Oracle's Default Insurance Hit a Record 227 Basis Points in September. Microsoft, Amazon, and Alphabet's Barely Moved.

Oracle's cost of insuring its own debt against default touched a record near 227 basis points in late September. That number was well under 100 basis points for most of the past two years, and the move higher has been sharp enough that Oracle's five-year credit default swaps now trade closer to a junk-rated company's than an investment-grade one's. Alphabet, Amazon, and Meta are running the same AI spending race and paying almost nothing extra for it. Their CDS spreads sit in a 57 to 78 basis point range, roughly where they've been for most of this year.

[[IMG1]]

What Changed

Five companies control most of the cloud infrastructure that AI runs on, and all five are spending at a pace none of them have matched before. Amazon spent $54.2 billion on capital expenditures last quarter. Alphabet spent $44.9 billion, up roughly 100% from a year earlier. Microsoft spent $35.8 billion. Meta spent $31.1 billion, up 83%. Oracle, the smallest of the five by revenue, spent $16.5 billion. Add them up and the group spent $182.5 billion in three months, a record for the industry.

For most of this year, bond investors treated that spending as one undifferentiated story: AI infrastructure is expensive, and the hyperscalers are all paying for it with debt. Oracle alone has raised bond deals large enough to push the yields on its existing debt wider with each new issue, and the five companies combined have sold more than $220 billion in bonds this year through early August, roughly double the pace of 2025.

The split showed up this summer. Oracle's five-year CDS spread, essentially the market's price for insuring against an Oracle default, climbed past 200 basis points in August and set a fresh record near 227 by late September. S&P did not downgrade Oracle to BBB-, one notch above junk, until July 9, and credit markets had already started pricing in the risk before that announcement. Alphabet, Amazon, and Meta didn't follow. Their spreads stayed inside a tight band, and in some cases are actually lower than where they started the year.

Why It Matters

The gap comes down to balance sheets, not ambition. Oracle's debt-to-equity ratio runs around 4 times. Microsoft, Amazon, Alphabet, and Meta are all somewhere between 0.18 and 0.51 times. S&P rates Microsoft AAA, Alphabet AA+, Amazon AA, and Meta AA-, four companies sitting comfortably in the upper tiers of corporate credit. Oracle sits at BBB- with a stable outlook of its own, but one notch above where investment grade ends entirely.

That matters because the AI buildout isn't pausing while credit markets sort out who's who. Broadcom already agreed to lend Anthropic the same $42 billion Anthropic lost last year, and SoftBank has taken on its highest-ever bond yields to keep funding OpenAI's compute needs. Oracle's exposure runs through similar commitments: S&P's own downgrade cited OpenAI as roughly half of Oracle's $638 billion backlog of remaining performance obligations, a concentration risk that shows up in Oracle's numbers in a way it does not for the other four. When a company with Oracle's leverage profile takes on that kind of obligation, bond investors price in the risk that the customer on the other end of the contract doesn't pay as expected. Microsoft, Amazon, and Alphabet are funding comparable buildouts largely out of operating cash flow and existing balance sheet strength, which helps explain why their spreads haven't budged.

This is also a lesson in how credit markets actually work. Nobody is arguing AI capex is overbuilt in aggregate, and no one is predicting a hyperscaler default this year. What changed is narrower than that: investors decided the five companies spending on the same technology at a similar pace are not equally capable of absorbing a slowdown, a rate shock, or a disappointing return on all that AI investment, and they're charging accordingly.

What to Watch Next

Oracle's next bond sale will be the clearest signal of where this goes. If the company has to pay up further to borrow, or scales back its own capex guidance, that tells the market the spread widening reflects a real repricing of risk rather than a temporary air pocket. Amazon's recent 20-year nuclear power deal is the kind of long-dated commitment that would normally draw scrutiny from credit analysts, and the fact that it didn't move Amazon's spreads at all says something about how much balance sheet cushion the market thinks Amazon has to work with.

Watch Microsoft's next earnings call too. The company with the tightest spreads and the only AAA rating in the group is also the one every other hyperscaler gets measured against going forward. If Microsoft's capex guidance keeps climbing without any credit market pushback, that sets the bar the other four will be judged by, and Oracle's situation starts to look less like an outlier and more like a preview of what happens to anyone whose balance sheet can't keep pace.

The Pulse24 Take

Five companies, one technology bet, five very different report cards from the bond market. That's the real story underneath the AI capex headlines this quarter. The spending totals get the attention because they're enormous and because they keep climbing, but the spending totals were never going to be what determined who gets hurt if AI returns disappoint.

Credit spreads are a cleaner signal than almost anything else available right now, because they represent real money being bet on real outcomes, not an analyst's price target or a CEO's guidance call. Bond investors have already told you which four of these five companies they'd lend to without much worry, and which one they're starting to treat differently. S&P's downgrade of Oracle to BBB- in July only caught up to where the CDS market had already been pricing Oracle's risk for weeks. Rating agencies tend to move slower than markets, and that gap is worth remembering the next time a stable outlook gets treated as the final word on a company's credit health.