Why the first GPU financiers are turning to inference chips in a $400 million deal

General Compute, an AI inference cloud startup, has landed a $400 million loan from Upper90, a tech investment firm. It might be the first deal to put up inference-specific chips as collateral — chips built to run already-trained AI models quickly and efficiently, rather than the more expensive chips used to build the models in the first place.

The financing is the latest signal that markets are responding to concerns over the price of AI tools and tokens by turning to infrastructure that runs open-source models more cheaply than the newest LLMs from frontier labs.

Founded by CEO Finn Puklowski, General Compute raised a $15 million seed round in May to build an inference neocloud around silicon from SambaNova, an Intel-backed chipmaker. (Neoclouds are purpose-built for AI workloads, unlike the general-purpose infrastructure offered by traditional hyperscalers like AWS or Azure.)

The company’s SN50 chips are designed for inference. They’re power-efficient and don’t require expensive water-cooling systems, which means they can be deployed more quickly than GPUs across a larger variety of data centers. General Compute says the new chips will provide 16 times faster inference than GPU-based clouds.

The challenge is getting a lot of these chips, especially when you’re a brand-new company.

Upper90 co-founder and CEO Billy Libby, a former Goldman Sachs quantitative trader, had a playbook for this: In 2021, his firm financed GPU purchases by Crusoe, the energy-focused data center startup, which he believes was the first loan against the value of advanced chips.

Traditional lenders eschewed such deals at the time because of the risks and uncertainties around GPU depreciation. But as CoreWeave made chips-backed loans into a business model and then the basis of a blockbuster IPO, this kind of financing has become common.

“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.”

Now that GPUs are comparatively well understood and perhaps over-bought, Upper90 is turning to companies like General Compute to ride the next wave of the AI boom. “We think open source models are going to be important, and we went and looked for a player last year that was in inference,” Libby said. “Everyone doesn’t need a supercomputer, but they do need inference and AI.”

That thesis has been growing stronger, with companies that provide access to open models, like OpenRouter and Fireworks, raising new rounds at huge valuations. New models like Kimi’s K3, recently just this week, have proven to compete with the latest releases from Anthropic and OpenAI on coding benchmarks. And new chipmakers like Groq and Cerebras have drawn interest from acquirers and public markets alike.

General Compute’s ability to access chips outside of Nvidia’s ecosystem matters for the same reason. TensorWave, another AI infrastructure company, is making a similar bet on a partnership with AMD. As more alternatives to Nvidia emerge, compute providers that aren’t locked into Nvidia deals may have an advantage in providing cost-efficient inference.

“There are a bunch of chips that are starting to scale that have amazing [total cost of ownership], or that can operate much faster than Nvidia, but there’s not too many buyers for them,” Puklowski said. “By getting together with Upper90, this is not just, ‘a cool startup got some money to buy some compute.’ Like, this is the first signal of capital organizing itself and the fragmenting of Nvidia’s monopolistic dominance.”