Key Takeaways

  • CoreWeave’s roughly $50 billion debt load is testing investor confidence in debt-funded AI infrastructure.
  • A $104.2 billion revenue backlog signals strong demand, but cash flow, refinancing costs, and GPU replacement cycles remain critical.
  • Rising electricity use and potential asset impairments could intensify pressure across the specialized GPU-cloud market.

CoreWeave’s bond selloff is turning the Nvidia-backed cloud computing specialist into a test case for the economics of the AI infrastructure boom. With debt of about $50 billion, CoreWeave faces growing scrutiny over whether its contracted revenue can produce cash quickly enough to cover interest, build data centers, and replace expensive GPUs as newer hardware arrives.

The warning from credit markets is hard to miss. Yields on some CoreWeave bonds due in 2031 have risen above 15% to 16%, while the cost of five-year default protection has approached 1,000 basis points. Those levels indicate that investors see substantial credit risk, even as demand for AI computing capacity remains elevated.

CoreWeave does have an unusually large commercial cushion. The provider reported a $104.2 billion revenue backlog at the end of the second quarter, according to IDNFinancials. Yet backlog is not the same as immediately available cash. Revenue recognition can stretch across several years, and delivering contracted capacity typically requires more spending on chips, facilities, networking equipment, cooling, and power.

CoreWeave’s central challenge is not simply finding customers; it is matching the timing of customer payments with a heavy and recurring capital burden. If financing costs stay high, projects that looked attractive when debt was cheaper can generate thinner returns. Refinancing existing obligations may also become more difficult, particularly if lenders demand stronger collateral or tighter contractual protections.

That tension reaches beyond CoreWeave. IDC forecast global AI-infrastructure spending of $318 billion in 2025, up from $153 billion in 2024, and projected $487 billion in 2026. The increase reflects real demand from model developers, enterprises, and cloud operators. It also creates a crowded race in which CoreWeave, Nvidia, Nebius Group, and other participants are committing capital before the long-term revenue profile is fully settled.

What happens if AI workloads grow, but not fast enough to support all the capacity being financed? Some assets could operate below expected utilization, reducing cash generation. Under the U.S. GAAP impairment model, businesses generally assess whether the carrying value of long-lived assets remains recoverable when adverse conditions or changes in expected performance emerge. Gartner recently estimated that more than $3.5 trillion of AI spending could face impairment by 2030 if revenue fails to support returns, a metric that could become highly relevant for servers or facilities whose economics deteriorate faster than anticipated.

Hardware turnover adds another wrinkle. GPUs can remain useful for years, but customers often prefer newer systems that deliver better performance or efficiency. CoreWeave may therefore need to refresh equipment while continuing to service debt tied to earlier generations. Used hardware can be redeployed, although its value depends on software compatibility, workload requirements, and secondary-market demand.

Power is an equally important constraint. The International Energy Agency reported that global data-center electricity consumption reached 485 TWh in 2025 and could rise to 950 TWh by 2030, with AI-focused facilities expected to triple their electricity use over that period. Access to generation, grid connections, and cooling infrastructure can influence whether a planned cluster comes online on schedule. Delays still carry financing costs.

Operational efficiency could soften some of that pressure. The Green Grid’s Power Usage Effectiveness metric compares a facility’s total energy consumption with the energy delivered to computing equipment. A lower PUE can improve operating economics, though it does not remove exposure to electricity prices, GPU utilization, or debt costs.

For enterprise customers, CoreWeave’s credit signals may warrant closer examination of contract terms, capacity portability, and contingency plans. For investors, the focus is likely to remain on free cash flow, customer concentration, refinancing schedules, and utilization rates. Strong AI demand helps, but the bond market is signaling that revenue growth and durable returns are not necessarily the same thing.