Could an AI debt crisis end up benefiting bitcoin? Arthur Hayes, co-founder of BitMEX, believes that the “Safety First” discourse adopted by several players in the sector also masks questions about the demand for their services. He describes a possible sequence: slowdown in spending, deterioration of data center financing, public intervention then return of liquidity to risky assets. This reading, however, remains a personal hypothesis, and not an established forecast. But we're used to it with Mr. Hayes.
Key Points
Arthur Hayes suspects AI companies are using security to justify a possible slowdown in their investments.
He mentions more than $1,000 billion of investment-grade debt exposed, without publishing a detailed calculation.
Oracle, Meta and TeraWulf illustrate the growing use of debt to finance data centers.
According to him, a credit crisis would first cause bitcoin to fall before triggering monetary intervention that could support it.
Behind “Safety First”, Arthur Hayes suspects a slowdown in AI
The concerns expressed around the security of artificial intelligence are very real. Several industry leaders have recently called for slowing down certain research or strengthening controls. However, there is nothing to establish that these announcements mainly serve to hide a drop in demand. This is Arthur Hayes' interpretation.
His reasoning is based on the gap between investments made and income generated. Microsoft, Alphabet, Amazon, and Meta spent about $380 billion on capital spending in 2025. A large portion was on data centers and AI, but not all of that was spending on artificial intelligence.
He also cites a study from MIT's NANDA project, which found that about 95% of the generative AI pilots studied were yet to produce a measurable effect on the bottom line. This figure does not mean that 95% of projects have definitively failed: the researchers mainly highlight problems of integration, organization and adaptation to businesses.
Finally, skeptics are watching deals between capital providers and chip vendors. Nvidia has announced plans to gradually invest up to $100 billion in OpenAI, depending on the deployment of new infrastructure using its own systems. The operation fuels criticism of the circular nature of certain financing, without however constituting proof of manipulation of orders.
Data center debt focuses concerns
The need for financing actually constitutes a more concrete risk. Morgan Stanley estimated as early as 2025 that the global development of AI infrastructures could leave an external financing gap close to $1.5 trillion by 2028. Global debt issuance linked to AI could, for their part, approach $570 billion in 2026 alone.
Several operations illustrate this evolution. Oracle has placed $18 billion in bonds in 2025. This issuance should not be confused with the approximately $18 billion in loans related to the Jupiter data center project, some of which are now trading at a discount.
Meta, for its part, created a joint venture with Blue Owl to finance the Hyperion campus in Louisiana. The debt is mainly in an unconsolidated structure, but Meta publishes the existence of this agreement and maintains economic commitments. Talking about a debt completely hidden off balance sheet therefore seems excessive.
TeraWulf, which came from bitcoin mining and is now very involved in computer hosting, has also placed more than $3 billion in guaranteed debt to finance its infrastructure. The company, however, retains a mining activity: it is diversifying towards AI rather than abandoning it entirely.
Arthur Hayes places insurers and their reinsurance subsidiaries above all at the center of his scenario. It includes an estimate of $1,540 billion in reinsurance between affiliated entities. This amount, however, represents neither direct exposure to AI nor potential loss: it measures a much broader set of risk transfers within the insurance sector.
Bitcoin: An initial decline before a possible return of liquidity
In Mister Hayes' scenario, a downgrading of debt linked to AI would first cause a decline in the markets. Bitcoin would decline along with tech stocks and other risky assets, while companies combining mining, data centers and debt would come under particular pressure.
An intervention by the Federal Reserve or the Treasury could then change the situation. Our expert of the day draws on March 2020, when bitcoin fell sharply before benefiting from the Fed's asset purchases, as well as the banking crisis of March 2023. The BTFP (Bank Term Funding Program) program launched after the bankruptcy of Silicon Valley Bank, however, constituted a counter-guaranteed loan mechanism, and not a classic quantitative easing program.
