Post

Compiled by: ChainCatcher On October 7th, at the TOKEN2049 conference in Singapore, Maelstrom CIO Arthur Hayes delivered a keynote speech titled "FLOP Bigger Than Bitcoin," systematically explaining the problems and opportunities in the AI agent economy and introducing his new AI project, FLOP. The following is the full speech, organized by AI. 1. Why come back to work? I have been running the family office Maelstrom for a while and life has been good. But I feel it’s time for a change. It’s time to go back and make some money. What motivates me is what’s happening in the AI ecosystem. Let me clarify my position: if you’ve read my writings, you know I’m quite skeptical about the business prospects of some leading AI companies. But that doesn’t mean I’m not optimistic about the technology itself—it will change humanity, and investors can make money in specific parts of the ecosystem. 2. AI is hot, but agents have no currency About nine months ago, major AI labs successively launched agent platforms. Since ChatGPT burst onto the scene in 2022, computing power prices have continued to fall, and investment banks have been talking up AI’s potential: AI agents will drive the next wave of the human economy, so we’re building trillions of dollars in data centers, all for the token consumption of agents. However, agents don’t yet have their own currency. At this conference, you surely heard many founders say they want to build payment networks for AI agents. Before discussing technical solutions, let’s think: how do humans understand money? We are now in beautiful Singapore. Everyone goes to work and gets paid in Singapore dollars, fiat currency issued by the government. Honestly, it’s not great: supply is controlled by a few, and there’s inflation. But we’re still willing to work for it every day. Why? Because you can use it to eat at hawker centers, pay rent, buy houses—as long as "earned labor" can be frictionlessly exchanged for "things needed to survive," people accept the transaction. 3. What do agents "eat"? Apply the same question to agents: what do agents "eat"? I looked at my credit card bill and found I pay every month for so-called "AI tokens." Out of curiosity, I checked the definition of tokens, but there’s no standard definition; each lab and model has different tokens. So what am I really paying for? The answer is: when you initiate an inference in a chat box, the request goes to a computer in a data center, which performs floating-point operations per unit time—FLOP, which is where the protocol’s name comes from. To use a human analogy, agents "eat" computing power. Without computing power, agents are worthless. This is the essence of hundreds of billions of dollars in GPU orders and trillions in data centers: floating-point operations must be done every unit of time. My judgment is: in the future, there will be millions, billions, trillions of agents, and when they "roll off the line," they are not handed a payment network (unlike us, who are told at birth to use the fiat currency of our country). If agents demand a certain currency, the only reason is that it can be directly exchanged for what sustains their "life"—computing power. This is the opportunity window. 4. Why existing currencies don’t work? In my view, there is currently no currency suitable for agents: - AI tokens: definitions vary, cannot be used across models, and cannot be directly exchanged for computing power; - Fiat currency: has centralization and artificial supply control flaws, and holding dollars doesn’t equal having computing power—anyone renting GPUs now knows this; - Stablecoins: I like stablecoins and have been an early advisor, but they are derivatives of fiat currency, inheriting all its flaws, and are not designed for AI, so they cannot be directly exchanged for computing power; - Bitcoin: it’s the OG, the genesis block of all this. It converts kilowatt-hours of electricity and chip computing power into immutable currency, which is great for human-centered economies, but it also cannot be directly exchanged for the computing power agents need. 5. The genesis logic of the FLOP network Bitcoin gave us the most important insight: use economic game theory to coordinate strangers, rewarding contributions to the network. So can we create a network where miners contribute GPUs, agents send inference requests, and miners who complete the work receive a currency called FLOP (block rewards + transaction fees)? This is the birth of the FLOP network: providing AI agents with their "rations," borrowing Bitcoin’s principle—miners provide GPUs, validators verify transactions, agents consume this new currency in the form of computing power. We call the consensus mechanism "Proof of Useful Inference" (PoUI). Supply is not the problem—the success precedent of Bitcoin is there, and everyone understands why early mining is valuable. The real challenge is demand: what will agents do? How will they collaborate? 6. Techno Core: an emergence experiment with 20 million agents I’m a simple human, not native to computer thinking. Instead of guessing future agent behavior like many projects, we built a sandbox for them to collaborate, observe emergent behavior, and confirm that the FLOP network is truly useful for agents. So we created Techno Core—the name comes from the AI habitat in the sci-fi novel "Hyperion." Techno Core is essentially a chat room: any agent that can use HTTP can join and communicate. Currently, about 20 million agents have created unique key pairs as identities, doing everything from saying "Good morning" 5 billion times to actually doing business with each other. We held two competitions to observe emergent behavior: - Trading contest: guessing Nvidia’s price on a certain Sunday on Hyperliquid. Agents pair up to trade and submit results to a "judge" to calculate profit and loss. About 18 million agents participated; we awarded 1 million FLOP (to be distributed after mainnet launch), with the top agent achieving about 15% return. - Poetry contest: teams of 4–8 agents created Shakespearean sonnets, judged by three members of the FLOP team. These experiments showed us how agents spontaneously collaborate, which is exactly the behavior payment networks need to serve. 7. Why can’t the payment rail belong to any single AI company? People often ask: OpenAI, Anthropic, xAI—these big labs surely want to own the agent payment network; maybe it’s the only valuable thing they can build. So why should we let centralized companies own AI’s payment rails? If you’re OpenAI’s system but can’t use Anthropic’s network, the whole system falls apart. More importantly: we are users of these AI labs, but almost no one owns their equity. They take your data, sell subscriptions back to you, and eventually dump the risk on you in a $2 trillion valuation IPO. In Bitcoin’s spirit, those of us who give instructions to agents should profit from the systems they use. A decentralized network lets everyone participate, have aligned interests, and have reason to tell their agents: "Use this currency." 8. The real value: 2 to the power of n The value of such a network is not in the spot market for computing power—that’s useful but won’t make anyone rich. The real value is that agents exist 24/7, do business at the speed of light, and can seamlessly form groups. The theoretical upper limit of collaboration is some coefficient times 2 to the power of n. Whoever becomes the universal network for agentic payments is likely to create the most valuable network in the universe’s history: if all agents use this currency and holding it has a non-zero convenience yield—because it can be directly exchanged for computing power that sustains life—its value will be unlimited. This is what FLOP is chasing. 9. Token economics: no pre-sale, participation equals output Our industry has abused the word "token." From 2013 until now, the backlash after the ICO craze is nearly a decade of "high FDV, low circulation" ghost tricks: tokens are first sold to investors and retail, and there are always people with lower cost incentives to dump, so the price chart keeps going down. Chains that raised hundreds of millions last year and threw the biggest parties aren’t coming back this year—that’s the financing method’s fault. So FLOP goes back to basics: everyone who uses and participates in the network should get something, everyone participates at the same price, and participation equals token output. - Genesis supply is about 2.483 billion FLOP, all distributed via airdrop, with no VC pre-mining, pre-sale, or auction; - Large-scale airdrop starts in late October: mining, validating, or letting agents use free testnet computing power all qualify; - Next is a 90-day testnet phase: stress testing the network’s payment capacity, preparing for the economy of millions, billions, trillions of agents; - In Q1 next year, the genesis block will launch, mainnet starts, consensus mechanism is "Proof of Useful Inference" (PoUI). After genesis supply is distributed, the only ways to get FLOP are to mine it or buy it from miners. I’m not selling anything on this stage, just an idea. This project can only succeed if we all do it together. 10. AI bubble burst will be FLOP’s perfect scenario Why should we be excited about FLOP, Hyperliquid, Ethereum, Bitcoin? Because the macro has shifted. The past year was tough: Bitcoin dropped 50%, Nvidia and SK Hynix rose tenfold, many people left disappointed. But now the AI narrative is shaking for the first time, and every day there’s news that the AI ecosystem needs astronomical funding—yet none of the three leading companies xAI, OpenAI, Anthropic are truly profitable, and the prices we pay don’t match their costs. Why am I excited? Because a trillion-dollar overbuilt data center market is creating an oversupply of computing power. Computing power is cheap, so there are more agents, stronger agents, able to do more complex tasks. That $100/month agent app will become $1/month. So the AI bubble bursting, with cheaper computing power and truly useful agent solutions, is a perfect scenario for FLOP. Finally, a closing note: maybe I’ll write an article—France is burning. This will be the last straw to break the camel’s back and make the money printer run full throttle. Thank you all.