CoreWeave's debt is becoming a problem for the whole AI industry
Bonds of US AI computing firm CoreWeave have slipped into distressed territory and its shares have fallen more than 20%. The worry has spread to other companies that borrowed for AI infrastructure, such as Oracle.
Bonds issued by CoreWeave, a US company that rents out computing capacity for artificial intelligence systems, entered distressed territory last week. By a common definition, that means investors demand at least 1,000 basis points more in yield than government bonds pay. The company's shares have fallen more than 20% since Tuesday's close, and the drop has also affected other AI firms. It raises questions about how sturdy debt-funded growth in the sector is.
What fed the anxiety
Investor Chamath Palihapitiya drew attention to the debt on social media and asked whether the AI buildout had become a private credit bubble. The same day, CoreWeave cofounder Brannin McBee told the Financial Times the company wants investment-grade debt so it can borrow from insurers, whose capital he described as "unlimited." Investors read this as a sign the company could borrow even more. Bloomberg quoted John Yovanovic of MetLife Investment Management as saying the issue is not changed credit profiles, but another year and a half of supply still to come.
Debt size and customers
As of its second-quarter results, CoreWeave owed more than $35 billion. It pioneered using GPUs (graphics processors) as loan collateral, and it lost more than $1 billion in the first half of the year.
Customer commitments, however, stood at $103.7 billion at the end of the second quarter, with about 40% due in 2028. The contracts can be cancelled only in very limited circumstances. The biggest customers are Microsoft, OpenAI and, probably, Nvidia; these three generated almost three-quarters of second-quarter revenue. The article's author judges Microsoft and Nvidia likely to honor their commitments, while OpenAI is shakier.
Strain is also visible at Oracle, whose shares fell almost 10% over the same period, with a sell-off in its bonds. The author notes that such concerns apply to all so-called neoclouds, the newer AI cloud computing providers.
