Arthur Hayes, chief investment officer of the crypto fund Maelstrom and chief executive of Flop Labs, is arguing that the artificial-intelligence buildout is less a durable earnings story than a credit boom whose eventual strain could force official liquidity back into digital assets.
In his September 21 essay “Safety First,” published on Crypto Trader Digest,
Hayes treats recent pledges by leading US labs to slow work toward artificial general intelligence as a commercial signal rather than a purely ethical one. Anthropic, OpenAI, and SpaceX, he writes, are not mainly pausing because of concern for human welfare.
He contends the market wants AI, but closer to the much lower prices associated with Chinese models—on the order of one-hundredth of prevailing U.S. prices—and that “safety first” is a convenient cover for weaker demand at current rates.
That distinction matters, in his account, because the labs’ projected compute needs sit under a large financing structure.
Hayes says demand from the three dominant US labs underpins more than $1 trillion of investment-grade debt and hundreds of billions of dollars in lower-rated loans, raised to lease data centers and buy chips.
He argues the labs remain unprofitable on a full-cost basis, and that a shift from training spend toward efficiency would cut orders for data centers, semiconductors, and cloud capacity.
Hardware that depreciates quickly, paired with long-dated obligations, is the mechanism he compares with earlier credit excesses rather than with the equity-only bust of 2000.
The holders he highlights are not only banks and private credit funds.
Hayes says US insurers have accumulated exposure to AI-linked private credit, so a mark-to-market repricing could leave parts of the insurance sector impaired.
From there he sees a binary official response: Washington becomes a buyer of last resort for surplus compute on national security grounds, or it prints money to support insurers sitting on distressed AI debt.
Either path, he argues, expands the dollar supply.
“Trump has a choice, print or print,” he wrote when promoting the essay.
He has separately said that sequence could carry Bitcoin to $1 million and beyond, with capital-expenditure growth possibly losing momentum from mid-to-late 2027 and the slowdown becoming clearer in 2028.
Those price and timing claims are forecasts, not observed outcomes, and Hayes’s earlier calls on crypto markets have often missed.
He restated the liquidity side of the thesis this week in Seoul, during Korea Blockchain Week.
AI firms, he said, need trillions of dollars for data centers even as the price of their services falls, and policymakers have “not really given themselves a lot of options other than print money and make it less bad.”
He tied that pressure to the financing of government debt as well, and said a shift in China away from what he called “austerity lite” could add another bid for scarce assets.
A second strand of the argument runs through Flop Labs.
Hayes has proposed Flop Network, a draft blockchain on which AI agents would pay miners for verified inference under a model he calls proof of useful inference, with the FLOP token meant as a claim on compute—“food for AI agents.”
He has also said the token would be worth little more than a compute marketplace unless agents begin transacting with one another on the network, and that the team has not yet worked out how agent disputes would be settled. In the “Safety First” framing, a glut of cheap compute left after a financing bust is the feedstock for that agent economy, while the official rescue of the debt is the fuel he hopes will reprice crypto.