CoreWeave's 9.625% Bond Now Yields 14.8%. What the Market Is Pricing That a $104 Billion Backlog Doesn't Show

On 11 June 2026, CoreWeave priced $1.25 billion of 9.625% senior notes due 15 July 2032 at par, according to the company's own announcement. This week the same bond traded at a yield of about 14.8%, briefly touching 15%, per figures CNBC and investinglive reported on 14 and 15 October. A bond that pays a fixed 9.625% coupon can only yield 14.8% if its price has fallen well below face value. By our estimate, using a settlement date of 16 October and the 15 July 2032 maturity, that price is roughly 80 cents on the dollar. Anyone who bought at issue in June is sitting on a paper loss of about 20 points, partly offset by the coupon already earned.

The headline version of this story is an AI debt scare. The more useful version is narrower and more interesting: the market is separating AI infrastructure borrowers into tiers, and it is pricing CoreWeave's tier on questions its $104 billion revenue backlog does not answer. This piece works through the arithmetic, sets the backlog against the company's own interest bill, gives the company's rebuttal, and says plainly what the evidence does not establish.

What a 14.8% yield on a 9.625% coupon means

Bond yield and bond price move in opposite directions because the coupon is fixed. When investors demand a higher return, the only way to deliver it on a bond with a set coupon is to pay less for it. We priced the 2032 notes with semiannual coupons and about 5.75 years to maturity. At a 14.8% yield the clean price comes out near 80.4. At 15% it is near 79.7. At a yield equal to the coupon, 9.625%, the price would be par. The roughly 5.2-point gap between the coupon and the yield is what turns into the 20-point price discount. These are our estimates from standard bond arithmetic, not quoted trade prices, and the exact figure depends on the settlement convention and on which dealer quote the headline yield came from.

A second CoreWeave bond tells the same story. The 9.75% notes due 2031 yielded about 14.9%, according to investinglive's summary of the CNBC reporting. S&P rates the company B+, which is sub-investment-grade, so a double-digit yield is not unusual in itself. What changed is the distance between a B+ issuer pricing a bond at par in June and the secondary market demanding nearly 15% four months later.

Two credit markets, not one

The sharpest comparison available this week is with Oracle, which is also an AI infrastructure borrower and also fell sharply on the same day. Oracle's 2032 notes yielded about 7.24%, per investinglive. That puts CoreWeave's 2032 paper roughly 7.6 percentage points wider, or about 760 basis points. Both companies are spending heavily on data centres and both stocks dropped: Oracle closed down nearly 9%, Nebius fell about 10% and Dell about 11%, while CoreWeave's shares fell roughly 16% by the close. Share-price figures differ by outlet (CNBC's body text cited about 10% intraday, and 24/7 Wall St's headline cited a different number for Nebius), so treat the moves as approximate.

The bond market, however, did not treat the two the same way. A 7.6-point gap on the same maturity year is the market saying that the identity and balance sheet of the borrower matter more than the shared label of AI infrastructure. Large, diversified, investment-grade-style borrowers can still raise money cheaply. Alphabet, Amazon, Meta and Oracle together issued more than $228 billion of bonds in 2026 as of 13 October, more than double all of 2025, according to LSEG data cited by investinglive. JPMorgan estimates hyperscaler spending at about $800 billion this year, rising above $1 trillion a year by 2027. That money is being raised, and it is being raised at prices that do not look like distress. CoreWeave is a different credit: smaller, rated B+, and funded largely by debt secured against chips and contracts.

The backlog and the interest bill answer different questions

CoreWeave's strongest argument is its backlog. In its second-quarter results, released 11 August, the company reported revenue backlog of about $104 billion as of 30 June, a figure that excludes more than $25 billion of net new customer commitments added in early July. Revenue in the quarter was $2.575 billion, up roughly 112% from $1.212 billion a year earlier. On those numbers the backlog is about 40 times quarterly revenue, or roughly ten years of revenue if the second quarter is simply annualised. That is a crude comparison, because backlog is delivered over multi-year contract terms and revenue is growing, but it shows how much of the equity story rests on contracts rather than on current earnings.

Bondholders, though, are paid from cash flow, not from backlog. The same release shows net interest expense of $640 million in the quarter, up from $267 million a year earlier, while adjusted operating income fell to $128 million, a 5% margin, from $200 million and a 16% margin. Net loss widened to $626 million from $290 million. On those figures, one quarter's adjusted operating income covered about one fifth of one quarter's interest. A caution is needed here: adjusted operating income is a company-defined measure, it is struck after depreciation, and lenders to asset-heavy businesses often look at earnings before depreciation instead, which would look far better. The direction is still the point. Interest more than doubled in a year while operating profit fell.

The balance sheet explains why. On our reading of the release, total debt at 30 June was about $35 billion, made up of roughly $31.4 billion of recourse debt and $3.7 billion of non-recourse debt, excluding finance leases. Cash and equivalents were $5.5 billion. Capital expenditure was $6.4 billion in the second quarter alone and $14.1 billion for the first half, about two and a half times the quarter's revenue. Debt totals are defined differently across data providers, and other figures circulated this week, so we describe ours as an estimate from the company's own tables rather than a verified headline number.

The company's case, and its limits

As CNBC reported, chief executive Michael Intrator has argued that the market reaction is overdone and that the contracted backlog more than covers the company's debt obligations. That is a claim by the company, and it should be read as one. It is also not absurd on its face: if $104 billion of contracted revenue is delivered, it dwarfs a debt stack of roughly $35 billion by our count. Co-founder Brannin McBee has separately said the company aims for investment-grade ratings on more of its chip-backed debt, which he said would open borrowing from insurers, according to 24/7 Wall St's account of an interview.

The limit of the argument is that backlog is a promise to buy, not cash received. Converting it requires CoreWeave to build and power the capacity on schedule, which is exactly what the capex line is paying for, and requires customers to take delivery and pay. Backlog carries counterparty risk and delivery risk, and neither is visible in the headline number. Nothing in the public reporting we reviewed shows a missed payment or a customer defaulting. The bond move is a repricing of risk, not evidence that anything has gone wrong yet.

What would change the picture

Three things are worth watching. The first is CoreWeave's next quarterly report, which on the 11 August pattern should land in November: the question for bond investors is whether interest expense keeps outrunning operating income. The second is any move toward investment-grade ratings on pieces of the debt, which McBee has described as a goal and which would lower the cost of the structure if achieved. The third is the rate backdrop. Inflation at 3.7% on the September CPI reading we covered earlier keeps long-dated yields high, and a company that must refinance several billion dollars of debt in the early 2030s is more exposed to that than one with a cheaper, longer-dated funding stack.

For readers trying to tell an AI debt scare from one weak issuer repricing, the evidence this week points toward the second reading. The tier gap between CoreWeave and Oracle is wide, hyperscaler issuance continues, and the stress is concentrated where leverage is highest and operating profit is thinnest. That could change if the repricing spreads to stronger borrowers, which it has not so far. It is a story about who gets to borrow cheaply, and why a large backlog alone does not settle that.