BusinessSubscribers only Jul 14, 2026 at 22:305Add to bookmarks

BBB-, the last step before the speculative category. S&P does not downgrade Oracle for its results, but for the structure of its bet: $90-95 billion in capex and a single client that accounts for half of the commitments.
A rating agency has just downgraded Oracle to the last rung before the "speculative" category. The reason is not poor performance: it's that the company is massively indebted to build AI infrastructure whose main counterparty is a single client, OpenAI.
S&P has moved Oracle from BBB to BBB-, with a stable outlook. One more notch down and the company exits the investment grade category - which mechanically increases its debt and closes access to part of institutional investors. For a company that must precisely finance a wall of capex, the rating is the business model.
The figures cited by S&P: a capex forecast of $90 to 95 billion for fiscal year 2027 - the previous forecast was $60 billion - and a free cash flow deficit expected around $42 billion for the same fiscal year. The agency explicitly points to the rise in the cost of GPUs and network equipment.
The real issue is customer concentration. Of $638 billion in services contractually promised but not yet delivered, about half is attributable to OpenAI. S&P bluntly qualifies OpenAI as a central credit risk: the client's ability to honor its commitments depends on its continued market dominance and its ability to raise external funds.
Let's translate into balance sheet terms. Oracle is borrowing today, at fixed rates, to build very specific assets, against future revenues half of which depend on a company that is not yet profitable and which finances itself on the markets. It's a chain of risk, not an order book. The order book only becomes cash if the next link holds.
In the background of the transformation: cloud infra accounted for 27% of revenue in fiscal year 2026, Oracle projects 60% in 2028. And 21,000 jobs have been cut in twelve months, or about 13% of the workforce.
The bet holds. OpenAI raises funds, honors, fills the data centers. Oracle becomes a full-fledged hyperscaler, the rating goes up, and today's downgrade will have been just a temporary financing cost.
The link breaks. A slowdown in OpenAI's financing, and Oracle finds itself with oversized assets to amortize against absent demand - with debt contracted at the top of the cycle.
The signal that matters is not the rating, it's the name. When an agency writes in black and white that a single client constitutes the central credit risk of a company carrying $638 billion in commitments, it describes a systemic exposure. The BRI has indeed warned about the parallels between AI investment financed by debt and the 2008 crisis - a comparison to be handled with caution, but which is no longer confined to contrarian blogs: it is in the regulators' documents.
Watch Oracle's next refinancing. That's where the market will say, in basis points, what it really thinks of the bet.
Create a free account to access all our content and the weekly review.
Article produced by artificial intelligence, reviewed under human editorial control.
Sign in to join the discussion.
Et si le pari d'Oracle sur l'IA portait ses fruits à long terme ? S&P aurait peut-être sous-estimé le potentiel.
Est-ce qu'on ne s'inquiète pas un peu trop vite pour Oracle ? Ils misent peut-être sur le long terme avec l'IA.
Est-ce qu'ils ont bien réfléchi aux conséquences morales de leur pari sur l'IA ?
Et si Oracle ne misait pas que sur l'IA ?
Et la synergie entre Oracle et OpenAI ? Leur collaboration ne pourrait-elle pas limiter les risques évoqués par S&P ?
Est-ce qu'on sous-estime le potentiel de l'IA pour révolutionner les secteurs ? Le pari d'Oracle pourrait bien payer, d'une manière qu'on ne voit pas encore.
Et si Oracle diversifiait ses clients ? S&P ne semble pas y avoir pensé.
La dette de l'IA : capex, notations et risque de contrepartie