GPUs as Bank Guarantees: CoreWeave Secures $8.5 Billion to Expand

CoreWeave has closed a financial deal worth $8.5 billion, marking an important step in how infrastructure for artificial intelligence is financed. This is a specific type of loan, called a delayed draw term loan (DDTL), which is not disbursed all at once but in several phases as the company needs it.

This financing, termed DDTL 4.0, allows CoreWeave to initially access about $7.5 billion, with the potential to reach $8.5 billion when the infrastructure is fully operational. The uniqueness lies in the fact that the loan is secured not so much by the company itself but by the assets being used: in this case, GPUs and contracts with customers.

This is one of the most innovative aspects. The debt is backed by graphic chips used for AI and by commercial agreements that ensure their usage over time. Among the involved clients are also major companies like Meta, with contracts estimated to be worth at least $19 billion. These agreements generate predictable revenues, which are used to repay the loan.

Thanks to this structure, the financing has achieved an investment-grade rating (A3 from Moody's and A low from DBRS), a rare outcome for transactions of this nature. In practice, it is considered relatively safe, allowing CoreWeave to obtain better terms and attract larger investors, such as insurance companies and funds.

The loan is divided into two parts: one at a variable rate (linked to SOFR plus a margin of 2.25%) and one at a fixed rate around 5.9%. The maturity is set for 2032. Compared to previous deals, CoreWeave's cost of funding has dropped significantly.

This is important because the company is investing heavily to expand its AI infrastructure. Over the past 12 months, it has raised about $28 billion in debt and equity but has also accumulated over $21 billion in debts. Lowering financing costs is therefore crucial.

This operation is set against a backdrop of rapidly growing demand for AI infrastructure that requires massive investments, estimated in trillions of dollars over the coming years. Chips, in particular, represent one of the major cost components.

The market for GPU-backed loans is still in its early stages but is attracting increasing interest. The fact that this operation was requested beyond expectations indicates that investors believe in the model and the growth of artificial intelligence.