CoreWeave’s Debt Scare Is Spreading Across AI

CoreWeave’s bonds have slipped into distressed territory. That’s the usual label for debt where investors demand 1,000 basis points (10 percentage points) more yield than they’d get from government bonds. The Verge AI reports that the stock fell more than 20 percent in the days that followed, and Oracle and other AI infrastructure names took damage too.

This matters beyond one company. CoreWeave is the poster child for debt-fueled AI buildout. When its lenders get nervous, everyone who has borrowed against the AI boom feels it.

📉 What Set It Off

Three things landed in the same week:

  • Chamath Palihapitiya pointed out the distressed debt and asked publicly whether the AI buildout is a private credit bubble. When the “SPAC King” starts asking that, people listen.
  • Cofounder Brannin McBee told the Financial Times that CoreWeave wants investment-grade credit so it can borrow from insurers, whose capital he called “unlimited.” To a nervous bond market, that sounded like a plan to borrow a lot more.
  • The supply problem got louder. John Yovanovic, co-head of leveraged finance at MetLife Investment Management, told Bloomberg: “The problem is not that credit profiles have changed in the last year, it’s that we’ve got another year and a half of supply coming.”

Yovanovic’s point is the sharpest one. Lenders aren’t saying CoreWeave suddenly got worse. They’re saying the market is about to be flooded with AI infrastructure debt, and they want to get paid more for holding it.

💸 The Numbers Behind the Nerves

CoreWeave’s balance sheet explains why investors are so jumpy:

  • More than $35 billion in debt as of its second quarter
  • Over $1 billion in losses in the first half of the year
  • One main business: selling compute, which doesn’t make money yet
  • GPUs as collateral: CoreWeave pioneered borrowing against chips, which lose value as newer generations ship

The risk here is a feedback loop. Expensive debt makes profit harder to reach. If borrowing costs rise further, the path to profitability gets steeper still.

🛡️ The Case Against Panic

The Verge AI’s Elizabeth Lopatto has long argued this business model may be unsustainable, but she isn’t calling it a collapse yet. Her reasoning is fair.

CoreWeave had $103.7 billion in remaining performance obligations as of Q2. That’s contracted revenue customers have committed to pay. About 40 percent of it comes due in 2028, and those contracts can only be canceled in very limited circumstances.

The customer list is also strong. Microsoft, OpenAI, and what’s probably Nvidia made up almost three-quarters of Q2 revenue. Microsoft and Nvidia can clearly pay. They also have good reason not to let CoreWeave fail, because the fallout would hit the whole market they depend on. OpenAI is the weaker link, but unless it’s in far worse shape than anyone knows, the payments should keep coming for a while.

As Lopatto puts it, a falling stock price isn’t the same thing as financial doom. As long as CoreWeave can service its debt, “the music keeps going.”

🌐 Why This Is Bigger Than CoreWeave

The spillover is what stands out to me. Oracle shares fell almost 10 percent over the same stretch, and its bonds sold off too. Oracle has also borrowed heavily to build AI data centers.

The Verge AI’s conclusion is blunt: the worries about CoreWeave apply to every neocloud. The core question is whether renting out bare-metal GPUs is a good business at all, or just a good business while demand outruns supply.

That ties back to Yovanovic’s warning. Supply is still coming. If capacity catches up with demand, rental prices drop, and companies that financed GPUs at junk-bond rates get squeezed from both ends.

🧭 What Practitioners Should Do

If you buy compute or build on AI infrastructure, here’s how to read this:

  • Diversify compute suppliers. Don’t run critical workloads on a single neocloud. Have a fallback with a hyperscaler or a second provider.
  • Read your contract terms. Know what happens to your capacity if your provider restructures debt or gets acquired.
  • Watch credit spreads, not just stock prices. Bond yields on AI infrastructure companies are an early warning that’s harder to dismiss than a volatile share price.
  • Expect pricing pressure. If supply floods in as Yovanovic expects, GPU rental rates could soften. That’s good news for buyers who stay flexible.
  • Track OpenAI’s finances. It’s a major source of neocloud revenue. Its ability to pay is now a systemic question.

🔭 What Comes Next

CoreWeave probably isn’t going under next quarter. Contracted revenue and powerful customers give it room. But the market just showed how fast confidence in AI infrastructure debt can crack, and how quickly that spreads to Oracle and others.

The real test comes when the AI boom slows. That’s when we’ll find out whether renting GPUs is a lasting business or just a byproduct of a gold rush. My take: the smart move is to plan for both outcomes now. Full details are available in the original reporting from The Verge AI.

Scroll to Top