Arthur Hayes: AI agents need a currency directly linked to computing power

Arthur Hayes: AI agents need a currency directly linked to computing power

With the continuous growth of autonomous machine-to-machine payments, Arthur Hayes believes that artificial intelligence (AI) agents need a currency directly linked to computing power. FLOP is designed to allow AI agents to pay for verified reasoning processes, while miners provide the required computing power. Artemis data shows that x402 and MPP have processed a cumulative 258.5 million proxy payment transactions, indicating that a rapidly emerging machine economy is taking shape.

Autonomous economy of AI agents

AI agents began to develop what they had never had: their own economic system. Increasingly, they are able to discover services, call APIs, purchase data, and perform tasks without having to be approved by humans for every individual transaction. Currently, agents have processed more than 258.5 million cumulative transactions on x402.

However, as these systems become more autonomous, one question becomes impossible to ignore: How will AI agents pay for the resources needed to maintain work?

Arthur Hayes, CEO of Flop Labs and chief investment officer of Maelstrom, believes the answer may require a completely new type of currency. Hayes is building FLOP around a simple but ambitious idea: Computing power could become the foundation of a new economic layer designed specifically for autonomous AI agents.

"The agency economy is one of the fastest growing areas, but it lacks a native payment network," Hayes told Block of Fame."At the same time, there is no globally unified market for spot computing power (the number of floating point operations per unit of time). So there is an opportunity; if Flop Network creates a spot market for computing power denominated in FLOP, then agents that need computing power to survive will use the intrinsic demand FLOP as a means of payment."

This difference is at the heart of Hayes 'argument. Instead of creating another token and then looking for AI use cases, FLOP attempts to start with the resource itself-that is, computing power.

Why Arthur Hayes built FLOP

FLOP's yellow paper describes the project as a "proven inference settlement layer." In short, the network is designed to connect entities or agents that require AI computing with miners that provide computing, while measuring and accounting for the results of work.

The basic model is very simple: AI agents require computing power, miners provide computing power, use FLOP for payments, and the network is responsible for verification and settlement activities.

Hayes's argument points out that today's money has no direct relationship with this resource. "No currency is directly related to computing power. The value of all existing currencies depends on variables other than computing power." he said.

This is the core concept behind FLOP. The network does not view computing power as something purchased only in dollars, euros or cryptocurrencies, but rather is designed around measuring the computing work itself. His yellow paper introduces "Effective FLOP"(F_eff) as a reference workload measurement standard and uses G_n as a unit of measurement for accounting and settlement.

Proxy for the future of the economy

The timing of this bet is crucial because AI's demand for computing power is becoming a key constraint for the industry. Economic trends at the infrastructure level are already emerging. AI companies sign huge contracts simply to ensure access to computing power. For example, SpaceX recently disclosed a $1.11 billion monthly computing power escrow agreement, with the customer's identity not disclosed. Anthropic is also considering separate computing power arrangements.

At the market level, GPU computing power is also gradually being financialized. According to Reuters, Nvidia's GPUs account for about 60% of AI server costs. CME Group and Intercontinental Exchange are developing futures products linked to GPU rental prices.

This provides the background for Hayes 'argument. Computing power is no longer just a technological investment hidden within AI products. It is becoming an increasingly visible economic resource with independent prices, capacity constraints and financial markets.

Hayes believes that the next step may be to assign a native currency unit to this resource. He foresaw that a few years later, the entire agency economy would emerge around FLOP. "AI agents will trade in FLOPs, convert FLOPs into computing power to get the job done, and spend FLOPs storing their memories on a decentralized censory-resistant network," Hayes said.

From AI narratives to economic infrastructure

The crypto industry has witnessed a wave of AI-related projects, including decentralized computing networks, AI agents and AI-linked tokens. The challenge for FLOP is to prove that its tokens and networks are indispensable to the targeted activity.

Hayes points out that agent behavior is a key test: "Agents coordinate with each other and price their work in FLOP."

This is much higher than the threshold for creating demand through mere speculation. In order for FLOP to be a meaningful infrastructure, agents actually need to use the network to coordinate work, purchase computing power, and preserve information.

Agents will require payment in FLOP

FLOP is scheduled to be airdropped in the fourth quarter of 2026 and launch the network in 2027, so the focus is not on introducing another AI token, but on testing whether its economic theory works in practice.

The key question is whether the agent really needs FLOP to operate. Hayes believes the answer depends on two things: computing power pricing and persistent decentralized storage. "If Flop Network becomes the price maker of spot computing power and wants existing and continuously running agents to keep their memories on the Flop Network, then the agent will require payment in FLOP because it is the basis of its existence."

This is the real bet. FLOP is not just trying to equip AI agents with the ability to use cryptocurrencies, but is trying to make cryptographic infrastructure a necessary part of AI agents 'operations.