CoreWeave’s Distressed Debt Tests the Financing Behind the AI Boom
Yields of 15% to 16% on parts of CoreWeave’s debt are forcing lenders to reprice AI infrastructure risk.
CoreWeave’s bonds have fallen to distressed levels, turning one of artificial intelligence’s fastest-growing infrastructure companies into a live test of whether lenders priced the compute boom too generously. LSEG data cited by Reuters put the company’s debt near $50 billion, while the cost of default insurance climbed to about 1,200 basis points and yields on some bonds reached roughly 15% to 16%.
Those prices do not establish that CoreWeave will default. They show that investors now demand far more compensation for risks that were easier to overlook when capital was cheap and enthusiasm for AI capacity appeared limitless. The repricing matters because private lenders and structured vehicles have financed much of the data-centre expansion that public equity alone could not support.
CoreWeave illustrates the model. It raised a $2.3 billion facility backed by graphics processors in 2023, followed by $7.5 billion in 2024 and an $8.5 billion facility anchored by Blackstone Credit & Insurance in March 2026. The collateral can include GPUs, infrastructure assets and long-term customer contracts with Microsoft, OpenAI and Meta.
Contract backing can make a loan safer than one dependent solely on resale values for chips that may become obsolete. It can also concentrate risk. If a small group of customers accounts for most contracted cash flow, repayment depends on those companies honouring and renewing commitments while their own AI economics remain uncertain.
The wider system is becoming exposed. Carlyle estimated in an October paper that private credit could finance about $1 trillion of the AI-compute buildout, more than half the industry's current assets under management. It warned that rapid lending could recreate concentration problems seen in software credit and urged lenders to set exposure limits before the buildout accelerates further.
Market discipline is already changing. Investors favour data-centre projects closer to completion and demand wider spreads for developments facing construction, power or customer hurdles. If elevated funding costs persist, lenders can require lower leverage, stronger covenants and tighter limits on technology and customer concentration.
This correction arrives when direct-lending funds have abundant cash but fewer attractive transactions. Funds raised a record $129.15 billion in the first nine months of 2026, up 35% year on year, while third-quarter lending volume fell 28% to $49.77 billion. Competition to deploy capital can weaken terms just when sector risk is rising.
The syndicated market adds another pressure. In the third quarter, $16.39 billion of direct loans were refinanced into broadly syndicated debt, compared with only $2.60 billion moving the other way. Better borrowers can leave private lenders for cheaper markets, potentially worsening the mix of opportunities that remain.
Why it matters
AI investment is increasingly a credit story, not just an equity valuation story. Data centres require huge upfront spending on land, power, cooling and chips before they produce revenue. If credit investors retreat, capacity plans can slow even when technology demand remains strong. If they continue lending without tighter standards, losses may accumulate in funds, insurers and structured vehicles.
For CoreWeave, the key variables are customer performance, project delivery, refinancing access and cash generation. For lenders, the episode tests whether collateral and contracts genuinely protect downside. For the AI supply chain, higher borrowing costs can separate projects with durable demand from speculative capacity.
The debt figures and market prices are snapshots, and individual facilities have different collateral and protections. Yet the repricing is material new information: markets are no longer treating AI infrastructure debt as a simple proxy for growth.
Sources: