Unsecured CoreWeave bonds now trade like distressed debt. Its loans backed by blue-chip contracts, and the data centers it rents, cost far less. The pricing shows where lenders see real risk.
Stock investors chase upside. Bond investors price what can go wrong. This month, bond investors sent a blunt message about one corner of the artificial intelligence boom.
On Oct. 14, CoreWeave’s unsecured bonds due 2032 fell below 80 cents on the dollar, Bloomberg reported, citing Trace data. Traders widely treat that level as a distress signal. Yet debt tied to hard assets and strong customers held up much better. That gap tells investors where the weak points sit.
A Brutal Week for a Neocloud
CoreWeave rents AI computing power to tech companies. It belongs to a new class of firms called “neoclouds,” and it competes with Amazon and Microsoft.
The stock fell about 15% on Oct. 14 and closed at $65.78. By Oct. 16, it had lost nearly a quarter of its value for the week. Meanwhile, the yield on its October 2031 bond briefly topped 16%, Reuters reported. The cost of insuring its debt against default also set records above 1,000 basis points.
The balance sheet explains the fear. S&P Global rates CoreWeave B+, solidly in junk territory. Debt and lease obligations totaled about $52 billion at the end of June. The company plans to spend $35 billion to $39 billion on capital projects this year. Furthermore, J.P. Morgan expects its free cash flow to stay negative until 2030.
Interest costs show the strain. In the second quarter, CoreWeave paid about $640 million in interest. Its operating loss, by comparison, was $49 million. Debt stood at roughly 5.8 times annualized second-quarter EBITDA, the Financial Times reported.
The CEO Says the Market Has It Wrong
CEO Mike Intrator rejected the verdict. He sent a memo to staff, which Bloomberg News reviewed. In it, he wrote that markets sometimes misunderstand new kinds of companies and act irrationally.
His case rests on the backlog. CoreWeave reports $104.2 billion in contracted future revenue. Intrator says that figure more than covers the company’s debt. He also blamed delayed stock offerings and shaky debt deals at rivals for souring sentiment.
Some on Wall Street agreed. Goldman Sachs upgraded the stock to Buy on Oct. 19, though it cut its price target from $139 to $106. Analyst Gabriela Borges estimated CoreWeave’s borrowing costs could rise 100 to 200 basis points, but likely no more. Nvidia CEO Jensen Huang also said CoreWeave was getting stronger and that Nvidia saw no need to backstop it.
Skeptics remain. Davidson analyst Gil Luria has warned CoreWeave’s stock could become worthless if the AI investment boom reverses. Bondholders also stayed cautious. By Oct. 20, the 2031 notes still yielded about 15%.
One Company, Many Prices
Here is the detail most coverage misses. CoreWeave does not pay one price for money. Instead, it pays very different rates depending on what backs each loan.
Its best-rated debt sits in a separate, bankruptcy-remote subsidiary. In March, that unit closed an $8.5 billion loan secured by computer chips and a six-year Meta Platforms contract. Moody’s rated it A3, which is investment grade. Its fixed-rate slice was priced at about 5.9% when the deal closed.
The ladder climbs from there, according to company announcements and ratings records:
| Financing | Backed by | Rating | Price at issue |
|---|---|---|---|
| $8.5B loan, March 2026 | Chips plus six-year Meta contract | Moody’s A3 | About 5.9% fixed, or SOFR + 2.25% |
| $3.1B loan, May 2026 | Chips plus two non-investment-grade customers | Moody’s Ba2, Fitch BB+ | SOFR + 4.50% |
| $2.6B loan, August 2026 | Chips plus contracts averaging about three years | Moody’s Ba2, Fitch BB+ | SOFR + 5.50% |
| Unsecured notes due 2032 | The whole company | S&P B+ (issuer) | 9.625% coupon; now yielding mid-teens |
The ladder prices customer quality and legal separation, not the CoreWeave name. In practice, a Meta contract is worth more to a lender than CoreWeave’s own promise to pay.
A word of caution, though. That 5.9% rate does not mean CoreWeave can borrow cheaply across the board. It applies only to a walled-off pool of assets with a top-tier customer. The company’s general creditors stand behind those structures and absorb whatever risk is left.
The August loan also shows lenders stretching. It runs about five years, while the contracts behind it average about three. CreditSights analyst Jordan Chalfin flagged that kind of mismatch to the Financial Times. If the industry overbuilds, he asked, why would customers renew?
Chasing Insurance Money
CoreWeave wants more of the cheap kind of debt. Co-founder Brannin McBee told the Financial Times the company seeks investment-grade ratings on more chip-backed loans. Eventually, he hopes CoreWeave itself will earn one. That would open the door to life insurers, whose capital he called “unlimited.”
However, McBee conceded that contracts with lower-rated AI labs are not ready for that market yet. The FT also reported that CoreWeave’s founders have sold about $2 billion of stock this year. It cited S&P Capital IQ data.
Why the Buildings Outperform CoreWeave Bonds
The same logic runs through the data centers CoreWeave leases. Applied Digital is building a campus in North Dakota for CoreWeave. A bond tied to that project yields around 8%, according to MarketAxess data cited by The Wall Street Journal. CoreWeave’s 2030 notes, by contrast, yield close to 15%.
Same AI boom, very different borrowing costs
Approximate yields on debt tied to CoreWeave and AI data centers, October 2026
Sources: CoreWeave; Moody’s; MarketAxess via The Wall Street Journal; Morningstar; Forbes analysis of FactSet data; Reuters. The Meta-backed loan figure is its fixed rate at issue in March; other figures are market yields in mid-to-late October 2026. Chart: Midpoint News.
Several protections explain the gap. Lenders typically hold a claim on the building itself. Some deals repay principal within the lease term, so lenders never need a second tenant. Moreover, a finished, powered data center keeps value if one tenant fails, because demand for capacity remains strong.
Applied Digital also benefited from the Meta-backed structure. CoreWeave shifted part of its lease to the investment-grade subsidiary, and the related bonds rallied. Pimco strategist Lotfi Karoui summed it up for the Journal: “You should not view the AI trade in credit as one big trade.”
Even Secured Debt Has Limits
Still, the safer link carries risk. Applied Digital’s 7% secured notes, sold in June, slipped to about 94.5 cents in mid-October, Morningstar reported. That pushed the yield to roughly 8.4%. J.P. Morgan analysts also cut those bonds to Neutral, citing a coming wave of new supply. Applied Digital may need about $8 billion more in project debt.
The pile of risky data-center debt keeps growing, too. Of roughly $80 billion in U.S. data-center bonds sold this year, about $37 billion carries junk ratings. That figure comes from a Forbes analysis of FactSet data. By comparison, all of 2025 saw only about $8.6 billion in junk issuance. Average yields on the junk group neared 10% in mid-October.
Tallies vary by source. Dealogic and Goldman Sachs data count more than $100 billion in data-center dollar bonds through Oct. 16. That equals about a quarter of all AI-related issuance, up from about 16% last year.
Big Tech Borrows Big, Too
The hyperscalers are borrowing on a historic scale as well. Alphabet, Amazon, Meta and Oracle sold more than $228 billion in bonds this year as of Oct. 13, according to LSEG data. That is more than double their 2025 total.
Their credit is far stronger, and Microsoft still holds a rare triple-A rating. Nevertheless, investor appetite has cooled. In July, a Reuters analysis checked 91 hyperscaler bonds issued in 2026. Of those, 78 traded at higher yields than at issue. Apollo also said order books shrank sharply. They fell from nearly five times deal size in February to under two times in July.
Higher Rates and Local Opposition
The macro backdrop adds pressure. The Federal Reserve under Chair Kevin Warsh raised rates in September for the first time since 2023. As a result, the 10-year Treasury yield hit about 5.2% in mid-October, a level last seen in 2007. Traders have since trimmed bets on an October hike after Warsh declined to prejudge the next meeting.
For heavily indebted builders, higher rates mean costlier refinancing. In other words, every new dollar of debt costs more.
Local politics is another drag. Data Center Watch counted at least 75 U.S. projects worth about $130 billion delayed or blocked in the first quarter. Organized opposition groups more than doubled, from 396 to 833. In New York, Gov. Kathy Hochul ordered a pause of up to one year on permits for large new data centers.
Opposition draws support from both parties. Residents commonly cite electricity bills, water use and noise. That raises the odds a project never gets finished. On the other hand, scarcity can make finished projects worth more. Barclays analyst Andrew Keches has said completion itself can push spreads tighter.
Analysis: Markets Are Pricing Risk, Not Ending the Boom
None of this means the AI boom is over. Rather, lenders are pricing risk borrower by borrower, and contract by contract.
That is how credit markets should work. Lenders charge less for hard collateral and blue-chip customers. They charge more for leverage and future promises. That sorting is happening without a government rescue. Indeed, Nvidia says it sees no need to backstop its most prominent customer.
Ordinary savers should still pay attention. Passive funds that track investment-grade bond indexes absorb more tech debt as these firms borrow. Vontobel has estimated AI-related bonds could eventually make up 15% to 20% of most corporate bond indexes.
The next test comes in November, when CoreWeave reports third-quarter results. Intrator has promised to put the facts in front of the market. That report will show whether the backlog can outrun the cost of capital unsecured lenders now demand.
This article builds on reporting by The Wall Street Journal (“Weak Points in AI Bonds Are Becoming Clearer,” Heard on the Street), with additional reporting from Bloomberg, Reuters, the Financial Times, Morningstar, Forbes and company filings.