents in this intent architecture where affected some kind of transaction some kind of counterparty agreement. >> Got it. Uh and you're on mute. >> Yeah. If we think about uh wallets in as a as a similar analogy to like how we'll use AI in the future kind of across multiple applications, multiple websites and things when when it comes to how we're you know generating actual uh like yield using agents I think you guys in

introduced this or near 20 introduced this agentic yield concept um that I think is relevant to the private chat as well as relevant to kind of like what Rob's getting at here. Um, how how do you see agents then reshaping yield across DeFi over the next year or so? You know, whether it's I want to use near private chat to, you know, go and do strategies on chain or if it's just embedded into my wallet. Within the lens of that previous point, how do you kind of see this agentic yield and this uh AI agent yield kind of uh concept uh coming about?

>> Yeah, I think I mean it can come from both directions, right? It can [clears throat] be that you instruct your AI to effectively like hey every hour go check you know what's the best rates where the best uh things are given my risk tolerance and you know rebalance and reallocate and and so that you can actually again nearance is a great platform to do the execution of that exact system and we have you know something interesting coming on that side. Now the other side that um is similar is you know we've seen this with Morpho right there has been a lot of like kind of curators which may not been managing

risks very well and it's really just opaque from a blockchain perspective how it actually working and so you can imagine actually curators being AI agents autonomous agents and so they're very transpar you know you know exactly what they are right there's you know the prompt and the data feeds all verified auditable but they run inside you know neari cloud so they're verifiable and private so because you know you don't want to reveal the decision that they make when they allocate things so this was the example we had launched called mind share agent which was effectively using you know analyzing Twitter sentiment about different assets and rebalancing uh portfolio around that so you can have kind of this coming from

both directions and it will depend really like the the curator model is good when you have kind of way to access to you know various different assets and then that might be even offchain right it like this agent can actually have a legal representation and it can go and actually get you know like real credit from bank or whatever or you know invoices this travel insurance etc like all of those things like can be packaged and then provided as a yield or >> your AI agent which is more about being personal about understanding your needs like hey you know I'm spending this much per month so I need to like withdraw that uh to pay my bills like all of

those management like your AI agent will your AI assistant effectively will be doing all of those actions and kind of reallocating between different curators and and different assets. >> Yeah. Okay then. So that's kind of like the DeFi crypto native version of this. When we zoom out to, you know, because I know you're extremely well connected in kind of like the traditional AI space as as well. So I'm sure you've, you know, talked to a lot of your friends about this product launch and now that it's out there. Um, so kind of zooming out from like the the web 3 world, but into the more kind of like broad space here.

Um, I guess very bluntly like does anybody care about this private AI stuff outside of like us like cipher punk kind of people? Like how has like Silicon Valley San Franand type people reacted to this? Do you think that you can kind of push the needle towards this movement happening more broadly within like the big players? You know, just let's kind of have have a more holistic uh conversation about the the state of affairs um outside of crypto. Yeah. So great question and and this is why we you know we launched with few different like non-cryp non-pure crypto or non-crypto at all uh clients because that was kind of we wanted to validate this um and as I mentioned I think

there's few aspects where this is extremely needed right so robotics is one legal is another one because they literally have the confidential and privileged >> y >> that they need to deal with and right now you know some of them just like set up their like local models and do stuff on on their local computer which you know the model quality is really bad. Um and also you know you cannot share like there's a lot of challenges with that finance like non-crypto but actual finance as well they have a lot of like [snorts] actual limitations on how the customer data can be moved around. Um so again like if you want to do something something more than just like ask basic questions to AI but actually analyze the

internal data that that's uh highly valuable and then interestingly governments every every government outside of US wants to make sure they don't depend on just us for the AI and even between departments in a government imagine there's a government department that's IT and government department that deals for example like police right well police doesn't want all of their information be available to some employee of the IT department government like they have no idea who they are right so like in general it's extremely valuable for kind of enterprise government use cases and then for users I think it's a it's a nice to have for

sure it's not like hey we're gonna you know we're going to change our behavior per se uh except Except for you know the slice of users who are on signal th those kind of people uh which you know we are [laughter] um and uh and then that's why we're partnering with Brave we're partnering with like applications who really bringing this kind of experiences forward now medium-term it's about building better product as I said to me privacy is enabler where we now can kind of allow users to connect all kinds of data all kinds of access. You know, again, you like you as a company have internal

databases that you individually have access, but you probably want going to give it to a third party software. [snorts] Uh and so that that is, you know, really where the opportunity is like, hey, look, this is as if you were running stuff locally. You can now give access to, you know, tools like to databases, to your CRM, to your emails, to all of this information. Now, your AI is way smarter. can do a lot of actions on your behalf. You can like set up background, you know, update processes, etc. Which again, for example, we're using CRM. I don't trust that CRM provider with my email. Like my email is too sensitive, right, as as a co foundation. And so like I'm missing on all the features of

the CRM being able to track my meetings and stuff for example because of this. And so like you kind of need this type of tech to really kind of integrate all of the different aspects of your uh of your life really uh and and kind of trust with you know there's medical data, finance, you know your crypto balances etc. Like all of those can be integrated in one. >> Yeah. >> Yeah. For this to really work that that's how it has to happen. And one of the foundational pieces of this technology stack, you know, obviously there's the privacy point, but there's also there's also this model that is is feeding the responses and doing a lot of the work underneath the hood. And I'm curious how this works when you're

training a model that is still private like uh the near AI private chat uses zyorg. This is different from like Llama and Meta and GBT and like all of these other models that are you know on the on the more centralized uh uh uh you know platforms. Where did this model come from? Could you give us just a little bit of history about what is what makes this model special? Why you guys chose to use it for near AI and how it helps enable privacy?

>> Yeah. So we actually have uh well we have four models available in the chat. You can use deepseek 3.1 Quen GLM and uh GPOSS which is openAI's open source model. So you actually have an options and the idea is to be able to support um more and more over time including we want to support third parties being able to upload their models. So if for example OpenAI entropic want to upload their model we like we guarantee that it's not going to get decrypted right it's going to always stay encrypted at rest and then only gets decrypted inside secure enclave and so there's an idea actually that this is a better way to monetize maybe not for those big players

but for smaller players who may not have enough capacity to do in like to provide inference services and to deal with users data and do all the things like this is actually a way for them to serve models to users. Um the reason why we picked this models I mean this is actually you know set of best open weight models right now in the market and we've been kind of updating as as things change the yeah generally we just we use open weight models uh we're working on some improvements around like tool calling and a few other things to uh but right now relying definitely on the open open source community.

>> Cool. I mean in that in that lens Ilio if you're talking about if you're talking about a better product right not just the private aspect but a better product there are tons of uh kind of rappers where you can choose whatever model that you want to like or you can get the result from three models at the same time if from the better product lens to get this thing adopted outside of you know let's say the signal type people or the or the CEOs of foundations like what is required for you guys to to to do that like like what is on your product roadmap or what are you thinking about from the product perspective to make it a better product because there are just kind of like these model aggregators that exist. They may not be private. So you've got a private model aggregator

here but what is what is required to really make this better than uh what's already out there. >> Yeah. So we have few hypothesis and we're testing them right now. So that that that's an important part part here is like you know a lot of it comes from the users and you know we can only hypotheize but you know personally I think it's a combination of improving memory and context management. So right now outside of for I mean probably only open AI honestly everything else has you know very forgetful model it it doesn't really know much about you open done pretty good job at like really extracting a lot of the context background context and then providing it

to the next messages. So I think that's a big part and again the benefit here is like we can connect to more data sources and so we can actually get a lot more context and keep it private for you. Now that goes into the second part which is really enabling to connect to more tools to more MCP servers to more uh kind of external information and so that's uh really about kind of um I mean MCB server support um which which would in turn enable a lot of things like actually using crypto. So for example, we do have near intent MCP server and so your AI can actually start doing those actions. uh and there's like few other

interesting product kind of ideas we have around how to display the out outputs etc to really enable um kind of better like I think in a lot of cases the people use text as an output which is not always the best when you actually going like you know for example trading a lot of people like oh I want to say you know buy me $100 but in reality nobody really wants that you want the trading UI right and so you kind of want like you maybe want to start with like hey I'm I'm in the trading mode but then you want you want actual like UIs ch like charts etc. So there's like a good mix and match of you know UI and kind of text and tools that needs to be working

together. And again this is where potenti like relying on open source community to be able to build those uh you know let's say skills or plugins uh and really surface them uh would be very valuable. >> Looking for yield that stands out? Infinifi gives you [music] just that. Deposit stable coins and earn far more than the current savings rate. Infinifi is a battle tested, reliable DeFi platform that is ready to go. Head to infinifi.xyz. >> Yeet is built for big moments. The most fun you'll ever have with your crypto.

Go to yeet.com to yeet into your next game. >> Trade on anything, economics, weather, [music] sports, crypto, and more. All on couch.com 247 for the world's biggest prediction market. And and so the idea is that you kind of open up this uh this platform almost make it permissionless. Maybe there's some gatekeeping that that you guys can do initially. Uh but you allow this to be a platform where you can encrypt these other models and then you can help people route and move between these things. Um we've heard a lot about X42 and MCP and these different models that or these different standards that can help connect models together. ultimately use a lot of these things at our fingertips to but use it

more efficiently and seamlessly coordinate across these different things. How much weight do you put on these? Have you seen a standard that has really stuck out to you and says, "Yeah, this is something that you know we know is going to stick around. We're going to build around this particular standard. You guys creating your own standards? What do you what do you think is is the trajectory for these things?" >> Yeah. So, I think it's it's important to understand that yeah, there's very different levels of the stack, right? MCP server is really just an API kind of standard for for AI to call. I mean AI can figure out without this honestly but it it it did become kind of standard that so now the in in AI the thing is

like there's actually a really interesting refer referential loop because as something more people use it it's more in the training data it's more in training data it actually just becomes like AI gets better at using it so so that that's probably like for MCP already happening uh X42 I think the 402 itself itself is a very useful primitive right so 402 is the HTTP code which was not used historically it was there for like last 20 years and now we say like hey let's use 402 so that that was that's the conceptual idea [snorts] uh we were part of the X42 launch uh and kind of as near intense as well we have we have a demo

how it works um I think there's like more work needed to really improve this uh but I think direct actually like the 402 as a concept where effectively in the HTTP request you can establish a session and pay for it versus go and sign up for API key separately like that that's extremely valuable and so I think that that going to be a part and then the intense is again flipping the script where instead of saying hey I need I need to call this API it's calling hey I need this outcome I need weather I need weather for Lisbon for tomorrow and somebody says like cool I'll provide it to you for whatever a microscent so that's where 42 can be used and so

intense plus for for two plus MCP all of the things really working together that's kind of where we're going uh with this >> uh Ilia as we kind of come to a close here thinking about um the the kind of effect of of where we're headed with uh private AI privacy as a concept near the protocol you guys have had some also interesting protocol updates. Uh things like you know cutting inflation by half um these types of of general ideas. There was this uh there was this L1 debate yesterday with Hib and Santiago and talking about what you know the valuations are and people are going to be wondering how a lot of these products that you've developed in including the

intense product kind of retie in back to near the blockchain as well as near the asset. uh for the community uh out listening today kind of how are you thinking about uh 2026 as a year of kind of of reigniting uh near the the chain near the asset and kind of participating in this more pragmatic uh crypto environment let's say that we found ourselves in the return to fundamentals environment here um you know what what do you have to say to the community who have been alongside NE for so long and uh you know we can kind of come to a close here Yeah, I think uh I mean obviously at the end it's all supply demand, right? And so the indeed you

know we had a half halfing of inflation and so reducing reducing the upcoming supply future supply and then all this efforts one way or another driving more demand. um you know think of it as like the the decentralized confessionial cloud itself is going to use near as kind of a fee structure underneath it is a you know profitable business with you know 20 30 40% margin depending on like efficiency of the inference and so you know that kind of generates surplus of near uh from what it provides to validators the similarly with intense right there's you know actual revenue being generated

Um it's a it's more complicated because there is actually because the way the different UIs use it um it's not as as kind of visible right now like we we we show the total fees paid by the users but the actual revenue like we're working on how to display it properly because uh there's some agreements with some of the front ends because they like they want to charge a fee and so like there's a little bit of just like um structure but I think one thing I want to mention given your uh always return to fundamentals there is actually an interesting comparison let's say somebody you know let's say

there's whatever hundred parties that pay $10 million in fees you this is one model and you take this fees and you buy back the token and you know you burn it or you just keep it or You can imagine the same parties actually just end up buying the token directly with instead of paying the fees and you know maybe lock it for year, two years, three years etc. Economically the same amount of near or whatever a token was taken out from supply for at least some period of time but now you have this 100 parties super aligned with near right and their fees are zero. So from a competitive perspective, they're actually going to

choose this product versus another product which will have non-zero fees. >> So we're going to we're going to force token we're going to force people to be token holders as a result of using this awesome product. >> This actually sounds a lot like AI companies in the mag like these larger AI [laughter] companies that are trading token stocks for you know compute. >> Yeah. Yeah, but but I think I mean there there is something about that where >> like there is a value in having your customers to be token holders instead and and instead of them just directly paying and so one of the things we're working on is is proposing some of the dynamics around that for near token

where yeah we want we want more near token holders who are getting value out of all of the products we're building right I mean things simp like simple if you stake some amount of near you have you know priv private AI inference uh you know up to some limit for free >> like those kind of things now it's like hey there's a value in holding near and then you know maybe your fees are reduced when you're trading so like you're actually getting more and more benefits uh from doing that >> yeah last thing for you IA from me we actually had a really good conversation yesterday with S turnelo the CEO of sovereign who is uh building out the uh current near uh treasury company I think the largest or the or the only And basically uh in that conversation we

spoke about how uh because of the acquisition of of such a a sizable quantity of the supply of a given asset uh you know in this case he was speaking about up to 10% of the total near uh supply you become a large you know shareholder stakeholder and you know contributor to the network almost like a separate labs corp. We were speaking about how there's this labs and the foundation because of the regulatory uh you know body and now you've kind of got like like these third entities who are like Tom Lee now owns 4% of something or of ETH and um you know now Cell is going to try to get up to 10% of Near. How do you think about uh this dynamic when it comes to these treasury companies when it comes to them owning you know a a

substantial portion of these networks um and then hearing their opinions acting on them keeping them at you know as happy as possible etc h how do you think about this either with near and sell or just you know at at large >> so I actually think this this is something yeah I've been thinking a lot about when this stats starting to um come out and I think there that is actually the most aligned third party to the ecosystem right it's actually more aligned than even foundation from like a structural perspective because it effectively means that their market cap is directly

attached to the value of the token and they can go in and like their fiduciary duty is to maximize their market cap right so like foundations for example don't have that as fidure doing right I mean obviously like some foundations are very much looking enough to do that but regulatory environment etc it's been hard to even go and say that but the DATs actually don't have that challenge they are very much directly like hey we're public traded security our market cap directly depends on market cap of the underlying token and our fiduciary duty is to maximize this market cap and and usually the executives are incented directly with the equity of the company.

So, uh, now the challenge of that has, I mean, been that they mostly been just like a kind of ho hold and maybe stake type uh, vehicle. And so what you know with S the the exciting thing is really like hey we have you know a really executive team that can actually build interesting businesses in this that you know generate surplus that they can reinvest into buying more near or they can reinvest into providing more GPU compute to the network which in turn will get them more near right because GPU compute gets paid in near and so so the idea is like hey look this actually value additive to the ecosystem they can

build a business but this business is like normally if you build a business your goal is to maximize the business so you're going to go and do other stuff other ecosystems whatever here it's very directed like hey but our at the end goal is to maximize value of the token and so from this perspective yeah it's actually a better structure than even foundations uh be and so yeah it's pretty exciting and I'm I'm excited to working with s on on keep growing near >> yeah Yeah, S was great. Uh, one of the more active approaches, more intentional of driving value back to the token holder because of the incentive alignment. Like you said, >> it sounds like you're in a crime scene over there, I [snorts] [laughter]

looking for you. >> That's because I talked about maximizing value of the token like immediately. [laughter] Oh man, Ilia man, thank you for uh spending some time with us today. Have a great evening and congrats on the big launch. Uh it's been awesome working together, seeing you guys grow. It feels like this has been the perfect time for not only this show, but for all the pieces that you guys have been putting together from Near's perspective. So uh props to you guys and uh return to fundamentals. Thanks for having me and everybody should check out the private near the eye.