Key Takeaways
- • Arthur Hayes argues AI infrastructure is massively overbuilt, with "multi-trillion dollars" being wasted on data centers that will eventually crash and trigger government bailouts, creating tailwinds for Bitcoin and cryptocurrencies.
- • Hayes expects the reckoning in late 2027 or 2028, when infrastructure providers demand payment for compute capacity that AI firms like OpenAI, Anthropic, and SpaceX have committed to but cannot yet monetize.
- • Anthropic's financials support Hayes's thesis: the AI company reported $4.6 billion in 2025 revenue but posted net losses near $42 billion (including a $34 billion noncash charge), with compute and infrastructure costs consuming $7.33 billion—over half of $12.65 billion in total operating expenses.
- • Historical precedent frames Hayes's prediction: he argues that every major technological rollout has followed the same pattern—overbuilding, crash, and state bailout—citing the 2008 financial crisis as validation that excess liquidity will eventually flow into alternative assets.
- • Hayes leaves room for an AI bull case, acknowledging that stronger demand within a year could make infrastructure companies more profitable, but maintains his base case is a liquidity event that favors crypto.
Summary
Arthur Hayes, former BitMEX CEO and co-founder of crypto investment firm Maelstrom, told CNBC at the Gamma Prime Investing Conference in Singapore that the artificial intelligence infrastructure buildout represents a catastrophic misallocation of capital that will inevitably collapse and benefit Bitcoin. Hayes contends that humanity is "wasting multi-trillion dollars" on AI data centers that will become "extremely cheap and extremely plentiful" once construction concludes, setting the stage for a crash followed by government intervention.
Hayes's framework draws from financial history. He argues that every major technological rollout exhibits the same pattern: overbuilding, a crash, and a state-sponsored bailout. Under this thesis, the excess liquidity released during a rescue package would flow into alternative assets, with Bitcoin and crypto positioned to capture the majority of those flows. "Thankfully, we have Bitcoin and other crypto to soak up that excess liquidity," Hayes stated, urging investors to "just be patient."
The timing of Hayes's prediction centers on late 2027 or 2028, when infrastructure providers will demand payment for committed compute capacity from major end users—including OpenAI, Anthropic, and SpaceX—that currently operate at massive losses. Anthropic's recent prospectus filing ahead of a potential IPO substantiates this concern: the company reported $4.6 billion in 2025 revenue while posting net losses near $42 billion (including a $34 billion noncash charge). Critically, compute and infrastructure costs consumed $7.33 billion of the company's $12.65 billion in operating expenses, highlighting the unsustainable unit economics of current AI deployment.
Hayes acknowledges that stronger demand within a 12-month window could improve infrastructure provider profitability, but his base case remains a liquidity crisis. His thesis hinges on the observation that current AI end users cannot generate sufficient revenue to justify their capital expenditures, creating an eventual cash crunch that will force either private restructuring or public sector intervention—either scenario, in Hayes's view, would redirect capital toward cryptocurrencies.
Hayes's framework draws from financial history. He argues that every major technological rollout exhibits the same pattern: overbuilding, a crash, and a state-sponsored bailout. Under this thesis, the excess liquidity released during a rescue package would flow into alternative assets, with Bitcoin and crypto positioned to capture the majority of those flows. "Thankfully, we have Bitcoin and other crypto to soak up that excess liquidity," Hayes stated, urging investors to "just be patient."
The timing of Hayes's prediction centers on late 2027 or 2028, when infrastructure providers will demand payment for committed compute capacity from major end users—including OpenAI, Anthropic, and SpaceX—that currently operate at massive losses. Anthropic's recent prospectus filing ahead of a potential IPO substantiates this concern: the company reported $4.6 billion in 2025 revenue while posting net losses near $42 billion (including a $34 billion noncash charge). Critically, compute and infrastructure costs consumed $7.33 billion of the company's $12.65 billion in operating expenses, highlighting the unsustainable unit economics of current AI deployment.
Hayes acknowledges that stronger demand within a 12-month window could improve infrastructure provider profitability, but his base case remains a liquidity crisis. His thesis hinges on the observation that current AI end users cannot generate sufficient revenue to justify their capital expenditures, creating an eventual cash crunch that will force either private restructuring or public sector intervention—either scenario, in Hayes's view, would redirect capital toward cryptocurrencies.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.