S&P Global warns hyperscalers' hidden AI debt may be understated

On October 20, 2026, Mark Habib, head of digital infrastructure at S&P Global, said at a roundtable in London that the AI-related liabilities of the world's largest technology companies may be well above what has been disclosed, while the agency has no current plans to change its ratings. In a report that morning, Habib was cited as saying hyperscalers may carry substantial undisclosed AI-related debt, largely from implicit guarantees tied to data center construction or smaller AI companies. "It is reasonable to assume there is more that we cannot see," he said.

A later report the same day added detail. Habib highlighted what he called interconnected financing structures, in which large technology companies effectively backstop major data center builds or smaller AI firms, creating off-balance-sheet potential liabilities. If such liabilities are not reported, he said, they fall outside hyperscaler disclosures; if the resulting transactions sit outside S&P's rating scope, the agency may not see them either. He said the size of these contingent liabilities cannot yet be estimated. The report also noted that the biggest rating risk for hyperscale data center operators remains whether expected profits materialize.

Both reports said market sentiment on AI infrastructure financing has turned more cautious in recent weeks, with bond prices for AI infrastructure providers and so-called neocloud companies such as Nvidia-backed CoreWeave continuing to fall amid investor concern over aggressive borrowing. The second report also said S&P expects the top six US hyperscale data center operators to spend more than $7 trillion on AI-related capital expenditure between 2025 and 2030. The matter currently rests at the warning stage, with no rating action announced.