Insuring CoreWeave's debt costs over $10 a year per $100, about 17 times the investment-grade index

A swap quoted at 100 bps costs $1 a year for every $100 of debt insured ^[11]. CoreWeave's swaps are above 1,000 bps, according to S&P Global Market Intelligence data ^[5], so protection costs more than $10 a year per $100 ^[17]. An investment-grade index trades around 60 bps ^[9], which makes CoreWeave about 17 times the index ^[18]. Oracle, at nearly 300 bps ^[7], is about five times it ^[19]. Nvidia at around 90 bps is 1.5 times ^[20] and Meta at around 104 bps is 1.7 times ^[21]. Spreads widen with leverage: the source says the cost has risen even faster for the more heavily indebted competitors, suppliers and customers of the big names ^[4].

That supports a narrow version of the claim. For Apple, Amazon and Microsoft, the article says demand for AI-linked protection has surged ^[3] but does not report where their spreads trade, so we cannot tell whether they sit near the 60 bps index or well above it. "Even for" the big three describes volume of demand. The record cost belongs to the leveraged names.

The equity half of the thesis is also thinner than it sounds. AI shares still rank among the year's biggest gainers, though many have dropped steeply in recent weeks as investors weigh what capital costs now that interest rates and inflation are rising ^[2]. Equities have not been hiding the credit signal for long.

The best argument against reading much into the record is size. Single-name CDS is a market of about $9 trillion, according to ISDA ^[12], against more than $150 trillion of debt securities outstanding, per BIS data ^[13], or under 6% ^[23]. Average daily trades even for large companies can be in the single digits, so small transactions can move prices ^[14], and the deals are over-the-counter, with no central clearing house ^[15]. A record quote there can rest on a handful of tickets. The spread-implied odds of a CoreWeave default, roughly 63% over five years against around 30% a year ago ^[6], are about double ^[22]. They are the same quote restated, and they inherit the thin-market problem.

In our view the evidence supports a claim about heavily indebted data-centre borrowers and does not yet reach the whole buildout. Three outcomes would change that. If Apple, Amazon and Microsoft spreads converge on the index while CoreWeave and Oracle stay wide, the stress is about a few balance sheets. If Nvidia and Meta keep widening from 1.5 and 1.7 times the index, the claim broadens. If shares recover and the spreads hold, credit is leading equities after all.

None of the companies whose swaps have risen has defaulted or signalled that it may ^[10].