Credit default swaps are telling the market that AI infrastructure debt is no longer priced as one block. A CDS is an insurance contract on a company's bonds: the buyer pays an annual premium, quoted in basis points of the amount insured, and collects if the borrower defaults. The wider the spread, the more the market charges to insure that debt — and through 2026, the swap market has pulled Oracle and CoreWeave sharply apart from the hyperscalers financing data centers out of cash flow.
For investors, that split is the useful signal. It is a daily, tradeable read on how professional credit desks rank the borrowers behind the AI buildout, and it moves faster than a quarterly earnings call. It also has real limits — thin markets, and whole categories of debt that never show up in a swap price at all.
Table of Contents
- What the spreads actually say right now
- Why the swap market moved on Oracle
- How a CDS price becomes a real borrowing cost
- Where the signal is weakest
- Using credit signals in an equity decision
- Frequently Asked Questions
What the spreads actually say right now
Oracle's five-year CDS climbed above 215 basis points in mid-2026, up from roughly 145bp at the end of 2025. According to an Investing.com analysis of the move, that is a record for the company and wider than where it traded in the Great Financial Crisis. At 215bp, insuring $10 million of Oracle debt costs about $215,000 a year. Standard pricing models translate that spread into roughly an 8% chance of default within five years, as Real Investment Advice laid out.
That is still investment-grade territory. It reads as repricing — the market demanding more compensation for the same name — rather than a forecast of distress. CoreWeave sits in a different regime entirely. Its CDS blew out to about 855 basis points in late July 2026, implying a roughly 50% five-year default probability, according to Tech Times. A near-coin-flip implied default rate is how the market distinguishes a leveraged neocloud — a GPU rental operator financing its fleet with debt — from an established software company that happens to be building data centers.
Why the swap market moved on Oracle
The leverage behind the repricing is documented in Oracle's own filings. Total debt reached about $134.6 billion by fiscal Q3 2026, up from roughly $92.6 billion nine months earlier, with capital expenditure of $39.2 billion over the first nine months against $12.1 billion a year prior — figures from Oracle's Form 10-Q for the period ended 28 February 2026. That is a balance sheet reshaped in three quarters. Oracle funded the expansion with $43 billion of debt and $5 billion of equity in fiscal 2026.
The company has said it does not expect further debt issuance in calendar 2026 and guided to roughly $40 billion of mixed debt and equity in fiscal 2027, including a $20 billion at-the-market equity program. That last detail is the one the swap market cares about. Spreads are not a one-way alarm: Oracle's CDS fell sharply in early February 2026 right after it published that financing plan, as CNBC reported at the time. The market was reacting to funding structure, not just to the size of the capex number — equity in the mix lowers the risk to bondholders, and the spread said so within a day.
How a CDS price becomes a real borrowing cost
The swap market is not a scoreboard that companies can ignore. CoreWeave had to widen the spread on a $2.6 billion leveraged loan by 100 to 125 basis points and accept maintenance covenants — ongoing financial tests it must keep passing — to get the deal done, per Tech Times reporting on the loan. That is the transmission mechanism in a single deal. Lenders looked at where the swap market priced the credit and converted it into contract terms: a higher coupon plus tighter legal constraints.
Both reduce the cash available for growth. For equity holders, this is where credit signals stop being abstract. A borrower whose CDS has widened faces higher interest expense on new debt, covenants that can force asset sales or equity raises, and a shorter runway for capex-funded expansion. Watching the swap spread is a way of seeing next year's financing terms early.
Where the signal is weakest
Volume matters as much as price. DTCC data show Oracle's gross notional CDS outstanding doubled in three months to about $14.1 billion, with roughly $75 million in average daily notional in Q3 — up about 650% year on year — while CoreWeave's gross notional is only about $271 million, according to IFR's account of these new CDS markets. That asymmetry is the caveat on the headline number. CoreWeave's market is thin — hundreds of millions against Oracle's billions — so a handful of hedging or speculative trades can move the implied default probability far more than any change in the business warrants.
A 50% implied default rate drawn from a thin book is a weaker fact than an 8% rate drawn from a deep one. The second blind spot is structural. CDS only prices the referenced entity's own obligations. Borrowers who fund data centers through off-balance-sheet joint ventures and special purpose vehicles carry AI leverage that never appears in the swap price — a gap venture investor Tomasz Tunguz has written about. An absent or calm CDS is not evidence of a clean balance sheet.
Using credit signals in an equity decision
CDS data is not on a retail brokerage screen, but the conclusions it produces are checkable against filings. A practical sequence: The one thing the current pricing does not support is treating "AI borrower" as a single category. At 215 basis points Oracle remains investment grade with a stated plan to lean on equity in fiscal 2027; at 855 basis points, in a market a fiftieth the size, CoreWeave is priced as a distinct and far more speculative credit.
- Separate the funding models first. Cash-flow funders, on-balance-sheet debt funders like Oracle and CoreWeave, and SPV/JV funders are three different risk categories, and only the middle one is fully visible in swap pricing.
- Read the direction, not just the level. Oracle's February drop after a financing announcement and its mid-year widening are both information; a single snapshot is not.
- Discount thinly traded names. Weight an implied default probability by the notional behind it before treating it as a market consensus.
- Go to the 10-Q for the underlying numbers. Total debt, capex, and the debt-versus-equity split in new funding are disclosed quarterly and do not require a credit terminal.
- Watch new loan terms as confirmation. Covenants and spread concessions in a fresh financing are the swap market's verdict showing up in cash.
Frequently Asked Questions
Does a widening CDS spread mean a company is about to default?
No. Oracle's record 215bp spread implies roughly an 8% five-year default probability, which is investment-grade. Widening means the market wants more compensation to hold the risk, not that failure is expected.
Why is CoreWeave's implied default probability so much higher than Oracle's?
CoreWeave is a leveraged GPU operator financing its fleet with debt rather than cash flow, so the market ranks it as a far weaker credit. Its swap market is also small — about $271 million gross notional — so the figure is less reliable than Oracle's.
Can retail investors trade credit default swaps?
No. CDS contracts trade between institutions in the over-the-counter market. Retail investors use the published spread levels as a read on credit sentiment and then verify the underlying leverage in SEC filings.