Arthur Hayes has a theory, and it’s the kind that makes you either close your laptop or start aggressively buying Bitcoin. The BitMEX co-founder is arguing that the AI infrastructure boom is building a credit bubble with a very specific, very dangerous structural flaw: lenders are financing GPU hardware over 5-6 year terms, but the chips themselves become obsolete in roughly 2-3 years.
The GPU financing time bomb
Here’s the thing about lending against depreciating hardware. The math only works if the borrower generates enough revenue during the loan term to cover payments long after the collateral is worthless. Hayes points to Nvidia’s H100 chips as a prime example, hardware with a useful life of about 2-3 years that’s being financed on timelines roughly double that.
Hayes estimates approximately $1.5 trillion in AI-related debt has been issued between November 2022 and mid-2026. That figure isn’t random. It corresponds almost exactly to a $1.5 trillion increase in the US M2 money supply over the same period, suggesting the AI buildout has effectively absorbed the new liquidity that might have otherwise flowed into risk assets like Bitcoin.
Perhaps more alarming: 75-80% of that AI debt issuance reportedly occurred in 2025 alone. That kind of concentration signals a peak in capital expenditure, the kind of frenzied spending that historically precedes a correction.
Why Bitcoin benefits from financial crises
Hayes’ central thesis rests on a pattern that Bitcoin investors know well. When credit markets seize up, central banks and governments respond with massive liquidity injections. In Hayes’ framework, Bitcoin functions as a bellwether for global fiat liquidity. When the money supply expands, Bitcoin tends to outperform. When liquidity gets absorbed elsewhere, as it currently is by AI infrastructure spending, Bitcoin underperforms relative to what you’d expect given the monetary backdrop.
But when the AI credit bubble pops, Hayes argues, that dynamic reverses violently. Central banks inject fresh liquidity to prevent systemic collapse, and this time there’s no AI capex boom to absorb it. His price target for that scenario is $1 million per Bitcoin, which would imply a network value of roughly $21 trillion.
The timeline and the Ethereum angle
Hayes places the anticipated stress point in the AI credit market between late 2027 and 2028. That timeline aligns with when the earliest wave of 5-6 year GPU loans would start maturing against hardware that’s already been replaced two generations over.
He also sees this period potentially coinciding with political shifts around AI regulation and taxation, adding another layer of pressure on companies that borrowed heavily to build AI infrastructure.
On the Ethereum side, Hayes projects ETH could reach between $100K and $200K during on-cycle peaks following the anticipated bust. His near-term forecast is more modest but still bullish, placing Ethereum at around $5K by the end of 2026.
What this means for investors
If Hayes is right that AI capex is suppressing Bitcoin’s response to monetary expansion, then any slowdown in AI infrastructure spending, even before a full credit crisis, could release pent-up upside for crypto assets. Watching enterprise GPU orders, data center construction rates, and the credit spreads on AI-related corporate debt may prove just as useful as watching on-chain metrics for the next major Bitcoin catalyst.