Oracle's Credit Rating: What Moody's and S&P Said, and the 4.5x Junk Line

S&P Global Ratings cut Oracle to BBB- on 9 July 2026, one notch above junk. Moody's still rates the company Baa2, two notches above, but has carried a negative outlook since July 2025. Read side by side, the Oracle credit rating story turns on one number: 4.5 times leverage, the line S&P says would send Oracle lower.

The short answer

As of 25 October 2026, S&P rates Oracle BBB- with a stable outlook. Moody's rates it Baa2 with a negative outlook. Fitch rated it BBB with a stable outlook at its last reported action in February 2026. That puts Oracle one notch above speculative grade at S&P and two notches above at the other two agencies.

Oracle's credit rating today: three agencies, three distances from junk

A credit rating is an agency's opinion of how likely a borrower is to pay its debts on time. The line that matters most sits between BBB- at S&P and Fitch, or Baa3 at Moody's, and the next step down. Above it is investment grade. Below it is speculative grade, which the market calls junk.

Oracle is still investment grade at all three agencies. What differs is the distance to that line and the direction each agency is pointing. That difference is the most useful thing a reader can take from the three releases, because it shows where the agencies disagree about the same set of facts.

Notches counted to the last investment-grade step, BBB- or Baa3. The Fitch row is the most recent action this post could confirm; a later Fitch action may exist.

S&P: downgraded, then stabilised

S&P moved first and furthest. It had held a negative outlook on BBB since July 2025, and on 9 July 2026 it took the rating down one notch. The short-term and commercial paper ratings fell to A-3 from A-2 at the same time. The outlook reset to stable, which in plain terms means S&P does not expect to move again soon on its base case.

A stable outlook one notch above junk is a narrower position than it sounds. It means S&P has priced in the plan Oracle described. It does not mean S&P has room for the plan to get bigger.

Moody's: unchanged rating, persistent warning

Moody's has not cut Oracle. It revised its outlook to negative from stable in July 2025, and affirmed Baa2 with a negative outlook on 2 February 2026 around an equity announcement. It was reported to have affirmed again in September 2026. Three looks at the same company, one consistent message: the rating holds, the direction is down.

Fitch: the most patient of the three

Fitch rated Oracle BBB with a stable outlook when it rated new senior notes in February 2026. Reuters reported Fitch's thresholds in August 2026: sustained EBITDA leverage above 3.5 times, or above 4 times including rents, would trigger negative action. That is a tighter line than S&P's, which makes Fitch's stable outlook the one most worth checking at its next review.

Reading the S&P downgrade of Oracle line by line

The S&P action is the most detailed of the three public records, and it reads like a reconciliation. Every change in the rating traces to a number Oracle itself put out. That makes it the best single document for anyone trying to understand the Oracle credit rating.

What moved: capex up 54%, cash deficit up 75%

The trigger was guidance. Oracle told investors it expected fiscal 2027 capital expenditure of $90 billion to $95 billion. S&P had been working with $60 billion. At the midpoint, that is a 54% increase in the spending S&P had to fund in its model.

The cash consequence was larger in proportion. S&P now estimates a fiscal 2027 free operating cash flow deficit of nearly $42 billion, against about $24 billion before. That is a 75% widening, and it is after counting the working capital benefit of customer prepayments. Free operating cash flow is cash from operations minus capital spending, before dividends.

S&P also said its previous leverage estimate for fiscal 2027 had been 4.8 times, above the 4 times level it uses as the downgrade line for a BBB rating. Once that number crossed 4 times, the downgrade followed almost mechanically. The new BBB- rating carries a wider tolerance: 4.5 times.

The three conditions for a cut to junk

S&P names three things that could take Oracle below investment grade. They are worth reading as a checklist rather than as boilerplate, because each one is observable.

First, sustained adjusted leverage above 4.5 times. Adjusted means S&P adds lease and other obligations to reported debt before dividing by EBITDA, which is earnings before interest, tax, depreciation and amortisation. Second, a view that Oracle is not on a path to positive free operating cash flow by fiscal 2029. Third, a negative view of Oracle's AI strategy or of the wider AI industry.

The third condition is the unusual one. Agencies normally rate a company on its own metrics. Writing the health of an entire industry into a single issuer's downgrade trigger tells you how S&P sees Oracle. It reads Oracle as a debt-funded proxy for AI demand, not as a software company that also builds data centres.

S&P analyst Andrew Chang told Reuters that Oracle's leverage was 3.6 times in May and could peak at 4.4 times in the next two fiscal years. His base case does not cross 4.5. The gap between the peak and the trigger is 0.1 times. My reading is that this is not a buffer. It is a rounding difference.

S&P was also blunt about upside. An upgrade is unlikely over the next two years, and would need consistently positive cash flow with leverage brought toward the mid-3 times area. The asymmetry is the point: the path down is three conditions wide, and the path up is two years away.

What Moody's said about Oracle, and what it has not said

Moody's has published less that can be quoted, and what it has published is more qualitative. That is itself informative. The agency has kept its rating but used unusually direct language about the shape of the risk.

In July 2025 Moody's cited "continuing elevated leverage and increasingly negative free cash flow" as the reason for the negative outlook. At that point adjusted debt to EBITDA already exceeded 4 times and trailing free cash flow was negative $5.1 billion. Senior credit officer Matthew Jones said capex would likely drive significant negative free cash flow.

In a September 2025 report following the OpenAI contract, Moody's described Oracle's data centre build as effectively one of the largest project financings in the world. It warned that counterparty risk is a key consideration in any project financing, particularly with high reliance on a single counterparty. Moody's took no rating action in that report.

The February 2026 note, quoted by Reuters, said leverage "will increase materially in the near term and could approach 5x temporarily". Read against S&P's 4.5 times line, that is a striking sentence. Moody's is openly describing a level that would breach S&P's junk trigger, and still holding Baa2.

What Moody's has not done, in any source this post could open, is publish a numeric downgrade trigger for Oracle. Several aggregator summaries attribute a 4.5 times trigger to Moody's. Every one I traced led back to the S&P release instead. Until the Moody's rating action text is quoted directly, treat any Moody's number other than "could approach 5x" as unconfirmed.

My view is that Moody's word "temporarily" carries the whole rating. If leverage approaches 5 times and then falls as contracted revenue arrives, Baa2 survives. If the fall slips by a year, the negative outlook becomes a downgrade. The rating is a bet on timing, which is precisely the variable Oracle's management controls least.

The lease adjustment that changes the size of Oracle's debt

Oracle's 10-K for the year ended 31 May 2026 reports $129.5 billion of outstanding indebtedness, maturing between 2026 and 2066. That is the number most coverage quotes. It is not the number the rating agencies use.

S&P's analysis counts $260 billion of lease commitments expected to begin between fiscal 2027 and fiscal 2029, plus $13 billion of unconditional purchase obligations, mostly for data centre power. These are leases Oracle has signed for facilities that do not yet exist, so under current accounting they sit outside the balance sheet.

The ratio is simple and it is the original figure in this post. The uncommenced leases equal 2.0 times Oracle's reported debt. Add them to it and the nominal total is about $390 billion, roughly three times the headline figure. Agencies capitalise leases at a present value rather than at face value, so their adjusted debt will be lower than this sum. The direction is not in doubt.

The wider pattern across the hyperscalers is in the analysis of data centre leases still off the balance sheet. Oracle's version is extreme because its leases are large relative to its own cash generation, not because the accounting is unusual.

The risk factor that links the rating to the leases

Oracle's own 10-K says a downgrade could "affect the terms or availability of certain long-term commitments (including data center leases)". It also warns that a downgrade could reduce access to commercial paper or raise its cost. Reuters reported that funding on a $14 billion Michigan data centre debt deal is linked to Oracle's rating.

That creates a loop worth naming. The leases push leverage up, leverage pressures the rating, and the rating governs the cost and availability of the leases. In my reading, this is the single most important sentence in Oracle's filing for anyone pricing its credit, and it gets almost no coverage.

How Oracle is defending the rating with equity

A company that wants to keep investment grade while spending more than it earns has three levers: cut spending, borrow, or sell shares. Oracle has chosen the third, deliberately and in size.

In February 2026 Oracle issued a $5 billion mandatory convertible preferred, which agencies treat partly as equity. On its September 2026 earnings call, Oracle said it had completed its $20 billion at-the-market equity programme during the first quarter of fiscal 2027. S&P's July analysis had already credited both.

Set the $20 billion against S&P's $42 billion deficit estimate and the share sale covers 48% of one year's gap. The remaining $22 billion has to come from debt, more equity, or a smaller build. Every dollar of it that comes from debt moves leverage toward the 4.5 times line.

There is a cost to this route that the rating does not capture. Shareholders absorb dilution so bondholders keep an investment-grade issuer. That trade is rational for Oracle, because a junk rating would raise the price of the leases and the commercial paper at once. It is still a transfer, and it is worth seeing clearly. A related question about who carries data centre risk when it is moved off a balance sheet is covered in the Meta Hyperion bond and its risk transfer.

Oracle's credit rating vs what the CDS market charges

Ratings move slowly by design. Credit default swaps, which are contracts that pay out if a borrower defaults, reprice every day. When the two disagree, the gap tells you how much the market doubts the agencies' base case.

In early August 2026, Reuters reported Oracle's five-year CDS at an 18-year high of 215 basis points, against about 80 for peers. By 20 October 2026 it was 295 basis points, after touching 300 days earlier. That is another 80 basis points in about 11 weeks, and roughly 3.7 times the August peer level.

Gavan Nolan of S&P Global Market Intelligence said that range is "more in line with BB and B credit." BB is the first rung of junk. So the market is already charging Oracle at least one notch below where its ratings sit, and possibly several. Oracle's ten-year bonds yielded about 6.5% in July, according to Bloomberg data cited by TNW.

Oracle is not the only AI infrastructure name where this gap shows. CoreWeave's debt stack trades at CDS spreads several times wider again. The wider pool of high-yield data centre bonds shows how the market prices the same assets once the investment-grade label is gone. The private lenders behind much of this are mapped in the breakdown of AI debt and private credit exposure.

Where this reading of Oracle's credit is weakest

The case above leans toward a downgrade. Here is where it could be wrong.

S&P's own base case says Oracle stays investment grade

The agency that has studied the numbers most closely set a stable outlook. Its base case has the cash deficit peaking in fiscal 2027 and free operating cash flow turning positive by 2029. A peak of 4.4 times sits under the 4.5 line. If you accept S&P's model, the downgrade I am describing does not happen. My disagreement is with the margin of error, not with the model.

The backlog is real and partly prepaid

Oracle reported cloud revenue growth of 93% in the quarter before the downgrade. TNW reported $75 billion of prepaid and customer-supplied hardware tied to large AI contracts. Prepayments shift cash risk onto customers. Oracle's net cash capex guidance of no more than $70 billion, against reported capex of up to $95 billion, reflects exactly that. A credit analysis that ignores prepayments overstates the gap. How much of the backlog converts, and when, is examined in the lender's reading of Oracle's RPO.

The source base is mostly secondary

The full S&P and Moody's rating reports sit behind subscription walls. This post reads them through Reuters, trade press and earnings summaries, and quotes only what those outlets attributed directly. Two Moody's actions, the September 2025 report and the reported September 2026 affirmation, each rest on a single secondary source. Agency wording matters at this level of detail, and paraphrase can lose a qualifier.

Four numbers that would move Oracle's rating next

Each agency has told you what it is watching. You can track the same things from public filings, quarter by quarter.

The fourth row needs a note. S&P said about half of Oracle's $638 billion of remaining performance obligations at the time came from OpenAI. If that customer could not pay, S&P warned, Oracle "could be left with massive data center leases that it might be unable to exit or would have to re-lease". That sentence is the downside case in full. Oracle's cash flow picture after its cost cuts is set out in the Oracle layoffs and AI cash flow math.

Frequently asked questions

Where to start if your business depends on Oracle

If you buy Oracle Cloud Infrastructure on a multi-year commitment, or hold its bonds, one task this week is worth more than reading another analyst note.

Open Oracle's next 10-Q, due after its second fiscal quarter closes on 30 November, and write down three numbers in a sheet: notes payable, uncommenced lease commitments, and trailing free cash flow. Divide the first two by the last four quarters of operating income plus depreciation. If that rough ratio rises while capex guidance rises too, you are watching the 4.5 times line approach before the agencies say so.

For contract owners, check whether your agreement with Oracle has any clause tied to the vendor's credit standing or service continuity. Most do not. If yours does not, raise it at the next renewal, not after a downgrade.

Related reading