General Compute has locked in a $400 million credit line, using SambaNova's inference ASICs as collateral. The deal marks a clear departure from the standard practice of backing AI chip loans with Nvidia GPUs. It signals that lenders are now willing to bet on hardware designed specifically for running trained models, not just for training them.
A shift from training to inference
Most AI chip-backed lending has relied on Nvidia's graphics processing units, which dominate the training market. But the new credit line puts SambaNova's application-specific integrated circuits — chips built for inference — at the center. That's a significant change. Inference chips are optimised for low-latency, high-throughput execution of AI models after they've been trained. The deal suggests the financial industry sees real value in that part of the AI lifecycle.
Why SambaNova's ASICs fit the collateral mold
Lenders typically want assets that hold resale value. Nvidia GPUs have a proven secondary market, but inference ASICs are less common. SambaNova's chips are designed for large-scale inference in data centres, and the company has been winning contracts with enterprise clients. By accepting them as collateral, the credit line provider is betting that these chips will retain enough worth to cover the loan if General Compute defaults. That's a vote of confidence in both the hardware and the inference market.
The broader implications for AI infrastructure financing
The deal opens the door for other inference-focused chipmakers to seek similar financing. It also shows that the AI industry's financial backbone is expanding beyond the training-heavy GPU model. As more companies shift from developing models to deploying them, the demand for inference hardware is growing. Financing that hardware directly could accelerate that shift. General Compute, which builds and operates AI infrastructure, now has capital to scale its inference capacity without relying on Nvidia's ecosystem.
For now, the credit line is a single data point. But it's a concrete one. Lenders and investors will be watching to see whether other chipmakers — or other infrastructure providers — follow the same path. If they do, the collateral landscape for AI loans could look very different a year from now.
