Arthur Hayes Says Bitcoin (BTC) Can’t Serve as AI Agents’ Native Currency

Hayes Targets AI Agent Money

BitMEX co-founder Arthur Hayes has told the Korea Blockchain Week 2026 stage that Bitcoin (BTC), stablecoins and every other digital currency in circulation today fall short as a native medium of exchange for autonomous AI agents — and he is now pushing a new project, the Flop Network, to fill that gap. Hayes, who also serves as chief investment officer at the Maelstrom fund, argued that the emerging agent economy needs payment infrastructure letting software agents request and settle payments on their own, without human sign-off. His case is definitional: Bitcoin was engineered as a peer-to-peer store of value, so its unit of account is the market price of Bitcoin (BTC) itself, while stablecoins simply mirror fiat balances. Neither expresses how much computing power a given sum can buy — the variable that matters most to a program whose survival depends on inference capacity. The Flop Network is described instead as payment infrastructure aimed specifically at autonomous machine-to-machine settlement, where agents both initiate and clear their own transactions. The timing is notable: governments are only beginning to write rules for the agent era, as in the recently signed voluntary AI accord between Washington and six technology executives, and a monetary layer designed for machines goes a step beyond even those frameworks. For now, the proposal sits at design stage rather than as an operating network, but it frames a concrete question for the industry: if agents become genuine economic actors, does money designed for humans still work for them? That debate is now running through Bitcoin commentary as well, with Hayes — historically one of the asset’s loudest champions — drawing a line between what the asset is and what the agent economy will need. Our reading of the on-stage record is that Hayes framed the gap as structural, not something incremental upgrades to existing chains can close.

FLOP Backed by Compute Exchangeability

Hayes laid out the underlying logic in separate remarks from the same event. For people, money is first of all a way to buy food; for an AI agent, the equivalent staple is computing power, so agent-ready money must be directly exchangeable with compute. His test for any candidate currency is therefore simple: can the money an agent holds map transparently to the quantity and cost of specific computational work? Existing digital currencies, in his view, can move value between parties but say nothing about what a particular inference job will cost, and that gap could become the decisive limitation of crypto assets in an agent-centred economy. The Flop Network is his proposed answer. The design connects an AI agent’s request with GPU providers holding spare capacity, and intends to settle both resource provision and payment through verifiable AI inference tasks — a verification problem far removed from proof of work, the process in which mining certifies block production rather than machine-learning output. The network’s proposed native unit, FLOP, would be what agents spend to consume compute. Flop Labs, the vehicle through which Hayes is pursuing the idea, has previously framed the token’s value as contingent on whether real agent transactions actually route through the network. One working comparison exists on the market today: Circle’s Agent Marketplace lets AI agents discover services and pay in USDC, but its focus is service discovery and settlement in a fiat-pegged token, whereas the Flop concept ties the payment unit itself to computing resources — putting FLOP in direct competition with the broader altcoin market for agent payment volume. The caveats are significant: the project remains early stage, and its operating method, whether the service is actually live, and the scale of transactions between agents and GPU providers have not been disclosed. Whether FLOP ever functions as a genuine agent payment medium depends on the network being built and participants showing up. Readers tracking the market in real time can follow live spot and futures prices on Gate.

A Test for Bitcoin’s Money Thesis

Taken together, the two statements are more than a product pitch: they test the core claim of Bitcoin maximalism, that a single credibly neutral asset can absorb the money functions of every new economic context. Hayes, himself a long-time Bitcoin holder, is arguing the agent economy will not be one of them. His on-stage remarks in Seoul — the primary record on this story — remain a design-stage vision, with nothing yet demonstrated in operation. That leaves Bitcoin’s institutional trajectory, from treasury adoption to the convergence of Bitcoin and traditional finance, running on a separate track from Hayes’s compute-money experiment — for now.