How CoreWeave Actually Finances Its GPUs
And why a functioning derivatives market in AI compute will rewrite the entire capital stack
TL;DR: CoreWeave isn’t a cloud company. It’s a leveraged infrastructure vehicle that finances GPUs like power plants, collateralizes them like aircraft, and backstops them with customer prepayments that behave like short-term loans. Its most recent 10-Q shows an emerging asset class trying to bootstrap a capital structure without the markets it actually needs. A real derivatives market for AI compute would make this vastly more liquid, more transparent, and more stable. It would also make the next CoreWeave easier to fund.
I. The Reveal: CoreWeave’s 10-Q Reads Like Project Finance, Not SaaS
If you strip away the AI gloss, CoreWeave’s most recent quarterly filing looks nothing like a cloud computing company. It looks like a 1990s independent power producer that accidentally found itself at the center of the GPU supercycle. A few numbers tell the story.
Property & equipment: $20.7 billion. Of that, $14.6 billion is technology equipment. Translation: GPUs, networking gear, servers, the physical substrate of AI compute.
Construction in progress: $6.9 billion. They are still mid-build. The capex conveyor belt is running full speed.
Total debt: $14.2 billion. Up from almost nothing pre-2023.
Deferred revenue: $5.3 billion. Plus $50 billion in contracted future obligations from long-term customers.
Every dollar of the balance sheet is shouting the same thing: CoreWeave buys GPUs with other people’s money, then leases that capacity back to hyperscalers and training labs under long-dated, high-stakes contracts.
This is not AWS. This is ERCOT-with-H100s.
