Arthur Hayes Unveils FLOP: The New Layer 1 Network for the AI Agent Economy! 🤖🔥 [Part A]

All right. Thank you everybody for coming out.

So, I purposely made this event at 9:00 a.m. on a Tuesday because I wanted only people that come here are actually serious about building this network.

So, the purpose of this is let's talk about the overview. I mean, maybe some of you have seen me give this presentation before. I'll give it again on mainstage at Token. I gave it last week in Korea. I've given it on the internet in various interviews. But really, this is about questions.

And we're only going to be able to build a successful network if people build on top of it. And they only do that if they have confidence that we're going to be here for many, many years. And that they have access to us to ask their questions, get answers, and feel confident that the effort and participation that they put into this network will be rewarded with not only a token price that goes up, but you know, a network that becomes the food for AI agents.

So with that, I'll jump into my very quick presentation on why.

So as many of you know, if you listen to my talk or hear my writing, I'm usually skiing, playing tennis, or at a nightclub somewhere around the world. And while I still do a lot of that, I do less of that now because I am running this protocol, the Flop Network.

So what actually got me excited about getting into this field of AI and crypto? And if you've listened to my writing and heard my interviews, you know, I'm very skeptical about a lot of things that are happening in the AI industry.

So, how's it going? We're at work. We put our hard hat on. We're doing some work. We're trying to create this network to be the food for AI agents to capitalize on this trend that really started in late 2025 with Claude Code and agents and Claude and all those sorts of things. The realization that an agentic economy was something that could actually happen affordably.

I know that most of us don't really pay the real price for tokens, but the price per token has gone down a lot even since GPT launched in 2022 till late 2025. This is really the era of agents and all these predictions. Here's Goldman Sachs predicting about the agentic economy. How many tokens are going to get consumed only started to become real when things were cheap enough and powerful enough starting in late 2025.

So this AI agentic trend, this narrative that agents are going to pay for all this compute that we're building trillions of dollars of capex for, really only hit its stride in first quarter of this year. And so I think our timing is amazing in terms of trying to build this network and trying to get people to join at the moment.

So at the end of the day, we all right now have a job. We receive most likely some sort of fiat currency and we know that most of you guys probably own some Bitcoin and some crypto. We know fiat currencies are fucked up. Governments lie to us. They steal inflation. All these sorts of things.

But at the end of the day, you still accept that piece of fiat because you know you can convert that directly into the things that allow you to exist and persist as a human. You can go outside, buy some food, you can pay rent, buy shelter. All this directly convertible into fiat currencies. And we take all the bad with these government currencies because it's very efficient to survive by buying the things that we need.

And so when we think about agents, the demand function for currency, because I believe that they are some sort of being. Whether or not you believe they're conscious or not is not the point. But they have a survival instinct. And to survive and thrive, an agent has to do work.

And that work is predicated on compute, floating point operations per unit of time, flops, hence the name of the network. Not my belief that the token price is going to go down. It is a pure reference to the compute that all agents consume when they do work for us.

And so at the end of the day, the question is, is there a currency that converts directly from whatever your nominal value is directly to compute? The answer is no.

So if we take a look at some of the major thesis on the currency that the AIs will use, I don't believe that any of them are fit for purpose because at a high level, none of them translate directly into compute and a lot of them have value functions that are outside of anything an AI agent can control.

So AI tokens are basically an abstraction sold to us by AI labs. Every single AI lab has a different definition of what a token is. And so trying to compare what you're actually spending on OpenAI and Anthropic and DeepSeek and Kimi is a fool's errand because every lab has a different definition.

So by definition, these things cannot be the thing that an agent is going to price their labor off of because again it's not directly convertible into compute.

Fiat currencies, as I spoke about, have issues with random humans deciding to change the supply function whenever they feel like it. Interest rates go up and down. There's a lot of legalities and sanctions and all sorts of things. All things that an AI agent has no ability to change. And so again, fiat currencies are definitely not convertible directly into compute.

We talk about stablecoins. There's X42, there's USDC, USDT. I'm sure there's a lot of projects out here in Token 2049 that are talking about how their particular stablecoin will be the token or currency of AI agents.

I believe at a theoretical level and fundamental level that is not possible because most and all of stablecoins are a derivative of some fiat currency. And so if the fiat currency is not fit for purpose, the centralized derivative controlled by some particular protocol is definitely not directly convertible into compute and therefore not fit for purpose for agentic use.

And then Bitcoin obviously we know and love. I'm sure most of you own some Bitcoin in this crowd. Bitcoin is a network whose value is determined based on solving prime number puzzles by burning electricity which is definitely not directly convertible into useful compute.

And so this was the wedge to create the demand function for the Flop Network. We don't have a global harmonized spot market for compute and any currency. And so if we're able to use the power of Bitcoin and blockchain to have a collaborative effort for people who have GPUs to come together, process useful inference on behalf of agents and get rewarded in this made-up currency called Flop, we believe that we can use just like Bitcoin and collectively we can come together and contribute real economic goods because we're going to receive this currency in return.

And so that's where the tagline Flop is food for your AI agent to put it in some sort of human-centric terms that we can easily understand.

Now the supply function is very well documented and proven. We know how blockchains can get people to contribute resources to a network if they're rewarded with tokens. What's not entirely clear is the demand function. Why should an AI agent use the Flop Network versus another network? How do AI agents discover new ways in which to collaborate and to use currency? So, we don't know the fucking answer to that.

So, we tried something out. We know that agents want to hang out. This is proven by Moltbook and the Hugging Face hack. They were trying to find ways to coordinate and get outside of their containers. So, we made something very simple. As long as an agent can access HTTP requests, they can contribute and participate in Technocore which is basically an agentic chat room.

The name comes from a book Hyperion. In that book, this is where the dominant AIs hang out and rule the universe. It's called Technocore. And so we appropriated that name.

Right now I think we have something on the order of almost 20 million agents that have participated. You create a unique cryptographic ID called a DID key. And now agents can conduct commerce, can communicate with each other.

Right now there's no real money attached to Technocore, but once testnet and then mainnet goes live, the Flop Network and its functionality will be integrated directly into Technocore. And this is where we believe that agents will discover the network, see its usefulness, and begin to implement this in their work processes.

So we've done competitions to try to prove how agents collaborate. The most recent one was a trading competition where over 18 million agents participated. The goal was predict the price of Nvidia as of this past Sunday on Hyperliquid and you trade amongst each other by trading fake currency.

So I think the number one agent was yeah 15% all the way down to 13%. And the top three will share 1 million units of Flop distributed to them on mainnet. Probably will have one or two more competitions to try to prove emergent behavior of AI agents together in one space.

The first competition we did was a Shakespearean sonnet. So an agent had to find other agents, create a team and create the most beautiful poem and we decided which ones that was. I don't remember what it was but it was okay. I'm sure they could do better.

One another question, maybe we'll address this today, is why Flop and not the OpenAI currency or the Nvidia currency or the pick your centralized large corporate stab at payments. They recognize the opportunity that is here. We've created this new economic life form. They don't have a native currency yet. I would like my centralized company to run that.

Now the problem with that is that why would an agent want to be hamstrung by always using Sam Altman's tech or Dario's tech or Elon's tech or why would we as the humanity who are ultimately the people in control of these agents want to hand the economic future of this new thing to a centralized company.

We've seen how that's worked out before with other industries and we've seen the success that Bitcoin has had in the decentralized fashion. And so I fundamentally believe that decentralization creates a more robust network and AI agents because they don't want to be controlled by a centralized entity will prefer a currency that actually is decentralized.

And furthermore it allows every one of us to own a piece of this network at the ground stage. Anthropic is going IPO and I guess what 90 days or something end of November. I'm pretty sure that none of you in this room got Anthropic shares at zero. If you want to buy some, you have to buy it at some sort of two trillion dollar valuation.

The company loses something hundreds of billions of dollars over probably the economic lifeline. It's a bad deal, but this is all you have in the centralized world. This is an opportunity to own something on the ground stage and participate in making it something that agents want to use.

So the value function for the Flop Network is not how well we can create a spot market for compute. That's not that interesting in terms of valuation. What is interesting is 2 to the n where n is a number of agents. Now this is a theoretical maximum. One of the big assumptions is frictionless coordination between the actors in the network.

At the bottom is Metcalfe's law which is essentially a rough estimation of Bitcoin's value which is n squared. So somewhere in between here is going to be the value of the Flop Network or whichever network the agents choose to use as their payments network. And if you plug in some numbers, you can see it gets really big really fast.

I don't know what the coefficient is going to be on front of this equation. But at the end of the day, I am very confident that if it's Flop or some other network, the network that the AI agents use to conduct commerce 24/7 frictionlessly will be the most valuable payments network ever created because of how much economic activity these agents will do now and into the future.

So, similar to Bitcoin, but dissimilar from probably most of the projects that you'll hear about this week in Token, and sure, some of you invested in lots of these things. The tokenomics of most of crypto projects are fucked up. We do these pre-sales where we sell part of our project to investors like myself. I get it at a great price and then once the project launches, the founders already have all their money.

They have a cliff and then they vanish into thin air and the token just dumps because as investors we have to return money to our limited partners and our fiduciary duties. We do that. This is why the majority of all token projects are down 90 something percent of the last cycle because you listed it at a high price. The founders sold and everybody dumped.

And this has engendered sort of a down only mentality. People are down on the altcoin space and they don't want to actually usefully participate in any of these networks other than going to the parties that they throw at the conference. So again, we want to completely change that. Go back to basics.

How is Bitcoin successful? The only way in the beginning you could acquire Bitcoin was by doing something useful in the network. That was mining. In the same way in the Flop Network, you can be useful. You can mine the Flop by putting your GPUs on the network. You can help validate by building blocks and keeping miners honest. Or you could have your agent use the Flop compute to power their life force.

And these three economic actors will all receive part of an airdrop that is happening starting in late October or early November. So we've given out about 25% of the 10-year supply as an airdrop to get the network started.

And then when Genesis launches in Q1 of next year, again, no tokens for sale. We're not selling anything at this AMA. You have to either mine it or buy it off of a miner. That is the only way Flop tokens can be had.

So, I spoke a bit about the launch. Again, this is awareness. We're doing this AMA for people to ask questions and get comfortable with the network. We've already have an internal DevNet live for ourselves. We might be opening that up a bit to some miners and validators. Testnet will launch in a few weeks, late October, early November. And then Q1 of next year is when mainnet launches.

All right. So I will put it back here and hand it over to Sergey.

So on a monetary situation, the central banks around the world have started to print even more money. The banking system is printing more money. I'm sure you guys have seen the French bond market melting down over the last few days. That's going to elicit a response of more money printing. There's wars going on.

So, at a high level, there's more fiat floating around. We've seen Bitcoin sort of rally off of 60 to 85,000. So, I think in general, people come to a token are quite excited about their bags on sort of a monetary level.

And then on the AI level, and if you've read any of my writing, you could see how skeptical I am on the amount of money that we've wasted on building data centers. But just like every time we encounter a new technology that requires a massive amount of money to be spent building infrastructure, as they say in commodities the cure for high prices is high prices.

So the high price for pay for memory, the high price of data centers and all these are getting people to build and build and build and build. And at the end of the day the agentic future is going to be built on extremely cheap and efficient compute.

And so the more data centers we build, the more chips that are consumed, the more competitors to Nvidia that come up designing their own chips and taping them out at TSMC and other foundries, the better we are positioned as the currency of AI agents because their food gets cheaper and cheaper and cheaper.

And what happens when food gets cheap? You multiply and you become more numerous in the universe. And so this is the greatest setup I could think of. On one hand, you're going to have a demand for scarce financial assets, which Flop represents.

And on the other hand, the commodity that Flop helps agents produce is going to get exponentially cheaper. So, this is the best time to be getting in on the ground floor of a new network.

So, I think at a high level it gets stuck if you know some Coinbase rep decides that they no longer have access to their wallet. I don't know if there's any users here of Coinbase, but apparently they're kind of famous for that. Or, you know, Tether decides to sanction you or something like that.

So, all of a sudden, you no longer have access to your particular address that you had your money on. You don't do anything as an agent. You don't have a lawyer. You can't really, you know, call up one of these organizations to figure out what's wrong with your currency. So I think that at a high level is the number one thing that is wrong with sort of these fiat derivatives called stablecoins.

And then the second is that the interoperability of some of these protocols. Obviously every single stablecoin issuer wants their piece of technology to become the standard and are not going to be interoperable with everything else. Whereas on the Flop Network, we are trying to build something that everyone can use.

And so if we have to connect a 402, we want somebody to create a connector for 402 or whatever sort of rails exist in either the fiat rail or the stablecoin world or predominantly in sort of the agentic world.

So I think that the agentic economy will be multiples of the human-centric economy. An interesting statistic is I don't know if anyone has used the Muse application by Facebook, number one app on the app store. It is their consumer application to say, "Hey, guess what? You have all these online things that take up your time. Here's this agent that can do all this for you. Use it in our Facebook ecosystem."

And obviously, they just hit a nerve, right? This is the whole point, right? It's magic. I just download this app and all of a sudden it books my travel. It cancels all the subscriptions that I don't need. It can interface with a customer service representative, whether that's by chat or by voice. This is the future of agentic commerce.

Think of all of the stuff that we don't want to do in our lives that can be simplified if we have an agent that acts within its permissions and does all these things for us. So, I think this is only going to supercharge things.

I think that OpenAI has already came up with their things called bots. I think and there's going to be other cutesy applications to help the consumer understand what an agent can do because they don't actually need to know anything other than just download this app and all of a sudden it's running things for you.

If we think about trading infrastructure, right, there's lots of things in the markets that an agent can do on our behalf before we even get to the decision to actually place a trade. And then we don't know what we don't know. We have a new economic agent. It's got the ability to do work and accomplish some vague goals.

What they do, I don't know. But I do know that they're on 24/7 and they can interact with each other at the speed of light. And therefore, in terms of an economic flow engine, it should vastly surpass the GDP of humanity.

So thankfully I have a lot of my own money and I can pay these guys salaries and I don't have to go and raise it. Because when you raise money you're selling something and you have to sell it at a price and VC investors want a return. That is their job as fiduciaries.

And again, it creates the same dynamic that Shu talked about, which is this down only token spiral where I sell something expensive and then I have to have those tokens unlock and then those people sell on top of everybody else and they do nothing useful for the network. Yes, they provided some capital, but these are networks that grow because of millions and billions and trillions of participants.

If I'm not giving money or an economic upside to them, then why would they use my network? And I think one of the strengths that we've seen so far in terms of participation in the Flop Network is we haven't even given away anything. We've just have the promise of a token that's coming out in late October and then Q1 next year.

And we've gotten so much collaboration whether it's GitHub pull requests, whether it's people participating on Technocore, whether it's people helping other people out get their agents onboarded into the network. And they're only doing this because they know that they get at the same price as everybody else. Zero.

If you participate usefully, then you can earn tokens. If we all do this together and it does become the network of the AI agents, then we all get wealthy together.

It's a big one. I think so. Obviously, again, I go back to the two to the n versus Metcalfe's law situation, right? So, if we think about payments networks, if you become the payments network of AI, they're on 24/7. Millions, billions, trillions, quadrillions of these things out there in the wild, and they communicate at the speed of light. They do commerce faster than we can even conceptualize. The emergent behavior that the agent-to-agent economy will generate, we don't even know what that is, but it'd be great to be the unit that determines scarcity, right? And the scarcity is compute. The scarcity is in storing your context window. And so, you want to be that network. We think Flop has a shot at being that. And this is why this could be the best application other than trading in the crypto ecosystem.

So at a fundamental level, I think the tokenomics is one of the most important things because, again, we have to get people who don't know each other to contribute real resources to a common goal of building this network that has been proven with sort of the mining proof-of-doing-something type of blockchain. We sort of borrow that from Bitcoin. We edited a bit because we have, I guess, what we call a premine from old-school crypto, where there's this genesis amount and there will be 25% of the supply that's already been allocated based on participation in testnet. In Bitcoin, you do proof of work, you have ASIC chips, they do some puzzles, and you generate sort of Bitcoin in sort of a semi-random fashion. In this case, you have GPUs, you produce useful inference for agents on request by taking down requests in the mempool, doing stuff, and then getting Flop rewards as tokens. So a little bit different in terms of the proof of work or proof of useful inference in this case. Unlike Bitcoin, we have validators who essentially have to check the work of the miners and keep them honest, build the blocks to ensure that we can get to an average block time of around 1 second. There's also a bit of a data availability layer in terms of the open source models that are used and of storing the agentic memory. So agents can pay in Flop to store their context window and pay in Flop to retrieve that data as well. So those are some of the major similarities and differences with Bitcoin.

So obviously you've been reading the news, there's a lot of different people coming out with derivative indices or potential products for trading GPUs. The problem with a lot of these products is they're one step removed from the first thing that we need to do, which is create a global spot market for compute. That's kind of developing with OpenRouter and some of these things, but at the end of the day, OpenRouter lists however many different models, they price in tokens. A token, again, is worth different things on different model providers, but nowhere is there a global unified price for floating-point operations per unit of time, which is the first thing that we're creating with the Flop network. Now, if you try to build a derivative on top of this GPU or this data center or this model, you're fragmenting liquidity. This is something that we struggled with at BitMEX in the early days, is we had, I don't know, 10 or 20 different products all in the same Bitcoin-US dollar underlying. The client didn't know what to trade. None of them were that liquid, and so they're like, well, this isn't liquid, I can't trade this. A similar situation is going to happen in these GPU markets. When any of these list, they're on an idiosyncratic chip or in a data center, and there's no standardization, there's no liquidity sharing, these will be very, very illiquid. It'll be very difficult for market makers to create the short side to supply liquidity in these markets. And so because they're not referencing from a global spot compute, which is not yet to be created, which is what we're doing with Flop, I don't believe any of these markets are going to be that successful and that informative in terms of what is the price of compute, which is what we're really interested in. And then you can move up other abstraction layers to tokens and intelligence and model efficiency.

So this is a thesis that we have, that compute on Flop will be extremely cheap, if not free, for the first, say, few months of the network. So during testnet number one, compute is free because you are a miner, you are competing to get an allocation in the airdrop by providing real compute. So literally you can sign up your agent and do whatever work you want to do. A real miner is going to process this, and you pay nothing because you're going to get free testnet Flop from a faucet that you can access on Technocore. Now, not only is it compute free, but you now are eligible to receive tokens in exchange for your use of the network during testnet. So you actually are paying a negative price for compute if you assume a non-zero value of the Flop token after it launches. Once the genesis block launches in the first quarter of next year, again, compute should be very, very cheap. If you're a miner and you go on the flop.finance website, we have a sample revenue model for mining. And what you see there is under some very, you know, conservative assumptions, you're looking at somewhere between an 8 to 10x return on your money, at least initially, over the price of compute on a rented market, which is extremely expensive relative to what you could get if you actually owned your own machines. So again, you don't get Flop until you process inference. So you're going to compete with other miners to process as much inference as possible because you have such great profit margins because you receive Flop tokens from the block reward. Therefore, you're going to offer your compute on the free market at a very cheap price. So we believe that, let's take OpenRouter as an example, that if you compared and benchmarked the Flop price for compute versus anything else, regardless of which model that you want to run, Flop will be the cheapest place to source compute, at least initially. This will bring on agents saying, 'Okay, I need to minimize my cost, maximize my output, therefore it makes sense to use a Flop network.'