Startup Uses Inference Chips as Collateral for $400M Loan

General Compute, a fledgling AI infrastructure company, just secured a $400 million loan from tech investment firm Upper90. The deal might be the first of its kind to use inference-specific chips as collateral — a notable shift in how capital markets are valuing specialized silicon. These aren’t the expensive GPUs used to train massive models; they’re chips from SambaNova designed to run already-trained AI quickly and cheaply.

The loan signals a broader market pivot. With concerns mounting over the astronomical cost of AI tools and tokens, investors are pouring money into infrastructure that can serve open-source models at a fraction of the price of frontier labs’ latest LLMs. General Compute, which raised a modest $15 million seed round in May, plans to build an ‘inference neocloud’ around SambaNova’s SN50 chips. CEO Finn Puklowski claims these power-efficient chips — which don’t need elaborate water cooling — will deliver 16 times faster inference than GPU-based clouds.

Upper90 CEO Billy Libby is no stranger to chip-backed financing. Back in 2021, his firm was an early mover in lending against GPUs, financing purchases for Crusoe. At the time, traditional lenders balked at the depreciation risk. Now, that model is practically mainstream, fueled by CoreWeave’s blockbuster IPO. “When we financed Nvidia GPUs as the first group to do that, the market was inefficient,” Libby told TechCrunch. “We could really put together something as an early participant, and kind of get compensated for the risk.” With GPUs now better understood and arguably over-bought, Upper90 is hunting for the next wave — and sees inference as the smarter bet.

The thesis is gaining serious traction. Open model providers like OpenRouter and Fireworks are raising at huge valuations, and new models like Kimi’s K3 are proving they can compete with Anthropic and OpenAI on coding benchmarks. Puklowski frames the deal as more than just a startup getting cash. “There are a bunch of chips that are starting to scale that have amazing total cost of ownership… but there’s not too many buyers for them,” he said. “This is the first signal of capital organizing itself and the fragmenting of Nvidia’s monopolistic dominance.”