Timeline

CoreWeave shares and bonds fall on AI-debt concerns

The GPU-cloud firm's bonds due 2031 passed a 16% yield and credit-default swaps implied a roughly 63% five-year chance of default, up from about 30% a year earlier.

  • Money & business
  • Compute & infrastructure
  • Minor

CoreWeave shares fell a further 6% in premarket trading on Thursday 15 October, after dropping more than 15% the previous day, as the GPU-cloud provider was caught up in a broader rout in AI-infrastructure stocks. Bonds the company issued in April, due 2031, traded at a yield that passed 16%, a level Bloomberg reported is typically associated with financial distress. The cost of insuring CoreWeave’s debt against default, measured by credit-default swaps, rose to roughly 1,000 basis points, implying a nearly 63% probability of default within five years, up from about 30% a year earlier.

Chief executive Mike Intrator told staff in a memo that the company’s “deal backlog more than covers its debt obligations,” and that financial markets can sometimes “misunderstand and act irrationally.” CoreWeave carries roughly $50 billion in debt against an expected $13 billion in revenue this year, alongside a reported $2.13 billion loss.

The slide was not isolated to CoreWeave. The broader semiconductor index had erased some $1.6 trillion in market value over six trading sessions, and Gartner research cited in the same coverage suggested roughly 80% of so-called “neocloud” GPU-rental firms could shut down by 2030 if current spending and financing patterns continue, with warnings that over $3.5 trillion of AI-infrastructure investment could face impairment.

The episode is one marker of a wider argument running through the AI-infrastructure buildout in 2026: whether debt-financed datacentre and GPU capacity is being matched by durable revenue, or whether the financing structure itself — bonds, leases and vendor-financed deals — is more fragile than the headline contract backlogs suggest. CoreWeave’s own figures and management’s public response capture both sides of that argument without settling it.