Arthur Hayes, co-founder of the BitMEX exchange, stated that the artificial intelligence infrastructure boom, built on borrowed money, could lead to a credit crisis similar to 2008. In his opinion, the authorities' response to such a shock will be a massive injection of liquidity, which will push Bitcoin to the $1 million mark and beyond.
In his blog, Hayes wrote that investors mistakenly perceive spending on data centers and energy infrastructure as investments in rapidly growing technologies, while in essence, it is about leveraged real estate. He said that lenders will continue financing excessive construction until a slowdown in AI capital expenditures exposes weak borrowers.
Hayes characterized the AI boom as "a credit story like 2008, not an income story like 2000." According to his estimate, Bitcoin may remain in the $60,000–$70,000 range, with a possible drop to $50,000, before the credit cycle and the subsequent regulatory response trigger a recovery. Ethereum, he predicts, will reach $5,000 by the end of the year—the Maelstrom fund intends to increase its position in this asset and simultaneously sell out-of-the-money Ethereum put options.
How Hayes's Position Has Changed
These current statements complement Hayes's earlier reflections on the controversial influence of AI on cryptocurrency market liquidity. On May 13, he said that competition between the US and China in the field of AI would stimulate bank lending and the issuance of fiat money, which would be beneficial for Bitcoin. However, on June 4, Hayes sold his positions in HYPE and NEAR, warning that large equity placements by AI companies could divert capital away from cryptocurrencies.
The scale of the commitments underlying the AI boom is already visible in the reporting of the largest technology companies. As reported by Reuters, Microsoft, Meta, Oracle, Amazon, and Alphabet have already undertaken commitments of approximately $1.09 trillion for lease agreements that have not yet commenced—primarily for data centers.
This amount is almost four times the sum of lease obligations (about $285 billion) already reflected on the balance sheets of these companies. Reuters clarifies that the $1.09 trillion cannot be directly equated to debt: these are undiscounted payments spread over several years into the future.
The Debt Burden is Unevenly Distributed
Financial stress affects companies differently. According to a separate Reuters analysis, Oracle's debt is approximately 4.3 times its earnings before interest, taxes, depreciation, and amortization, while for Alphabet, Amazon, Microsoft, and Meta, this ratio remains below one.
S&P Global analyst Andrew Chang noted that Oracle's data center lease agreements are signed for terms of 15 to 19 years, while its contracts with its own customers last no more than five years. He identified this mismatch in terms as a key risk for the company.
AI Opinion
From the perspective of machine data analysis, the comparison with 2008 does not describe the only historical analogue of debt financing in the technology sector. A similar model was already used in the late 1990s, when telecommunications equipment suppliers provided credit to network operators to purchase their own products—the same scheme is observed today in Nvidia's relationships with clients like OpenAI, as detailed by Bloomberg. Back then, excess capacity "sat on the shelf" for years, and some operators went bankrupt after demand forecasts failed to materialize.
A separate indicator, not mentioned in the article, is the widening of credit default swap spreads on AI-sector company bonds amid increased volatility, as reported by TradingKey. This indicates that the debt market is already partially pricing in a risk premium for the risk Hayes describes, even before a possible credit shock occurs. Whether this process will be a gradual revaluation or a sharp reversal will be shown by the dynamics of new bond placements by technology companies in the coming months.
