AntSeed Founder: The AI Inference Boom That Is Set To Explode (Here's Why)

Transcript

All right, you guys. Welcome back to AI Supercycle episode number 40, our premier weekly AI show powered by our friends at Near. Uh visit near.com for one of the best portfolio interfaces in all of crypto, as well as near.ai for their private inference products. Today, we've got Amos Meiri talking about the AI inference explosion. Amos is a long-time crypto builder and investor. Was one of the first builders on Bitcoin with colored coins, and is now the co-founder of a company called Antseed, which is an open-source peer-to-peer marketplace for AI inference. Amos, welcome to the show.

Good to have you, buddy. >> Thank you, Jeff. Thank you for having me, man. Super happy to be here. >> Of course, man. Of course. So, we've kind of seen the iterations of the decentralized AI happen. We saw this iteration of decentralized training, compute, and inference about a year and a half 2 years ago. Then, we kind of ran through this like agentic framework with like these uh different forms of agents on chain, as well as having like X accounts, and and, you know, meme coins associated with them. Then, you know, then we've gone through through GPU-backed stablecoins like USDAI. Um and then now we're you know, we're quite firmly in this open-source AI era, where the uh CEO of Anthropic is saying that open-source AI is progressively moving

in a very scary direction, and is something that the US is also trying to onshore. Through this, there's been a persistent demand for compute across the board. We've seen memory stocks fly. We've seen um chip stocks fly. Semiconductor stocks fly. And through this, there's been a kind of new golden goose, which is inference. People are realizing the value of inference in the AI stack as being the most valuable part of the stack.

And so, in today's conversation, just want to understand what is happening at the at the inference level, why it's so valuable, what Antseed is building, open-source AI, and more. So, maybe Amos, you could just give us a start of what is inference and why is it catching so much fire and traction in the AI world as being the most valuable part of the AI stack? >> Yeah. All right. So, so many things to talk about. But, I'll start first, you know, for you. You mentioned like the crypto and now we devolved and the the terrible meme coin period for crypto, which was very hard for me seeing being in the space for almost 14 years.

Um so, so yeah, I'm very happy to start seeing actually things that make sense in in in crypto. And for me, it was you know, a very long journey. But, I got really excited about AI just I would say a year ago, something like that. And really went deep into understanding where crypto and AI could actually work. There are many places that I think they they won't. But, we can talk about them. Um and and yeah, and like you had a lot of different attempts and in in stuff that were early, you know, like for example, think about Akash and and project like that that were very early to market and

now are booming, which is very very nice to see, you know, people holding for for that long and building something that actually meaningful. And inference is is yeah, I guess one of the hottest topics because it's kind of like the new oil, right? Um There are basically two type of works, I would say, in terms of AI. The one is like the training part and the other part is the inference. The inference to to to very, you know, simplify it, it's like when you're talking with chat GPT, cloud, whatever it is, and that work that the AI the LM is doing is is basically the the inference.

So, you're paying for that service that compute. And I look at inference really like the new all because you know if you look at airline companies if they're we're hedging all prices because that's their biggest expense. What we're going to see is that inference is going to be one of the biggest expenses of the new tech the new company. So as we move forward this market is going to evolve a lot and we'll start seeing you know stuff and I think we're already starting to see that like people looking at hedging inference prices and you know you have

Venice doing also something interesting in this intersection with the DM which we can talk about as well. So yeah I think inference is going to be a big thing something with basically endless demand. If only like 5 or 7% of the population is paying for it I think about what will happen when when will have more and you will have robotics and everything will use compute and inference. >> Yeah and so just a way to kind of classify the value of inference when you're thinking about using these tools when they are like writing back to you like oh we're

thinking that is the inference you're you're producing tokens to get an outcome that like a response to your prompt. And so this the the inference explosion or the bull thesis for inference is the demand for tokens. Is that the right way to think about it? >> Yeah yeah I mean it's basically when people say the demand for AI what they mean is actually the demand for that. So yeah. >> Okay. And right now the token production is you know growing exponentially right like the amount of tokens being used on a day-by-day basis across Frontier Labs is growing exponentially. And there there is kind of starting to be this

financialization of the AI stack, right? Like we've heard about these compute markets being available for data center owners to hedge their inventory or for different AI programs to kind of think about going long or short on the price of compute based on what they're seeing in in their internal data. And so we're seeing like simultaneously the growth of both inference and compute demand, and then also the financialization of these markets to create like the best fair market value. How do you How do you think about what is happening right now with the financialization of the AI stack?

>> I think we're really really just starting right? We're really just starting because, you know, I sometimes it's really funny to me because I look at the the day I space like in general and, you know, you had like GPT and everyone were using GPT like, I don't know, Bitcoin. I really like I can compare it to to blockchains, right? And people think like, yeah, that's going to be like the main thing. And then you you had Claude, you had some others, Gemini, etc., etc. And now what we're seeing is basically a boom of of so many different models and I think this is just going to to increase. We're going to have also sovereign like countries, you know, having their own intelligence that's a

whole new, by the way, competition and and and that's happening behind the scenes for sure, right? Like that that's going to be a very big intelligence war, you know, between countries having their own proprietary intelligence going to be the most lucrative and important thing for a country in my opinion. But anyway, so you know, you have all of these models and endless possibility where inference prices are going to be also pushed down. And in terms of the financial part, it's really really just starting. Yeah, it's it's it's I see some, you know, different applications. There was a big company, I

forgot its name, that kind of like raised now for to invest futures and and some derivatives on top of that. I forgot the name. So, yeah, it is just starting. It's going to be a very interesting space to be in. What is actually also what kind of like turn on um the idea of starting Anseed. Yeah. >> Okay, so maybe we can just get get into then what what Anseed is. Um for someone new to this, you know, what exactly is Anseed and how does it differ from, you know, platforms like Open Router and other kind of similar platforms?

>> Yeah, actually the the the best way to explain it is to compare it because people understand and use Open Router. Open Router is an amazing company. Um and and basically what they're doing is they're connecting between buyers of inference of any type of model that there is to sellers, providers who provide inference, different models, etc. So, it's a marketplace, right? People are looking to buy inference and other people try want to sell inference. And Open Router sits in in the middle, basically.

Curates the provider. You know, it's it's a centralized marketplace that basically control the supply, control the demand, and growing pretty pretty fast. And um Anseed is basically Open Router without Open Router in the middle. Okay? There are there is no gatekeeper. There is no one who decides who is going to provide and what is going to provide. And there's no one who who who is going to um consume that inference. And the way that it works it it it takes a kind of like an old idea which is the torrent network which by the way I think kind of like didn't work. It did work it had millions

of users yeah but because UI was very very important but in the edge of of agents like they don't care about AI but we we can get into it. Anyway so AnSId doesn't have anyone in the middle and the exploration of the network happens on the DHT network which is the same BitTorrent network that you use to download files from one another. It works in the same way and the settlement and the agreements basically and the payment happens on base on on the blockchain.

>> Got it. >> And all of this thing is so decentralized and permissioned that we when we went live just I think almost 3 months ago at the second day we had a provider like we never talked with him. We didn't know like who he is but if someone is just you know read something and connected and started to provide the inference. So it's it's really like a truly decentralized protocol for that. Now already with after only 3 months with more than 1,000 active users more than $100,000 in in in volume and

with a product that we're going to release which we can talk maybe a little bit here it's it's going to to go very very big. Yeah. >> And then almost maybe can just kind of walk us through practically how how like what the user journey looks like on AnSId. So I'm a buyer of inference. I come to the platform and I can use the platform and then I pay the seller in the marketplace. How How How does that work? Um and you know, what does that process actually look like for those who are listening who want to try this out?

What does that look like from the buyer perspective of wanting to buy inference to to use, you know, any LLM? >> Yeah. So, uh basically you can find all of the models basically that exist. There are more than 1,000 different services, AI services. And what you can find there is and and the idea behind um NCD, it's not only raw inference. Like you can have specialized inference, like skilled inference, or you can have even a router that you're paying for a specific router. So, it's any type of uh service. On the user side, the experience right now is a little bit uh still rough. And that's why yeah, I'll talk about how it's going to look like

very very soon. Uh but right now basically you download a um a client. You need to download the end station, which is a chat uh client chat, which is kind of like a node as well behind the scenes. And you interact with the network through that, or you can use a CLI, or you can connect it to your Hermes or your Claude like pretty That's pretty pretty easily. So, that's how it works right now. You know, for someone who's uh a little bit uh technical, I would say that's an easy thing to do. Um the way that it's going to work, it's going to work a little bit like a VPN.

So, you download um um a client, like a very uh small client just like a VPN. And you just click on. That's it. And you're connected to the network, and you can use whatever tool that you already using today. If it's open code, um Claude, well, whatever it is really. Um super super easily. Uh so, the onboarding is going to be just like any non-crypto, non-technical um product. >> Yeah, love it. And then so this inference marketplace that you built is, you know, still in the early days, you said, you know, it's the there's been a hundred thousand dollars of volume, thousand users, it's kind of growing. There's kind of more

marketplace oriented inferences, like oriented marketplaces for inference versus kind of like decentralized providers. How do you see this divide growing in the future? What are the advantages of using something like AnChain.AI versus a centralized inference provider or marketplace? And how do you guys leverage blockchain rails, open source nature to kind of build a more a more dynamic and a better product from this, you know, you know, foundation? And then what is that kind of divide and how do you anticipate these two separate paths playing out for the next 12-18 months?

>> So so the idea is that open market creates competition and competition creates better prices. That that's the whole idea behind this open market. Um and we don't care how people are providing their inference. Of course, as long as they're following terms and conditions and rule and everything, but they can run their own GPUs, they can they have credits to to a model that they want to resell. They can do basically whatever they want on on on on the network and provide inference.

And if their product is good, then people will use it and they will pay for it. The main challenges on on on on the network are basically two. One is, you know, how you make it easy for anyone to use, which we I think figured out very well and we'll release it pretty pretty soon. And the second one is is creating this trust layer because when you have a decentralized network, you know, there are people who are trying to do some uh tricky stuff, and it's not like we're a centralized platform who can do, you know, stuff. We need to cryptographically prove that someone uh saying that he's say is selling um I don't know, GPT or whatever model it is, that it is really what is uh uh

selling. And we're using for that, for example, uh black box API fingerprint, which is it's a way to understand which model uh you're using tea at the stations, uh some uh different things. So, so that's kind of like the main main challenge when when you create something that is uh uh that open. >> All right. So, almost let's get into what our audience is really going to want to hear about, which is this ANTS token. Um there's a 1.04 billion token hard cap, no minting beyond emissions, no admin functions. The current supply is 55 million, uh 5.29% available.

You're on epoch number 11, and there are basically emissions to track eligible real activity on the network. And so, um we've seen the success of VET. We've seen how how strong that token has performed, as well as uh just the actual growth of the inference token production. Excuse me, that they've done on a day-by-day basis. And so, you know, you guys have generated a a significant sum of revenue. You You guys have a pretty good business model. I'd like to hear about that. And then I'd like to hear about how you're thinking about this ANTS token as relates to to the system. Because again, a lot of us uh you know, have been holding VET and been bullish on VET. We had Eric on the

show in February when the thing was trading at like a dollar, and they were doing like maybe 5 million uh you know, tokens per day, or sorry, 5 billion tokens per day. Now they're doing like 80 to 90 billion per day. And so, you know, I think our audience would be very happy if this is a uh you know, VET style outcome. Not, you know, no financial advice here. What is the uh the An seed business model? How do you guys generate revenue? And then, what is the plan for the Ants token and how can we get involved there? >> Yeah. All right. So, it's the first disclosure. I'm also an investor through Node Capital in Venice and a big fan of what the team is doing and super happy for them today for raising that amount.

Um Okay. So, basically, um we decided to take a path where there is no conflict between a labs company and the network. So, we're not going to have a company, but we're going to have a foundation is going to basically help manage the the network and all of the revenues from the network are going back into the network itself through a buy and burn. Um So, nothing goes basically to a company. There is no destruction of value. We want to create it in a very, very decentralized way. And that's the only way that I know.

Um and the first and easy fee is the network takes already 4% fees for every transaction or consumption of inference on on the network and that will go to buy and burn. There are other things that are going to happen in terms of fees. There are going to be more fees more fees generated on the network. The token have two layers. The first layer we're going to release in in a couple of weeks and that layer focus the utility of the token is focused on the reputation layer. So, we talked about trust that it's hard to create trust in a decentralized network. So, staking into provider is one way to create that

trust. So, we'll enable staking of the Ants token into providers and the rewards will flow both to users who are using sellers and stakers. We also have another component into the tokenomics, which is verifiers. It's It's people companies who's doing the work to verify the work on the network and the authenticity of the the providers on the network. So, that's the first layer is going to be up and running in the next couple of weeks. And the second one is the interesting one, which I'll talk just a little bit to give you the view. We're looking into building

kind of like a liquid inference token on top of the ANTS token. And the idea is that every time you stake the ANTS into a provider, you're receiving an NFT. That NFT is going to trade and it's going to be some kind of a bond to create that liquid inference token. Um that's the second layer. We're still working on it. It will take us time to to release. But that's kind of like the utility of the ANTS. But in general like we wanted to be able to to do it as like as fair launch as possible. And this is why we didn't pre-mine, you know, we didn't pre-mint, sorry, tokens. Everyone get the opportunity to participate just

like us from day from day one. Everyone get a chance to earn the tokens. Um we self-funded. Uh in the beginning it's only me and Shakhauf, my co-founder. So, we're only two people. We're only now bringing on board another two people to the team. Uh so, yeah, like super excited about everything that that's coming. And um yeah, like try it out. >> Yeah. And then on the revenue side, you mentioned that that you guys are planning on having the revenue be directed to the token. What does that like monthly revenue look like right now?

>> Yeah, right now you just take 4% of the 100,000 that we we generate. >> Yeah. >> But there are going to be other mechanisms. Like for example, if someone is is because you're locking, for example, your ANTS on a provider for a period of time, and this period of time is like the strength of that position. Uh if you want to move it, you will need to pay fees. There are there are some other fees mechanisms, so it's going to grow. We look at it as a this protocol as a business. It's not, you know, it needs to generate money. It needs to create value to the people who participate in it. Uh really like a win-win. And not not just saying a win-win. We're We're trying to create something that's truly decentralized and and a win-win for all

the players. >> Yeah. Okay. And yeah. On the bond token, right, that you mentioned is kind of NFT, in the Venice design, you have this DM token that gives you a guaranteed amount of credits to use on their platform, right? Is Is this bond token built with a similar kind of understand like a similar idea of this kind of guaranteed inference? >> Um it's a little bit more problematic to do when you're in the same model, because it's not one provider just deciding that he's going to provide the inference. You need to basically aligned a lot of different providers into accepting the same inference token. So,

the mechanism is going to be a little bit different, but I think the Venice mechanism is is is is one of the most innovative mechanisms I see in in the past 14 years. And by the way, like we have also the DM pool. >> Yeah. >> Where DM holders are able to pool in into a smart contract, and this smart contract is a provider on ANTSID. So, we have a Venice proxy provider on ANTSID, which is basically people pooling in their unused DM into a smart contract and [snorts] they're receiving basically the usage of the USDC uses, which by the way is growing like crazy on this

provider and and they're receiving also of course answers as a reward for that. So yeah, we're you can use basically vanish right now at 50% discount on AnSId. Yeah. >> So one of the things that you mentioned about the providers right where you stake the token into one of the providers pools. How do you prevent providers from being bad actors with no central control right given that it's like this kind of decentralized system. What what what rules do the providers have to maintain integrity in the AnSId system?

>> They don't have any rules because we can't create those rules. We can't control it. The only thing that we can control is to kind of like create this trust layer and have the wild west right? Like the wild west we we can't control but we can basically help users explore trust and all of these trust layers is not like us saying oh this one is shoots or this one is like you know a non provider. Like we have to cryptographically prove you know that they they had a good up time that they provided the model that they said they are going to provide, you know, that their identity is their identity like all of that.

It's a little bit harder than just building a centralized platform but that's why we're excited about it and I think it could scale. Yeah. >> Yeah and then just thinking about kind of like the market at large. How big is the inference market today? And if we're talking about you know AnSId getting a couple bits of this market to start, how big are we talking in terms of total token production and you know what is the scale of this market that that you're talking about? >> It's it's it's endless. It's I think where we're seeing now only that, even if you look at the biggest player, I think this market is going to explode like in in ways that, you know, there is I guess an AI bubble, okay? It's okay

that there is a bubble because any new type of market like that, but it's also not a bubble, you know? Because the the demand to inferences is just to sorry to intelligence is just it's it's endless. It's it's an endless Yeah, it's an endless. So, yeah, I think it's huge and I think there is a place in the world for for, you know, a permissionless a truly permissionless inference marketplace where everyone, you know, equally can consume or provide. And I think also, you know, looking ahead, one of the interesting things is specialized inference because think about for example designers.

You know, people who used to do UX UI and, you know, websites and stuff like that and they have like really good workflow for that, but they're going to be out of job like most likely, right? In in in a couple of years. Like I think those people are going to develop their own specialized inference where they create their own workflow, you know, connecting to different models to get basically the best output. And in the future, people will not care if it's, you know, Opus 4.8, GPT 5.5. Like, you know, I need now legal advice. I'm going to get the best legal advice, you know, that someone can provide me. So, I think the future is going to look a lot like that. And AnSyt is just, you know,

open to that and we already have some some things like that. >> Last thing for you from my side, Amos, what is the one big why for you for building AnSyt and what's your boldest prediction for how these open-source PDP AI marketplaces will evolve in the coming years. >> So, for me it's you know, I've I've been to crypto for a very long time. Very exciting early days of Bitcoin, was working with people like Vitalik on colored coins before starting Ethereum and all of that. So, so seen a lot of in the space and and I came there from really an emotional and and that belief, you know, in a new system and everything. And as time passed, you know, things like meme coins, like

sketchy token models, all that like really destroyed everything. I hate it. I hate what pump fund like did to to to to our crypto market. I just I I really I truly hate it. And And for me it's like I invest in in in crypto companies through Node Capital. We run three funds and I felt like, you know, this older generation which seen a lot of things and have the experience and everything need to step in again to start building stuff that are meaningful for crypto.

And I felt like the intersection between AI and crypto is the right place for me. So, that's why. >> it. Amos, thank you for your time and thank you for being part of the AI Supercycle episode number 40 here powered by our friends at NEAR. Amos is building an seed guys. You guys can check out an seed and their website. It is at anseed.com. You can learn about the an seed token and more and everything that is going on with the open-source inference marketplace that that you guys are building. Wishing you guys luck as you scale and come into this launch and appreciate your time Amos.

>> Thank you so much man. Bye-bye. >> Cheers buddy.