CoreWeave’s chip-backed AI debt boom

A $2.3 billion financing arranged for CoreWeave in 2023 by Blackstone and Magnetar Capital has become the reference point for a new model of chip-backed lending that could reach about $1 trillion in size. The deal provided CoreWeave, an AI-focused cloud provider, with capital to purchase Nvidia (NVDA) GPUs while using those chips and key customer contracts as collateral, turning semiconductor assets into a distinct class of security for lenders.

CoreWeave’s need for capital stemmed from large cloud-computing contracts, including with Microsoft (MSFT), that required significant investment in Nvidia (NVDA) hardware. By structuring the loan so it was secured by both the GPUs and CoreWeave’s agreements with Microsoft (MSFT), Blackstone and Magnetar gave creditors tangible assets and cash flows to rely on beyond the company’s own balance sheet. The structure is reportedly helping to establish a broader chip-backed financing market for AI infrastructure.

Following the initial $2.3 billion deal, CoreWeave attracted substantial additional capital using similar collateral arrangements. The company secured a $7.5 billion facility in 2024 that again involved Blackstone and Magnetar, underscoring continued institutional appetite for GPU-backed lending. By March 2026, it closed an $8.5 billion facility led by Blackstone Credit & Insurance, further scaling this financing approach.

CoreWeave’s March 2026 borrowing terms showed a marked improvement in funding costs compared with earlier deals. Part of the $8.5 billion facility carried an interest rate of 5.9%, compared with rates in the mid-teens on earlier financing in 2023. Later transactions, including a $3.1 billion and a $2.6 billion delayed-draw term loan, followed the same pattern of securing borrowings against GPUs and customer contracts, helping to entrench the structure across the company’s capital stack.

By the end of June 2026, CoreWeave owed about $35.6 billion in debt. Roughly half of this amount was secured against specific assets, mainly the GPUs it uses for AI workloads and the customer contracts linked to those resources, while the remainder consisted of unsecured obligations. The secured facilities have been relatively stable, reflecting the priority lenders’ claims on hardware and contracted revenues.

The unsecured side of the structure has been more volatile. CoreWeave’s unsecured notes have been trading at prices that imply yields of around 15%, levels often associated with higher-risk or distressed credit. One issue, carrying a 9.75% coupon and maturing in 2031, has fallen sufficiently in price that new buyers can earn a yield closer to 16%, and these notes are largely held by traditional high-yield and multi-sector bond managers rather than the private-credit lenders that dominate the secured facilities.

CoreWeave’s leadership has emphasized the company’s long-term contracted revenue as a key support for its growing debt load. Chief executive Michael Intrator has stated in an internal memo that the company’s backlog "more than supports" its debt obligations and described periods when a new type of technology company may be misunderstood by the market. CoreWeave points to a contracted backlog of roughly $104 billion as the main offset to its leverage.

This combination of large, secured GPU-backed facilities, a sizable unsecured layer, and a substantial contracted backlog illustrates the new financing model emerging around AI infrastructure. The trajectory from the original $2.3 billion Blackstone- and Magnetar-led loan to successive multibillion-dollar facilities has helped turn high-end chips and AI service contracts into collateral at scale, anchoring a chip-backed debt market that is rapidly expanding alongside demand for AI computing.