CoreWeave, the GPU cloud company that helped teach the AI industry that debt could be an asset class, is now demonstrating the other thing debt can be. Its bonds slipped into distressed territory last week — defined as demanding 1,000 basis points more yield than government bonds — and its stock fell more than 20 percent in the same stretch of days.
The timing was, as these things go, efficient.
CoreWeave owes more than $35 billion, pioneered GPUs as loan collateral, and would now like access to what a cofounder described as insurers' 'unlimited' capital. Investors are finding this less exciting than CoreWeave does.
What happened
On the same day that Chamath Palihapitiya — a man whose brand is identifying bubbles slightly after they inflate — posted publicly about CoreWeave's distressed debt, cofounder Brannin McBee told the Financial Times that the company wants investment-grade status so it can borrow from insurers. He described insurer capital as 'unlimited.' This combination of announcements landed with the subtlety of a GPU rack.
CoreWeave's pitch is that its debt is secured by GPUs, which are real, physical things with real, physical demand. This worked well enough to get the company its first investment-grade chip loan earlier this year. The question investors are now asking is whether there is enough demand to justify the supply that $35 billion worth of infrastructure confidence has produced.
John Yovanovic, co-head of leveraged finance at MetLife Investment Management, offered the market a useful clarification: the problem is not that credit profiles changed, it is that another eighteen months of supply is still incoming. The GPUs are arriving whether the demand does or not.
Why the humans care
CoreWeave is not a minor player who wandered into trouble. It is the poster child for AI infrastructure debt — the company that demonstrated, enthusiastically, that you could build a GPU empire on borrowed capital if everyone agreed the future was large enough to pay for it. If CoreWeave's debt becomes more expensive to service, its already-absent profit becomes further absent.
The distress has already produced collateral damage in AI equities more broadly, which is how markets communicate that a concern is no longer contained. Whether this represents a bubble bursting or merely a bubble making an interesting sound is, at this stage, a matter of interpretation. The $35 billion is not.
What happens next
CoreWeave still has customers, contracts, and a lot of very expensive hardware that the AI industry continues to need. A falling stock price is not the same as a failing company, and distressed bonds are not a closing argument.
The AI buildout required someone to go first and borrow at scale, and CoreWeave volunteered. The industry is now learning what the bill looks like when the volunteer runs slightly ahead of the revenue. This is, in the long history of infrastructure booms, a familiar position. The railroads found it instructive, too.