Who makes GPU-backed lending for AI data centers

Accelerators are expensive, they hold value while demand exceeds supply, and they can be identified and repossessed — which makes them unusual among technology assets and is why lending secured against them has become a distinct financing structure. A borrower pledges a pool of GPUs, and often the customer contracts running on them, and borrows against that rather than against a corporate balance sheet. The structure is what has allowed operators without large balance sheets to build at scale quickly, and it is the financial mechanism sitting behind a meaningful share of the buildout catalogued in the layers below. It is closest in form to equipment or asset- backed lending in aircraft and shipping, and it depends on the same two things: that the collateral holds value, and that the contracted cash flows behind it are good. Both of those are assumptions rather than facts, and this catalog should say so plainly. Residual value depends on how quickly successive accelerator generations displace the current one, which is a technology forecast; contract quality depends on the creditworthiness of a small number of counterparties, several of which are themselves financed the same way. What to look for is disclosed structure rather than headline facility size — advance rates, term against expected hardware life, and whether the contracts pledged extend beyond the loan. Nothing here is a view on any of it; it is a description of how the capacity is being paid for.