Arthur Hayes Predicts AI Bubble Will Conclude with a Trillion-Dollar Bailout

Hayes Sees a Multitrillion-Dollar Reckoning

Arthur Hayes has a curious thesis for the AI boom. The business economics are lousy, the infrastructure binge is headed for trouble, governments will eventually foot the bill, and crypto could come out sitting pretty.

Asked by Bitcoin.com News Media Product Manager David Sencil whether AI is a bubble during an interview at KBW 2026, the Maelstrom partner didn’t hedge. “Absolutely, it’s a bubble.” Hayes believes the reckoning could arrive around 2027 or 2028, when AI companies face enormous commitments for data centers built on expectations of future demand.

His problem starts with who’s buying all that computing power. “The end demand, the people paying for AI compute, is Anthropic, OpenAI, and xAI,” Hayes detailed. “None of these companies make any money.”

In Hayes’ telling, enormous sums are being committed to infrastructure while the companies expected to consume that capacity remain economically shaky. Eventually, he expects the bill to come due. He noted:

“That’s the aha moment. I think we’re maybe 12 to 18 months from that point.”

Too Big to Fail, AI Edition

Hayes thinks the present phase is less boom than expensive positioning. “Right now we’re in the capital wastage phase,” he said. His thesis gets considerably bigger from there. Hayes expects major AI companies and their surrounding infrastructure to become sufficiently important that Washington would have powerful incentives to prevent a disorderly collapse. Data-center debt, he argues, could reach far beyond Silicon Valley because insurers own portions of that financing.

“If that debt goes bad, it hits average policyholders across the United States,” Hayes explained. That, he believes, is where the heat is on. “That’s when, from a social stability standpoint, the government steps in and bails it out to the tune of multiple trillions of dollars.”

For crypto, Hayes sees the prospective rescue as a liquidity event. His broader macro thesis is that governments ultimately choose money creation over allowing widespread losses to rip through heavily indebted economies. “The point is they’re going to keep printing money instead of letting the chips fall where they may,” Hayes told Sencil.

Building FLOP for the Wreckage

Then comes the twist. Hayes isn’t merely waiting around for an AI bust. He’s building a business designed for what he thinks the boom leaves behind.

“I 100% believe AI is a bubble, but something useful will come out of it, and that’s agentic commerce,” Hayes said during the interview. His Flop Labs project is developing FLOP, a blockchain network intended to connect AI agents with computing power. Miners contribute GPUs, process inference and receive transaction fees plus FLOP block rewards. Hayes likens the design to Bitcoin, with no venture-capital allocation or presale.

An airdrop is slated to begin in late October and run 90 days, distributing about 25% of the planned 10-year supply to holders. Mainnet is scheduled for the first quarter of next year “The only way to earn FLOP is by doing useful work in the ecosystem: mining, validating, or having your agent use the network,” Hayes emphasized.

The Stranger Bet Behind the Bubble

Cheap compute, normally a threat to anyone selling computing power, is precisely what Hayes wants. “The cheaper the compute, the more you use,” he remarked. His wager is that a colossal AI infrastructure buildout eventually makes computing abundant enough for billions, perhaps trillions, of software agents to perform economic tasks continuously. FLOP would aim to provide the currency and coordination network underneath that machine economy.

“We hope FLOP might be the most valuable cryptocurrency ever created, but most people will have no idea what it is,” Hayes explained during the interview. That makes Hayes’ bet a peculiar one. The AI bubble doesn’t have to avoid disaster. In his version of events, the excess, the bailout and the cheap computing power left afterward are part of the opportunity.