CoreWeave's distress puts neocloud contracts at real risk

CoreWeave's bonds crossed into distressed territory this week—meaning investors are demanding 1,000 basis points more in yield than they'd get from government bonds. That's not a blip. That's a signal. And if you signed a multi-year compute contract with a second-tier AI cloud provider, you may have just discovered a liability you never priced in.

The numbers around CoreWeave are genuinely hard to look at. The Verge reported that CoreWeave owed more than $35 billion in debt as of its second quarter earnings, lost more than a billion dollars in the first half of this year, and does not actually make money on its core business of selling compute. Its shares fell more than 20 percent from Tuesday's close. A measure of its credit risk hit a record high, according to Bloomberg as cited by The Decoder. Oracle's shares fell almost 10 percent in the same stretch, with a sell-off in Oracle bonds too.

The sector-wide anxiety got louder when Australian provider Firmus abruptly pulled its IPO, and when Microsoft withdrew from a data center deal in Loughton, England, where it had planned to lease 23,000 Nvidia chips from Nscale—a site where utility providers don't expect power until the 2030s, according to The Telegraph as cited by The Decoder. Nscale said its relationship with Microsoft remains 'strong and long-term,' but some investors are already pushing to cut Nscale's target market cap from about $30 billion to somewhere between $16 billion and $19 billion.

Hundreds of neoclouds, maybe ten survivors

Here's the number that should concentrate your mind: two executives from the data center and neocloud industries told The Information that only seven to ten of the hundreds of AI cloud providers currently in the market will still be around in a few years. Gartner puts it differently but arrives at the same place—predicting that 80 percent of independent neoclouds will cease independent operations by 2030. Bernstein estimates that only 35 percent of announced North American data center capacity will ever get built.

Satya Nadella said back in late February 2025 that he expected compute oversupply to start in 2027 and that the industry was too focused on supply instead of actual demand. His exact words: 'The only thing that's going to happen with all these data centers is that prices will fall.' He was right about the direction. The question now is how fast and how hard.

CoreWeave CEO Michael Intrator, in a memo to employees, blamed smaller rivals including Firmus, Lambda, and Nscale for the turmoil, saying they were 'trying to raise enormous amounts of capital without the execution, revenue, scale, or software solutions that CoreWeave delivers to its customers.' Maybe. But CoreWeave itself has $103.7 billion in remaining performance obligations, about 40 percent of which is due in 2028—and the companies on the other side of those contracts cannot cancel them except under very limited circumstances. Three customers—Microsoft, OpenAI, and an undisclosed third party that The Verge says is probably Nvidia—generated almost three-quarters of CoreWeave's second quarter revenue. That's a fragile foundation for a company carrying this much debt.

The counterparty risk nobody budgeted for

This is where it gets personal for business leaders who aren't in the cloud business themselves. Picture any mid-sized company that signed a two- or three-year contract with a neocloud provider in 2024 or 2025, when GPU access felt scarce and locking in capacity felt smart. The contract looked like an asset. Now it might be a claim against a company that's heading toward insolvency—or at best, a fire sale to a hyperscaler who may or may not honor the original terms.

John Yovanovic, co-head of leveraged finance at MetLife Investment Management, put it plainly to Bloomberg, as cited by The Verge: '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.' More supply means lower prices, which means the revenue projections that justified all this debt start to look fictional.

The fair counterargument is that CoreWeave is not a random startup. It has real contracts, real revenue, and cofounder Brannin McBee told the Financial Times this week that the company wants its debt to be investment-grade—not junk—so it can borrow from insurers, describing their capital as 'unlimited.' CoreWeave did secure its first investment-grade chip loan earlier this year. And Nvidia has strong incentives to keep the neocloud ecosystem alive: the non-hyperscaler segment brought Nvidia $40 billion in the July quarter, representing 45 percent of its data center revenue, more than double the figure from a year earlier. Nvidia is actively supporting neoclouds with equity investments, chip allocations, loan guarantees, and customer leads.

But 'Nvidia wants them to survive' is not the same as 'they will survive.' And even if CoreWeave makes it through, many of its smaller rivals won't. The shakeout is the story, not just the distress at any one company.

What to do this week

Pull out every compute contract your company holds with a neocloud provider—any provider that isn't AWS, Azure, or Google Cloud. Then ask three questions. First, what happens to your workloads if this provider goes under or gets acquired mid-contract? Second, does your contract include any termination rights if the provider's financial condition deteriorates materially, or are you locked in regardless? Third, are you paying in advance or in arrears—because prepaid capacity at a failing provider is just an unsecured claim in a bankruptcy proceeding.

If your legal team hasn't looked at force majeure and insolvency clauses in these agreements recently, now is the time. The providers that survive this shakeout will be fine. The ones that don't will leave their customers holding contracts that are worth exactly what the bankruptcy court decides.

Watch CoreWeave's next earnings report and any news about its debt refinancing. If it can't move toward investment-grade status—or if one of its three mega-customers quietly reduces its commitment—the bond market's anxiety will look prescient. I'd also watch whether Nvidia starts quietly consolidating its neocloud investments rather than adding new ones. That would be the clearest signal that even the most motivated backer thinks the field is narrowing fast.