Arthur Hayes has warned that a pullback in spending by leading U.S. artificial intelligence labs could expose a chain of debt and insurance-market risks built around the rapid expansion of data centers, chips, and cloud-computing capacity.
In Hayes’ account, the immediate trigger would be a shift in AI companies’ priorities. He said major labs have slowed work toward artificial general intelligence while emphasizing a “safety first” approach, even as buyers of AI services resist prevailing U.S. price levels. Hayes characterized the mismatch as demand that exists but is seeking “China pricing,” which he estimated at roughly 1% of U.S. prices.
If labs respond by focusing on more efficient models rather than ever-larger training runs, they could buy less computing power. That reduction would reach far beyond the companies operating the models, Hayes argued, because projected demand for compute has become a foundation for an extensive financing structure behind data centers and semiconductor procurement.
Compute spending supports a large debt stack
Hayes estimated that AI-related compute demand supports more than $1 trillion in investment-grade debt, along with several hundred billion dollars in lower-rated loans. His argument places particular attention on debt used to finance data-center construction, equipment leases, and chip purchases.
The structure depends heavily on expectations of continued revenue growth from AI services. Hayes wrote that many AI labs have yet to generate profits and rely on financially stronger technology companies to support obligations that sit outside their own balance sheets, including data-center leases and financing linked to hardware purchases.
A fall in compute orders would not automatically lead to defaults. Debt obligations would remain in place, while the assets and expected cash flows supporting them could be revalued lower. That sequence could put pressure first on the market prices of AI-linked credit, especially debt held by leveraged buyers or packaged into structured products.
Hayes compared the risk to a credit repricing rather than an immediate collapse. A downgrade of debt tied to AI infrastructure could reduce the value of securities before borrowers miss payments, creating capital strains for institutions required to mark holdings more conservatively or maintain regulatory buffers.
Insurance portfolios could carry indirect exposure
Hayes’ more controversial claim concerns the route through which AI infrastructure debt could affect U.S. insurance policyholders. He argued that private credit and structured debt linked to technology infrastructure can enter insurance portfolios through captive insurers and affiliated reinsurance arrangements.
He cited analysis by Substack writer Nick Nameth, who has written about the use of affiliated reinsurers to obtain capital relief. According to Hayes’ summary of Nameth’s work, “phantom” capital buffers associated with affiliated reinsurance arrangements could total as much as $1.54 trillion.
Hayes also pointed to an example cited by Nameth involving Brookfield-related reinsurance assets. In that example, assets were recorded at $1.48 billion while a reporting entity listed an associated claim liability as zero. The example was used to illustrate Hayes’ concern that internal reinsurance structures can make an insurer appear better capitalized than it would be if the obligations were backed by cash or independently valued assets.
The argument depends on whether the assets held by insurers retain their assumed credit quality. Hayes wrote that a downgrade of securitized investment-grade data-center debt could force insurers holding senior tranches to post additional capital. Affiliated reinsurers without readily available cash, he said, could struggle to meet those demands.
That would place insurance balance sheets under pressure even if the underlying AI assets did not fail all at once. Insurers commonly hold long-duration assets to meet long-term policy obligations, making sudden losses in highly rated credit particularly problematic when capital requirements rise at the same time.
Policyholders face limits on guarantee coverage
Hayes also focused on the gap between large insurance policies and state guarantee protections. He noted that insurance guarantee coverage in most U.S. states is generally capped at roughly $250,000 to $300,000, leaving holders of policies with multimillion-dollar promised benefits exposed beyond those limits if an insurer fails.
Unlike the Federal Deposit Insurance Corp., which maintains a pre-funded insurance fund for bank deposits, state insurance guaranty mechanisms generally assess surviving insurers after a failure. Hayes argued that such a system could face a more difficult test if stress hit several major insurers or reinsurers simultaneously.
His warning echoes concerns raised during the 2008 financial crisis, when American International Group became a central point of failure because of its exposure to credit derivatives tied to lower-quality mortgage debt. The subsequent federal response included support through the Troubled Asset Relief Program and other emergency facilities.
Hayes expects any comparable response to AI-credit stress to focus on preventing a broad recognition of insurance-sector losses. He outlined two possible channels: federal procurement of compute capacity under a national-security framework, or monetary support aimed at stabilizing insurers holding impaired AI-related assets.
Federal purchases of computing capacity could provide revenues to data-center operators and support debt-service assumptions. Liquidity measures directed at insurers, meanwhile, could ease the pressure created by falling asset values and capital requirements. Both approaches would rely on additional government borrowing, central-bank liquidity, or banking-system balance-sheet expansion.
Hayes links the scenario to monetary conditions
Hayes framed the AI-credit warning against what he described as already accommodative financial conditions. He cited nominal year-over-year U.S. GDP growth of 6.6% in June 2026, against an effective federal funds rate of about 3.6%.
He also wrote that the Federal Reserve raised rates by 0.25 percentage points at its latest policy meeting, while the RMP short-term Treasury purchase program ended on Aug. 14. In Hayes’ view, commercial banks subsequently expanded their balance sheets, creating hundreds of billions of dollars in new money.
Hayes estimated that the latest quarter-point rate increase would add $7.5 billion annually in interest paid on excess reserves. He argued that higher interest income could support more lending and risk-taking by banks, though the eventual effect would depend on credit demand, bank capital, and lending standards.
The broader case rests on a fragile connection: lower AI compute spending could weaken the credit supporting data centers, and losses in that credit could reach insurance portfolios through structured-finance and reinsurance channels. Hayes’ scenario is far from a forecast of imminent insurer failures, but it identifies a pressure point in the AI buildout that extends beyond technology-company valuations and into the balance sheets backing long-term insurance promises.
Concerned about AI-driven financial risks? Explore how on-chain private credit could reshape exposure and systemic stability.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
