USD.AI has closed a $128.9 million GPU financing facility, its largest deal so far. The borrower is a publicly listed GPU cloud provider in British Columbia, Canada, which USD.AI didn’t name. The loan finances 32 NVIDIA GB200 NVL72 systems. They’ll run under a multi-year contract with an investment-grade customer.
The deal beats the $98.1 million facility USD.AI announced in June.
Lending against the chips themselves
USD.AI makes non-recourse loans secured only by the GPU hardware. If the borrower runs into trouble, the lender’s claim is on the GPUs and the income they earn, not on the rest of the company. Lenders get direct exposure to income-producing compute. Borrowers can finance hardware without putting their whole balance sheet on the line.
The company funds these loans partly from crypto-native capital. It has a $40 million revolving facility from K3 Capital and $100 million from Bullish.
The compute middle market
USD.AI targets what it calls the compute middle market. These are GPU clouds too big for small loans and too small to get the massive credit lines hyperscalers and the largest neoclouds can arrange. Demand for financing there is huge, and few lenders serve it.
Building GPU clouds takes a lot of borrowed money. Each GB200 NVL72 rack costs millions, and newer platforms are already shipping. How lenders price that risk, and how fast they expect the hardware to lose value, is one of the key questions in the AI build-out. USD.AI is betting that a signed multi-year contract with a strong customer makes the collateral bankable.