USD.AI Secures Record $128.9M Asset-Backed Financing for GPU Deployment

USD.AI announced on September 23, 2026 a $128.9 million asset-backed GPU financing facility for an undisclosed borrower, a transaction the company described as the largest loan originated through its platform to date. The facility supports the deployment of 32 NVIDIA GB200 NVL72 systems in British Columbia, Canada.

The borrower is a publicly listed GPU cloud provider that runs large-scale NVIDIA clusters as a managed service for enterprise customers, with deployments across North America, South America, and Europe, according to the announcement. The 32 systems are contracted under a multi-year agreement with what USD.AI described as a blue-chip investment-grade counterparty. The company said the transaction surpasses its previous largest facility, a $98.1 million GPU financing facility announced in June 2026.

The Financed Hardware

NVIDIA’s GB200 NVL72 is a rack-scale, liquid-cooled system that connects 36 Grace CPUs and 72 Blackwell GPUs into a 72-GPU NVLink domain, a design NVIDIA describes as acting as a single massive GPU and calls an exascale computer in a single rack. Each GB200 Grace Blackwell Superchip in the system pairs one Grace CPU with two Blackwell GPUs through the NVLink-C2C interconnect.

NVIDIA’s published specifications list 13.4 TB of HBM3E GPU memory with 576 TB/s of memory bandwidth, 130 TB/s of NVLink bandwidth across the domain, 2,592 Arm Neoverse V2 CPU cores, and 17 TB of LPDDR5X CPU memory. The system carries a second-generation Transformer Engine that enables FP4 precision for inference workloads, paired with fifth-generation NVLink providing 1.8 TB/s of GPU-to-GPU interconnect bandwidth.

NVIDIA claims the system delivers 30 times faster real-time inference on trillion-parameter large language models than its H100 GPU, four times faster training, and 25 times greater energy efficiency. NVIDIA publishes the test configurations behind those comparisons, including input and output sequence lengths and cluster size, and marks the performance figures as projected and subject to change.

Loan Structure and the Compute Middle Market

USD.AI said the transaction supports its thesis that the fastest-growing demand for credit lies in what it calls the compute middle market, a segment where significant AI infrastructure financing demand is met by a relatively limited pool of dedicated capital providers.

In an essay published on USD.AI’s website in July 2026, Conor Moore, chief operating officer and co-founder of Permian Labs, the developer of USD.AI, defines that middle market as financings of $50 million to $250 million. Moore wrote that over an eight-month span USD.AI saw $13 billion in deals across more than 200 neoclouds in the segment, which he estimated at $2 trillion in spending through 2030.

The essay describes the structural protections applied to these financings: a bankruptcy-remote special purpose vehicle with an independent manager; customer contracts assigned to the vehicle, with bank accounts governed by a deposit account control agreement; a first-priority charge over the GPUs plus a pledge of the vehicle’s equity, perfected through a Uniform Commercial Code filing; step-in rights to the colocation facility, including the right to sell the servers in an event of default; and a debt-service reserve account funded upfront with one to three months of principal and interest.

Sponsors typically put 20 to 30 percent cash down, Moore wrote, and he argued that the resulting packages carry net coupons 400 to 600 basis points above investment-grade-backstopped deals while maintaining comparable structural protections.

Earlier Facilities and Platform Scale

The earlier facility, a $98.1 million three-year debt financing announced on June 5, 2026, supported the deployment of 2,304 NVIDIA B300 GPUs. That infrastructure is operated by Duos Edge AI, a subsidiary of Duos Technologies Group, and managed by Hydra Host through its Brokkr platform. The June facility was structured as non-recourse, non-dilutive, and off-balance-sheet, secured by the underlying GPUs and associated offtake contracts.

On the capital side, digital asset platform Bullish announced on August 28, 2026 a $100 million stablecoin-based liquidity facility for USD.AI, with plans to onboard the sUSDai token across multiple Bullish Exchange trading pairs supported by a dedicated market-making program.

USD.AI’s website reports $609 million in total value locked, a $183 million active loan pipeline, more than 80 partnerships, and 77,057 active users, and describes its USDai stablecoin as collateralized by GPUs and compute infrastructure.