S&P Global: AI-Related Debt of Hyperscale Cloud Providers May Be Underestimated
nashnova research
S&P Global warns that the six largest hyperscalers carry AI-linked liabilities likely exceeding disclosed levels — capex will top $7 trillion from 2025 to 2030 — but says no rating changes are needed yet.
What debt is invisible?
When hyperscalers — Amazon, Microsoft, and peers — build data centers, they often guarantee debt for smaller AI firms or large projects.
These guarantees are contingent liabilities — obligations that only hit the balance sheet if the borrower defaults — and do not always appear in public filings.
S&P's Mark Habib, global head of digital infrastructure, put it bluntly: "You can reasonably assume there is more out there that we can't see."
This means → the balance sheets investors rely on may show only the tip of the iceberg.
Where does $7 trillion go?
S&P projects that Amazon, Microsoft, Alphabet, Oracle, Meta, and SpaceX will spend a combined $7 trillion-plus on data centers and AI-related capex between 2025 and 2030.
The spending is so large that free cash flow will stay negative through at least next year.
In plain terms = these companies are spending far more on AI than they earn in the short term, and they must keep borrowing to cover the gap.
Will ratings be cut?
Habib stated clearly: "We don't see a material disconnect that would force a rating change right now."
He added two caveats — the call could shift if AI spending proves unsustainable or if revenue competition becomes too fierce.
This reflects an S&P stance of "watch but don't act": the risk is acknowledged, but the trigger has not been pulled.
What does Oracle's downgrade signal?
S&P already cut Oracle to BBB- — the lowest investment-grade notch — in July, citing aggressive AI spending and uncertain profitability.
This means → S&P is willing to act; it simply scores each company separately. Oracle is the only hyperscaler downgraded so far.
For the other five, Oracle serves as a mirror: if hidden liabilities keep growing while earnings lag, a downgrade could come at any time.
市场有风险，内容仅供研究参考，不构成投资建议。