S&P Global flags potentially hidden AI debt at hyperscalers as AI spending surges

Hidden AI debt means financial obligations tied to data centres or smaller AI companies that are not clearly shown on a hyperscaler’s balance sheet. The risk matters because investors and credit-rating agencies may underestimate the company’s total exposure.

The exposure can arise through complex financing arrangements. A hyperscaler might support a project, partner with an AI provider, or become responsible for obligations through an unrated transaction. If the arrangement is not reported, S&P may lack visibility into the potential liability. Mark Habib said, “there is more debt out there that we are not seeing.”

S&P has not estimated how much extra debt these structures could add. For now, it is taking a holistic view rather than changing ratings. The issue could become more serious if AI spending rises further or competition makes expected revenue and profits less certain.

Hyperscalers are the very large technology companies building and operating extensive computing infrastructure for digital services and AI. In this article, the group consists of Amazon, Microsoft, Alphabet, Oracle, Meta and SpaceX.

These companies matter because their AI plans involve enormous capital spending. S&P expects the six to spend more than $7 trillion on data centres and other AI-related capital expenditure between 2025 and 2030. Such spending can require borrowing, partnerships and other financing arrangements, creating questions about debt and future cash flow.

Their credit strength therefore matters to investors and rating agencies. S&P says their free cash flow is expected to remain negative until at least next year. Ratings will depend especially on whether these investments generate enough revenue and profit, and whether the scale of spending begins to look unsustainable.

S&P Global expects Amazon, Microsoft, Alphabet, Oracle, Meta and SpaceX to spend more than $7 trillion on data centres and other AI-related capital projects from 2025 through 2030. This is the article’s central measure of the AI infrastructure buildout.

The spending is directed toward the physical and technical capacity needed to support AI. It also helps explain why companies may use borrowing, partnerships or complex financing structures. Larger investment plans can increase financial exposure before the resulting revenue and profits arrive.

The scale has consequences for cash flow and credit analysis. S&P expects the six companies’ free cash flow to remain negative until at least next year. The agency’s larger concern is whether the investments produce sufficient returns. Ratings could face pressure if spending grows further or competition makes revenue harder to achieve.

A financing structure can connect a company to a project’s debt without placing every obligation directly in its ordinary balance-sheet debt. This can happen when companies support partners, guarantee payments, lease infrastructure, or arrange funding through separate entities. The company’s economic exposure may therefore be greater than its headline borrowing suggests.

For example, a hyperscaler could help finance a data-centre project or support a smaller AI company. If that project struggles, contractual commitments or guarantees could leave the hyperscaler responsible for payments. If the transaction is not reported or rated, outside observers may not see the potential liability immediately.

The article does not quantify this exposure. Mark Habib said S&P was trying to assess the risks holistically because it might lack visibility into unrated transactions. The concern grows as AI borrowing and infrastructure spending increase.

AI companies’ bonds could lose value if investors conclude that expensive infrastructure will not generate enough revenue and profit. Lower bond prices usually signal that investors view the borrower as riskier. New borrowing may then become more expensive, making it harder to fund additional data centres or AI projects.

Credit ratings could also be downgraded if weaker expected profits damage a company’s ability to manage its obligations. S&P downgraded Oracle to BBB in July, citing AI spending and an uncertain path to profitability. CoreWeave’s debt has also faced pressure as investors assess whether aggressive borrowing can produce sufficient returns.

For now, S&P does not see a major disconnect requiring ratings changes. Its ratings take a longer-term view. That stance could change if investment rises further or intense competition reduces expected revenue growth.

Free cash flow shows how much cash remains after a company’s operations and capital spending. Negative free cash flow means the company is spending more cash than it generates during that period. That matters here because AI infrastructure requires huge investment, potentially increasing the need for borrowing or other financing.

An accounting loss is different. It is calculated under accounting rules and can include non-cash expenses, such as depreciation. A company can report a loss while still generating cash, or report an accounting profit while using cash heavily for new data centres. The two measures therefore answer different questions.

The article says the six hyperscalers’ free cash flow is expected to remain negative until at least next year. This makes the speed of converting AI investment into revenue and profit especially important for credit ratings.

A credit rating is an agency’s assessment of a borrower’s ability to meet its debt obligations. Higher ratings generally indicate lower perceived default risk. Rating agencies consider factors such as cash flow, debt, profitability and business risks when forming that assessment.

Investment-grade status usually helps a company borrow at lower interest rates because investors view its debt as relatively safer. It can also widen the pool of institutions allowed to buy the bonds and improve access to financing. A downgrade can raise interest costs and make funding more difficult, although the exact effect depends on markets.

The article identifies BBB as Oracle’s current rating after S&P’s July downgrade, calling it the lowest rung of investment grade. S&P has not changed the other ratings for now. It is watching profitability, spending levels and competition over the longer term.