CoreWeave’s Financial Struggles Impact the Entire AI Sector

CoreWeave had a rough week recently, with its bonds slipping into distressed territory—defined as when investors demand a yield that’s 1,000 basis points higher than government bonds. Last Tuesday, Chamath Palihapitiya, known for his SPAC ventures, pointed out this distressed debt and raised concerns about whether the AI expansion is creating a private credit bubble. Since then, CoreWeave’s shares have plummeted over 20 percent, which is quite significant! This downturn also seemed to ripple through the market, affecting other AI companies.

While I’ve pondered whether the AI sector is in a bubble, Palihapitiya’s echo of concerns from a prominent figure questioning the financial sustainability of AI projects sends a stronger signal. When I express concern, it’s one thing; when someone like Elon Musk does, it carries a different weight.

To add to CoreWeave’s troubles, co-founder Brannin McBee mentioned last week that the company aims to elevate its debt to investment-grade status instead of junk, allowing it to borrow from insurers, whose capital he described as “unlimited.” This ambition isn’t entirely outlandish—earlier this year, CoreWeave secured its first investment-grade chip loan. However, the desire to tap into the vast resources of insurers raises fears among investors that CoreWeave, already heavily indebted, might take on even more debt.

John Yovanovic, co-head of leveraged finance at MetLife Investment Management, noted that the issue isn’t just a shift in credit profiles but the looming supply increase over the next year and a half.

Some are suggesting this indicates the AI bubble is bursting, but I’m not convinced. A falling stock price doesn’t automatically spell financial disaster. Stocks reflect confidence in a company’s future rather than its day-to-day operations. However, CoreWeave’s substantial debt becomes a heavier burden if borrowing costs rise, complicating its path to profitability.

CoreWeave exemplifies the AI debt situation, with over $35 billion owed as of its second-quarter earnings. This is partly due to its innovative use of GPUs as loan collateral and the fact that its business model revolves around selling compute power, which isn’t currently profitable—the company lost over a billion dollars in the first half of the year.

So why am I hesitant to declare this a victory? The answer lies in the “remaining performance obligations” from its customers, totaling $103.7 billion as of the second quarter. About 40 percent of that is due to CoreWeave by 2028, and the companies involved can’t easily cancel these contracts except under specific conditions. Its largest clients include Microsoft, OpenAI, and likely Nvidia, although CoreWeave hasn’t disclosed its third-largest customer. These three accounted for nearly three-quarters of the company’s revenue in the second quarter.

Microsoft and Nvidia are likely reliable for their payments—as long as it serves their interests. I suspect they understand that if CoreWeave fails, it could lead to significant collateral damage in the market, so they’re unlikely to push the company to the brink. OpenAI’s situation is less certain, but unless it’s in worse financial shape than we realize, CoreWeave should be stable for a while. As long as it can manage its debt, it can keep operating. The real test will come if the AI boom starts to wane, revealing whether CoreWeave’s business model is truly sustainable.

Nonetheless, the uncertainty surrounding CoreWeave has created a shaky environment for other companies with substantial debt tied to the AI expansion—Oracle’s shares, for example, have dropped nearly 10 percent in the same period, and there’s been a sell-off in Oracle bonds as well. The concerns investors have about CoreWeave are echoed across the neocloud sector, raising questions about the viability of the bare-metal business model.

In any case, some investors are seizing the opportunity to buy the dip.