AI Agents Will Need a Currency Tied Directly to Compute

AI agents are beginning to develop something they have never had before: an economy of their own.

They can increasingly discover services, call APIs, purchase data and execute tasks without requiring a human to approve every individual transaction. Today, agents are processing over 258.5 million cumulative transactions across x402.

But as these systems become more autonomous, one question becomes harder to ignore: how will AI agents pay for the resources they need to keep working?

Arthur Hayes, CEO of Flop Labs and CIO of Maelstrom, believes the answer could require an entirely new kind of money. Hayes is building FLOP around a simple but ambitious idea. Cmpute could become the foundation for a new economic layer designed specifically for autonomous AI agents.

“The agentic economy is one of the fastest growing but lacks a native payments network,” Hayes told Block of Fame.

“The agentic economy is one of the fastest growing but lacks a native payments network. At the same time, there is no global unified market for spot compute (floating point operations per unit of time).

Therefore, there is an opportunity; if the Flop Network creates a spot market for compute denominated in FLOP, then agents that need compute to survive, will intrinsically demand FLOP for payment.”

That distinction is central to Hayes’ argument.

Rather than creating another token and then searching for an AI use case, FLOP is attempting to start with the resource itself – compute.

Why is Arthur Hayes Building FLOP

FLOP’s Yellow Paper describes the project as a verified-inference settlement layer. In simple terms, the network is designed to connect the people or agents that need AI computation with the miners providing that computation, while measuring and settling the work.

The basic model is straightforward. An AI agent needs compute. A miner provides that compute. FLOP is used to pay for it. The network verifies and settles the activity.

Hayes’ argument is that today’s money does not have a direct relationship with that resource.

“There is no currency that has a direct link to compute. Every existing currency’s value is depending on variables other than compute.” he said.

This is the central idea behind FLOP. Instead of treating compute as something that is simply purchased with dollars, euros or crypto, the network is designed around measuring computational work itself. Its Yellow Paper introduces Effective-FLOPs, or F_eff, as a reference-work measurement and G_n as the metered unit used for accounting and settlement.

What’s Next for Agentic Economy

The timing of that bet is important because AI’s appetite for compute is becoming one of the defining constraints of the industry.

The economics are already visible at the infrastructure level.

Other Articles…

AI companies are signing enormous contracts simply to secure access to computing power. SpaceX, for example, recently disclosed a $1.11 billion-a-month compute hosting agreement with an undisclosed customer. Anthropic’s is also eyeing a separate compute arrangement.

At the market level, GPU compute is becoming financialized too. Nvidia’s GPUs account for roughly 60% of AI server costs, according to Reuters. CME Group and Intercontinental Exchange are developing futures linked to GPU rental prices.

That creates the backdrop for Hayes’ argument.

Compute is no longer simply a technical input buried inside an AI product. It is becoming an increasingly visible economic resource with its own pricing, capacity constraints and financial markets.

Hayes believes the next step could be giving that resource a native monetary unit. In few years, he sees an entire agentic economy around FLOP emerging.

“AI agents will transact in FLOP, convert FLOP into compute to accomplish work, and spend FLOP to store their memories on a decentralized censorship resistant network,” Hayes said.

From AI Narrative to Economic Infrastructure

The crypto industry has already seen multiple waves of AI-related projects. This includes decentralized compute networks to AI agents and AI-linked tokens.

The challenge for FLOP will be demonstrating that its token and network are necessary for the activity it is targeting.

Hayes points to agent behavior as the key test.

“Agents coordinate between themselves and price their work in units of FLOP.”

That is a much higher bar than simply creating demand through speculation.

For FLOP to become meaningful infrastructure, agents would need to actually use the network to coordinate work, purchase compute and preserve information.

“Agents will Demand to be Paid in FLOP”

FLOP’s planned Q4 2026 airdrop and 2027 network launch will therefore be less about introducing another AI token. It will be more about testing whether the economic thesis can work in practice.

The key question is whether agents will actually need FLOP to operate.

Hayes believes the answer depends on two things: compute pricing and persistent decentralized memory.

“If the Flop Network becomes the price setter for spot compute, and if agents that wish to exist and persist save their memories on the Flop Network. Then agents will demand to be paid in FLOP because it is the cornerstone of their existence.”

That is the real bet.

FLOP is not simply trying to make AI agents capable of using crypto. It is attempting to make crypto infrastructure part of what AI agents need in order to operate.