S&P Global Says Hyperscalers Likely Carry More AI-Linked Debt Than They Disclose, Sees No Rating Change Yet

Since last week's warnings from the IMF and Bank of England on leverage, and Morgan Stanley's projection of weak hyperscaler free cash flow, a fourth voice has weighed in: S&P Global said today, Oct. 19, that the biggest tech companies likely owe more for AI than their filings show.

Mark Habib, S&P's Global Head for Digital Infrastructure, made the comments at a roundtable in London. "It is safe to assume that there is more (debt) out there that we are not seeing," he said.

His explanation was about disclosure. "If it's not reported, then it's not part of the disclosure by the hyperscalers," Habib said. "Also, if we are not rating the transactions where this liability may accrue, then the fact of the matter is that we might not have the visibility."

What S&P is worried about

Habib pointed to what he called "interconnected financing structures." In these, a hyperscaler effectively guarantees a huge data center build, or backs a smaller AI company.

S&P has no number for it. Habib said the agency has no estimate of how much extra debt these circular arrangements could add to balance sheets. He said S&P is trying to take a "holistic" view.

The scale of spending around it is large. S&P expects the top six U.S. hyperscalers, which it lists as Amazon, Microsoft, Alphabet, Oracle, Meta and SpaceX, to spend more than $7 trillion on data centers and other AI capital expenditure between 2025 and 2030. It expects their free cash flow to stay negative until at least next year.

In July, S&P cut Oracle to BBB-, the lowest rung of investment grade. It cited the company's AI spending and what it called "an uncertain path to profitability."

Ratings stay put, for now

Habib said the biggest risk to hyperscaler ratings is that expected profits fail to show up. Asked whether the recent selloff in AI-linked bonds means investors see more risk than the ratings imply, he said no.

"For now, we don't have the view that there is a major disconnect that needs to force our ratings to change at this point," he said. S&P's scores, he added, are built to take a longer-term view.

What would change that? Habib said evidence that the spending is no longer sustainable. That could come from "further increases in the investments" by AI firms, or "a view that the revenue rate of competition is getting too intense."

The IMF's version of the same concern

S&P is not alone in flagging circular deals. In an analysis published Oct. 12, the IMF reviewed 14 such transactions. Since 2023 they involve about $300 billion in equity investment and $650 billion in debt, in exchange for $1.4 trillion in future revenue commitments.

Jason Wu, an assistant director in the IMF's Global Markets Analysis Division, told a press conference in Bangkok that if there are questions about AI monetization, "the equity investments will deplete in value, the debt may be difficult to service, and the revenue commitments in the future may not materialize."

In its World Economic Outlook released Oct. 13, the IMF forecast global growth of 3.1% in 2026 and 3.4% in 2027. One downside scenario, a steep drop in AI investment, would cut global GDP by 1.1% by 2028, including 1.8% in the U.S.

Managing director Kristalina Georgieva put it this way: "We now have an economy that is standing on one leg, the AI leg."

Markets already moving

The debt questions arrive after a rough week. The Nasdaq Composite fell 2.14% last Friday, Oct. 16, for a weekly loss of roughly 4.5%, its worst week since June. The S&P 500 lost almost 2% on the week. Both indexes had set record highs 10 days earlier.

The pressure is most visible in neoclouds, which rent computing power to AI companies. CoreWeave, backed by Nvidia and public since its March 2025 IPO, carries tens of billions of dollars of debt. Its bond yields recently climbed to levels that signal distress. Its stock fell almost 12% Friday, lost about a quarter of its value on the week, and is down more than 65% from its post-IPO peak.

Gil Luria, head of technology research at D.A. Davidson, called neoclouds "the subprime tranche of the AI trade." He said they are selling off first because it is unclear CoreWeave will earn enough to repay what it borrowed. Luria also said he believes AI demand is real and growing.

Nvidia CEO Jensen Huang told Bloomberg on Thursday that he is optimistic about CoreWeave. The stock rallied that day and gave the gain back Friday.

A chip-stock index fell 3.8% Friday, its eighth straight daily decline, and dropped 11.4% on the week. Brent crude touched $110 a barrel for the first time since May, feeding worries about inflation and further Fed rate hikes.

What remains open

Habib's argument cuts two ways. S&P sees no mismatch between market stress and its ratings, yet it also says part of the picture sits outside what hyperscalers report and outside what S&P rates.

The companies named have not disclosed an aggregate figure for guarantees tied to data center builds or smaller AI firms. Until they do, or until S&P rates the underlying transactions, the size of the unseen debt stays unknown, and so does the point at which S&P would revisit its ratings.