What GPU-backed lending is

GPU-backed lending is financing secured against the accelerators and clusters it pays for, rather than against the borrower's general credit. The hardware is the collateral, and its resale value is what the loan ultimately rests on.

How it works

A lender advances capital against a pool of GPUs at a discount to appraised value. Reported advance rates for recent facilities cluster around 50 to 70 percent, over terms of a few years, sized to stay inside the hardware's expected useful life. If the borrower defaults, recovery depends on what that hardware is worth at the moment it has to be sold.

That last clause is the whole underwriting problem. The loan is only as sound as the lender's view of resale value, and that view has to survive credit committees, auditors, and a refinancing two years out.

How large it has become

The market crossed from experiment to structure in August 2023, when CoreWeave raised $2.3 billion in a facility led by Magnetar and Blackstone, secured by NVIDIA H100s. It was the first GPU-collateralized financing at that scale, and the company's total debt has since grown past $20 billion in its public filings.

The direction of travel is larger still. JLL estimates the data-center buildout will require roughly $870 billion of new debt financing over the next five years. Hardware will not secure all of it, but credit at that scale needs collateral it can value, and GPUs sit at the center of the spend.

Where lenders hesitate

GPU values move with supply, power availability, refresh cycles, and deployment timing, and most of the evidence lives in private conversations. That makes internal marks hard to defend and recovery assumptions hard to explain.

The speed of the AI cycle magnifies the gap. Capital wants terms long enough to make infrastructure projects work, while hardware economics can shift inside a single product generation. For a bank or private credit fund, the hard question is what the collateral is worth if deployment is delayed, the use case changes, or the borrower needs to refinance.

Why it depends on price discovery

Every one of these deals requires a view on what the collateral is worth, today and at maturity. Yet no market-wide reference exists for the underlying hardware. Lenders, insurers, and the funds behind them mark exposure off internal estimates and whatever quotes they can gather.

Stoa Markets is built for that gap. When qualified participants transact in one venue, lenders can underwrite GPUs the way they underwrite other real assets, and hardware value becomes something to point to rather than argue for.