S&P Warns: Hyperscale Cloud Providers Face Opaque AI Returns, Rating Headroom Narrows
nashnova research
S&P Global Ratings warns that the six U.S. hyperscalers will spend over $7 trillion in capex through 2030, yet none discloses quantified AI returns — rating cushions are thinning, and some companies could hit downgrade triggers as early as 2027.
$7 trillion in spending — where is it all going?
S&P projects that Microsoft, Amazon, Alphabet, Meta, Oracle, and one other hyperscaler will spend a combined $7 trillion-plus in capex from 2026 through 2030.
Over the past year, roughly half of U.S. private-sector economic growth was tied to AI-related activity. This means → AI is no longer a future story; it is already one of the main engines of U.S. economic growth.
Every time S&P digs deeper, it finds the same pattern: capex growth outpaces expectations, financing structures grow more complex, and transparency drops.
All that spending — what are the returns?
The critical gap: none of the six hyperscalers has quantified its AI investment returns.
S&P speculates the reason is either that the companies cannot calculate them, or that they "choose not to share the data."
In plain terms = the rating agency needs return figures to assess credit risk, but the companies simply will not provide them. S&P's fallback is to assume the world's largest firms "most likely know what they are doing" — and rate on that assumption.
This reflects a deep tension: the world's leading rating agency is relying on trust, not verification, for the single most important data point.
Whose rating cushion is thinnest — and thickest?
Microsoft (AAA): thinnest cushion — can absorb only about one year of EBITDA (earnings before interest, taxes, depreciation, and amortization — the cash a company generates from core operations) in incremental debt and lease commitments before a downgrade threat materializes.
Alphabet (AA+): thickest cushion, with the most headroom at its current rating.
Amazon (AA): if capex stays on its current trajectory and earnings do not keep pace, it could hit downgrade triggers around 2027. S&P stresses that the final call rests with the rating committee.
Why is Oracle in the most dangerous position?
Oracle's current rating is BBB- — the lowest rung of investment grade (the minimum credit quality that most institutional investors are permitted to hold).
A one-notch downgrade would push it straight into junk territory (high-yield). This means → index funds and insurers that are barred from holding junk-rated bonds would be forced to sell.
Oracle has roughly $117 billion in index-eligible U.S.-dollar bonds. In plain terms = the selling would not be driven by the market losing faith in Oracle — it would be rules forcing the sale — and that kind of mechanical selling often hits harder than a fundamental deterioration.
Beyond opaque returns, what other risks are hiding?
S&P flagged several risk categories: revolving financing structures, overbuild risk, and the potential threat of open-source models to commercial business models.
Some financing arrangements involve residual-value guarantees, take-or-pay agreements (contracts requiring payment whether or not a service is used), chip financing, and lease liabilities. This means → the true scale of off-balance-sheet commitments is hard to pin down, and reported figures may understate the real burden.
This reflects a structural problem across the industry: spending keeps rising, but the portion outsiders can actually see keeps shrinking.
When does this "trust bet" get called?
The central question: can the hyperscalers deliver returns on schedule against S&P's EBITDA growth benchmarks?
The verification window will open gradually with quarterly earnings over the next few years — not a single reveal, but report by report.
In plain terms = S&P's current ratings are essentially a "we'll take your word for it" call. If earnings reports over the next two to three years show returns lagging capex, downgrades will move from warning to action.
市场有风险，内容仅供研究参考，不构成投资建议。