AI Infrastructure Faces a Credit Test as CoreWeave Yields Top 16%

Lead: The AI infrastructure boom is still attracting enormous amounts of capital, but credit markets are becoming less forgiving. CoreWeave, a leading neocloud provider, has become the clearest example of that shift: the yield on its 2032 bond, which carries a 9.625% coupon, has risen above 16%. Investors are demanding a much larger return for holding debt linked to data-center expansion.

Key takeaways

• CoreWeave’s five-year credit default swaps briefly jumped above 1,289 basis points before easing about 8% to roughly 1,155 basis points.
• CDS costs rose 76% for Nvidia, 49% for Broadcom and 44% for Amazon over the past month, although their absolute spreads remain far below CoreWeave’s.
• Goldman Sachs counted $67 billion in AI-related bond deals since Oct. 29, taking hyperscaler debt issuance this year to $296 billion.
• Potential data-center restrictions in several states could affect as much as 45% of planned U.S. capacity, according to Goldman.
• Hyperscalers collectively reported negative free cash flow in the third quarter as capital spending exceeded operating cash flow.

Why credit markets are reacting

Data centers require heavy upfront investment, while the revenue generated by compute contracts, cloud services and AI applications may arrive over a longer period. Higher bond yields therefore have a direct impact on project economics. For companies that depend on prepayments, long-term contracts or repeated refinancing, a narrower financing window can also delay equipment purchases and construction.

The pressure is not limited to CoreWeave. Default-protection costs have risen across several major AI-linked issuers, suggesting that investors are reassessing the entire capital-spending chain. The percentage increases, however, should be interpreted carefully. A large move from a low base does not imply that a company has reached CoreWeave’s level of credit risk. The reported spreads were approximately 139 basis points for Nvidia, 298 for Oracle and about 1,100 for CoreWeave.

Policy and cash flow add uncertainty

Local politics are becoming another variable for data-center developers. Ohio, Texas and Georgia contain a significant share of planned U.S. capacity, and some incoming Democratic-led administrations may seek construction pauses or stronger community protections. Goldman estimates that a broad set of moratoriums could affect 45% of planned U.S. capacity. Federal officials continue to support data-center growth, so the immediate risk is more likely to be slower approvals and project delays than a sudden collapse in financing.

The more fundamental question is when AI investment will generate enough cash to support further borrowing. Goldman expects hyperscaler debt issuance of $440 billion next year, alongside $300 billion in project-level financing. Bank of America estimates that total AI-related credit issuance could reach $685 billion in 2027 before falling to $531 billion in 2028. If monetization takes longer than expected, interest costs will consume more of the returns from new capacity.

Why it matters

The credit-market move does not prove that AI demand has disappeared. It does show that the next stage of the boom will be judged on more than chip supply and server deployment. Capital costs, permits, contract quality and free cash flow will determine which projects can proceed.

For hyperscalers, higher yields could encourage a slower rollout, revised capital budgets or tighter purchasing commitments. For neoclouds, shorter contracts, smaller prepayments or delayed construction could quickly amplify liquidity pressure. The competitive question is therefore shifting from who can build the most compute to who can fund it efficiently and turn it into cash fastest.

Source: Hacker News