General Compute secures $400M as chip-backed lending moves beyond GPUs

General Compute has secured a $400 million loan from Upper90 to finance specialised chips for running trained AI models. The structure extends a financing model pioneered around Nvidia GPUs into a newer and less proven class of inference hardware.

What happened

The startup plans to build its cloud infrastructure around SambaNova’s SN50 processors, which are designed primarily for inference rather than model training. The loan is notable because the chips themselves serve as collateral. General Compute previously raised a $15 million seed round and is still at an early stage compared with established AI-cloud providers.

Chip-backed lending has helped neocloud companies acquire large amounts of expensive hardware without funding every purchase through equity. Lenders are comfortable with Nvidia GPUs partly because they have broad demand and a relatively liquid resale market. General Compute is now asking lenders to make a similar bet on more specialised processors.

Why it matters

Inference—the repeated process of generating outputs from trained models—is becoming one of the largest cost centres in AI. Hardware designed specifically for inference could offer better economics than general-purpose GPUs for some workloads.

If the financing works, it could give startups another route to build compute capacity and create a new asset class for private credit. It could also help alternative chip companies compete with Nvidia by making their hardware easier for customers to finance.

The bigger picture

The key risk is collateral value. Specialised processors may depreciate quickly, support fewer workloads or attract fewer buyers if model architectures change. A lender can recover more easily from a failed GPU cloud because the hardware has many potential users; the same may not be true for inference-specific chips.

This deal is therefore both a compute bet and a financial experiment in how AI infrastructure will be funded.