GPU-Backed Lending: What Credit Teams Should Underwrite

GPU-backed loans and data center financings are a fast-growing credit market. The collateral behaves unlike real estate or traditional equipment. Here is what to underwrite.

Lending against GPUs has moved from a niche to a meaningful part of private credit. The structures look familiar, often delayed-draw term loans or equipment facilities secured by hardware and the contracts that pay for it, but the collateral behaves differently from anything most credit teams have underwritten before.

The collateral is the contract, not just the chip

A GPU cluster without a customer is a depreciating asset with a narrow buyer base. The real security is usually the offtake: a compute contract with a creditworthy customer. Underwriting starts with that contract.

• Who is the counterparty, and what is its credit quality?
• How long is the term relative to the loan, and can the customer terminate early?
• Are payments fixed or tied to usage?
• Can the contract be assigned to the lender, and is it in the security package?

Re-leasing risk

When the initial contract ends, can the hardware be re-leased, and at what price? That depends on how quickly newer generations arrive, how the hardware has been maintained and whether the facility can support the next generation's power and cooling needs. Rating agencies have published frameworks for this risk; a credit team should score each financing against one consistently.

Depreciation and residual value

Accounting useful lives for GPUs vary by company, and market values can move faster than book values. Underwrite to a conservative residual curve, test the loan against a sharper decline, and size amortization so the loan is well covered before the next generation is widely available.

Covenants read from the agreement, not the summary

• Debt service coverage and how it is calculated, including any add-backs.
• Customer concentration limits.
• Collateral coverage tests and what happens when they are breached.
• Restrictions on moving hardware between facilities or customers.

Press releases summarize these terms; the credit agreement defines them. Differences between the two are common and material.

Surveillance after close

GPU credits need closer monitoring than most equipment loans: utilization, customer payment history, hardware condition, power availability and secondary market pricing. A surveillance plan tied to each facility's own covenants turns early warning signs into action before a default.

Start from primary sources

Comparable deals are the best guide to terms and pricing, but much of what circulates is secondhand. Building comps from SEC filings, issuer announcements and rating disclosures, and labeling each one as verified or reported, gives a credit committee numbers it can defend.