What CoreWeave's 16% bond yield says about AI debt

The debt market is charging AI data center builders more to borrow, and the clearest sign is CoreWeave. According to CNBC’s report, the yield on CoreWeave’s 2032 bond, which pays a 9.625% coupon, has climbed above 16%. The cost of insuring its debt against default also jumped. That gap between coupon and yield is the number to understand. It tells you how much lenders now demand to fund this part of the AI buildout.

The background you need

Neocloud. A neocloud is a newer company that rents out GPU computing power, mostly for AI work. It competes with the giant cloud providers. CoreWeave is the best-known example. Its business is buying expensive hardware, putting it in data centers, and selling access to that compute to customers.

Hyperscaler. This is the industry’s word for the biggest cloud operators, such as Amazon. They are the largest spenders on AI data centers.

Capex. Short for capital expenditure: money spent on long-lived assets like buildings, chips and power equipment. The AI capex arms race means these companies trying to outspend each other on that hardware.

Coupon versus yield. A bond’s coupon is the fixed interest rate the company promised when it sold the bond. Here that is 9.625% a year. Once the bond trades, its price moves. If investors get nervous, they pay less for it. A lower price with the same fixed payments means a higher return for whoever buys now. That return is the yield. So a 9.625% bond yielding above 16% is trading well below its face value. Buyers today want roughly 16% a year to hold CoreWeave’s risk.

This does not mean CoreWeave pays 16% on that existing bond. Its payments stay at the coupon. The yield matters because it shows what new borrowing would likely cost.

Basis points. One basis point is one hundredth of a percentage point. So 100 bps equals 1 percentage point.

Credit default swap (CDS). A CDS works like an insurance policy on a company’s debt. The buyer pays a regular fee, and if the company defaults, the seller covers the loss. The fee is quoted in basis points per year of the amount insured. A higher CDS spread means the market sees more default risk, or simply wants more pay to carry it.

Free cash flow. Cash from operations minus capex. If it is negative, the company spent more on building than its business brought in during that period.

The credit card analogy. According to CNBC, Jefferies strategist Christopher Wood framed the situation as a credit card in a Friday note: “The most plausible way the AI capex arms race will end is that the credit card will be taken away.” The picture is useful. Lenders can raise the rate, and they can lower the limit. But the analogy breaks in two places. First, there is no single card issuer. Thousands of bond buyers reprice the debt every trading day. Second, the old balance keeps its old rate. Existing bonds keep their coupons. The higher rate hits new borrowing and refinancing.

What actually changed

Some of this is not new. CNBC notes that CoreWeave’s CDS was already elevated, with a peak in October. Hyperscaler borrowing has also been heavy all year.

What is new, according to CNBC and Deutsche Bank:

• CoreWeave’s five-year CDS rose above 1,289 bps on Thursday, past its October peak. It eased 8% on Friday to around 1,155. CNBC also gives a figure of roughly 1,100 in a comparison table. These numbers come from different points in time, so they do not line up exactly.
• Default insurance got pricier for the strongest borrowers too. Over the last month, Nvidia’s CDS rose 76%, Broadcom’s 49% and Amazon’s 44%.
• Deutsche Bank’s Jim Reid said debt and stocks are now moving more in step, and that it remains to be seen whether credit weakens further.

CNBC attributes the shift to politics. Credit markets, it reports, see more hurdles for data centers under incoming Democratic governors in Ohio, Texas and Georgia, the three states with much of the planned capacity, plus a Democratic-controlled Congress. Goldman Sachs estimates moratoriums could affect 45% of planned U.S. capacity, much of it in Texas. Texas governor-elect Gina Hinojosa has said she is not opposed to data centers, but wants a moratorium until the legislature passes protections for communities.

There is a counterweight. National Economic Council Director Kevin Hassett told CNBC he expects data center growth to continue as forecast. And issuance has not dried up. Goldman counts five AI bond deals totaling $67 billion since October 29. That brings hyperscaler debt issuance this year to $296 billion.

What it means for developers who invest on the side

The key point is the spread between CoreWeave and everyone else. Nvidia’s CDS sits near 139 bps. Oracle’s is about 298. CoreWeave’s is roughly 1,100. The percentage jumps at Nvidia, Broadcom and Amazon look dramatic, but CNBC itself stresses that a large rise from a low base does not mean the same credit risk. The market is drawing a sharp line between the giants and the neoclouds.

Why might neoclouds sit on the wrong side of that line? CNBC gives one reason. If hyperscalers consolidate or cut back on capex, neoclouds could see fewer or shorter compute contracts, or delayed prepayments. The report does not spell out how that would flow through to CoreWeave’s balance sheet. So treat any link to its debt as a possibility, not a mechanism the sources describe.

The hyperscaler side matters too. Goldman says hyperscalers collectively had negative free cash flow in the third quarter, because capex exceeded operating cash flow. Goldman names delayed monetization, meaning AI spending taking longer to earn money, and growing purchase commitments as the key risks to more debt issuance, subject to market conditions. CNBC frames the big question as how quickly AI spending will generate enough cash to keep investors comfortable.

What to watch:

• CoreWeave’s bond yield and CDS. Is the gap to Nvidia and Oracle widening or narrowing?
• CDS at the big names. Watch the level, not just the percentage change.
• State actions in Texas, Ohio and Georgia once the new governors take office.
• Hyperscaler free cash flow in the next quarterly reports.
• Pricing on new AI bond deals. Do they still sell easily, and at what rates?

What we don’t know yet

• Whether politics is the main driver. CNBC links the move to the election results. The reports do not show how much of the repricing comes from politics versus worries about cash flow.
• Whether moratoriums happen, and how big they are. Goldman’s 45% figure describes what moratoriums could affect, not what has been decided.
• When peak borrowing arrives. The banks measure different things. Goldman forecasts $440 billion of hyperscaler debt issuance next year, plus $300 billion of project-level issuance. Bank of America estimates AI-related credit, public and private, rising 31% to $685 billion in 2027, then falling 23% to $531 billion in 2028. BofA flags downside to its forecast if the Democratic wins slow AI projects.
• How contracts respond. Whether neocloud contracts actually get shorter or prepayments slip is not yet reported.