AI stocks fell hard last week, but the bigger warning came from the bond market. As lenders charged AI borrowers more, they also began pricing in the customers behind each loan. These four stories show where that risk sits now.
CoreWeave Bonds Hit Distress Levels
CoreWeave’s stock fell 15.5% in one day as investors worried about how much the company has borrowed. The next day, the yield on its 2031 bonds climbed to 15.78%. The spread on its five-year credit default swaps passed 1,000 basis points, a level the credit market generally treats as distress. CEO Mike Intrator told employees the company’s backlog ‘more than supports all of our debt obligations.’
Five days after that memo, Goldman Sachs upgraded CoreWeave to Buy and raised its price target from $106 to $139. CoreWeave shares rose 3% after the upgrade but closed the day down 0.44%.
One recent loan explains why lenders watch that backlog so closely. CoreWeave pledged a set of customer contracts to secure its $2.6 billion facility, which prices at the Secured Overnight Financing Rate (SOFR) plus 5.5%. Under the loan’s terms, customer payments go into a lender-controlled account if one of those customers gains the right to cancel, or if projected contract value falls below 2.4x the loan balance. If either problem lasts for three consecutive monthly payment dates, CoreWeave must use the funds in that account to repay part of the loan early. The contracts also run shorter than the loan. They average three years, so CoreWeave will have to renew or replace them before the five-year loan comes due.
CoreWeave wasn’t the only AI cloud provider hit. Oracle, which also sells computing capacity to AI labs, fell nearly 9% the same day CoreWeave’s stock dropped 15.5%. Over the week, the spread on its five-year credit default swaps rose about 10% to 282 basis points.
S&P Global Admits It Can’t See All the AI Debt Hyperscalers Carry
S&P Global’s head of digital infrastructure said it’s safe to assume more AI-linked debt exists than the agency can see. Some of that debt sits in deals where a hyperscaler guarantees a data center build or a smaller AI company. When the hyperscaler doesn’t report the guarantee, and S&P doesn’t rate the deal, the agency has no record of it.
Those blind spots matter because of how much these companies plan to spend. S&P expects Amazon, Microsoft, Alphabet, Oracle, Meta and SpaceX to spend more than $7 trillion on data centers and AI from 2025 through 2030. Even so, it isn’t changing any ratings. Investors who rely on those ratings may be looking at leverage figures that leave out some of what these companies have guaranteed.
Volta’s $5 Billion Norway Data Center Loan Ties Its Rate to Anthropic
Volta is an AI infrastructure company founded in February, with plans for a data center in Norway. To build it, Volta is borrowing $5 billion through a loan that JPMorgan and Goldman Sachs are selling to lenders. Some of those lenders pushed back because they already had as much data center debt as they wanted. Before they’d commit, they asked for more collateral and a higher interest rate. The loan is now expected to price at 11% or more, which is unusually high for this kind of project. Private credit firm Apollo Global Management anchored the deal with a $1.5 billion order, though terms could still change before it closes.
Anthropic’s role made lenders even more cautious. Volta is counting on Anthropic to buy much of the data center’s computing capacity, but Anthropic shares little financial information with lenders. To win them over, Volta agreed to an interest rate that falls if Anthropic earns an investment-grade credit rating. That clause means lenders are pricing Anthropic’s credit into Volta’s loan.
OpenAI Races to Raise $30 Billion Before Anthropic’s IPO
OpenAI wants commitments for a round of about $30 billion before Anthropic’s expected listing in mid-November. The round could value OpenAI at more than $1.4 trillion, up from $852 billion in March. According to investors familiar with the company, OpenAI needs the new funding to cover its infrastructure commitments and strengthen its balance sheet in case its 2027 listing slips.
Investors comparing the two labs will run into an accounting dispute. OpenAI subtracts its partners’ cut before it reports revenue. On that basis, it told investors its annualized revenue is still about $50 billion, the same figure we covered last week. The $68 billion figure in earlier news reports also counted revenue from OpenAI’s partners. The company now expects to hit $70 billion by year-end. It has accused Anthropic of inflating revenue by counting the full value of contracts sold through cloud providers before the providers deduct their fees. Anthropic says its accounting follows established practice. Its IPO filing will lay out that method in detail, so OpenAI wants to finish its round before the filing goes public.
Paid subscribers get the AI financing exposure tracker, which maps who funds, guarantees, and owes on deals like CoreWeave’s and Volta’s.
Which AI company’s debt do you want audited next?