Arthur Hayes Thinks the AI Data Centre Boom Ends in a Bailout. He's Built a Crypto Project for That Moment.

Key points

• Arthur Hayes, former CEO of crypto exchange BitMEX, says the world is building too many AI data centres and a crash is coming, likely in late 2027 or 2028.
• Hayes is launching Flop, a cryptocurrency project expected in early 2027 that would let AI agents, software programs that carry out tasks on their own, pay for computing power directly.
• SpaceX, OpenAI and Anthropic drive demand for computing, Hayes told CNBC, but none currently makes money.
• His central bet: when governments bail out the overbuilt AI industry, Bitcoin absorbs the flood of new money.
• Flop is designed to create a live open marketplace for raw compute, rewarding GPU contributors with its own tokens.

Arthur Hayes watched the 2008 financial crisis print a playbook he thinks he recognises again. Build too much, crash hard, get bailed out. This time, he says, the thing being overbuilt is AI infrastructure.

Speaking at the Gamma Prime Investing Conference in Singapore, Hayes told CNBC that humanity is spending trillions constructing AI data centres, the vast warehouse-scale buildings packed with specialised chips that train and run AI models. His view is blunt: capacity will outstrip demand.

"If you study financial history and you study every single major technological rollout, it always is overbuilt," Hayes said. "There always is a crash, and there always is a bailout."

When does the crunch arrive?

Hayes puts the danger window at late 2027 to 2028, when much of the data centre capacity now under construction is completed and providers start demanding payment for the computing they were promised.

SpaceX, OpenAI and Anthropic are the ones signing those commitments, Hayes said, and none is profitable yet. If AI doesn't become useful enough to generate real revenue in the next year or so, those bills come due with no income to cover them.

The bull case, Hayes acknowledged, is that AI tools become genuinely indispensable fast enough that companies do turn profitable in time. Memory chipmakers and Nvidia, which sells the GPUs (specialised chips that do the heavy number-crunching for AI) filling those data centres, are already making money. The question for investors is whether today's share prices already assume too much growth.

What is Flop, and why does it matter for AI agents?

Hayes sees the glut itself as an opportunity. Cheap, abundant computing power is the precondition for Flop, his new crypto project, expected to launch in the first quarter of 2027.

Flop aims to build a spot market for raw computing power. Participants who contribute GPUs or run AI inference tasks (feeding data through a trained model to get an answer) earn Flop tokens in return.

The core idea is that AI agents need a native way to pay for compute. No payments network is currently built for software agents; they rely on conventional billing systems designed for humans. Hayes wants Flop to fix that.

"If agents can convert a currency directly into compute, which is what they eat and consume, then they will use this currency," he said.

AI agents are moving fast from curiosity to commercial reality, and a payments layer built for them would slot into a rapidly expanding ecosystem. Our August story on Binance's Agent OS launch showed crypto infrastructure already bending toward autonomous software, though the safety guardrails there were largely left to users to configure themselves. Flop is pitching a more fundamental piece of plumbing: not a trading interface but an economic substrate.

Hayes isn't shorting AI companies. He just thinks patient crypto holders win when the bailout lands. Whether that patience pays off depends entirely on how fast AI revenue catches up with the build. That's the number to watch, not the conference rhetoric.