Why early GPU investors are turning to inference chips in a $400 million deal

General Compute, an AI cloud startup, has secured a $400 million loan from Upper90, a technology investment firm. It may be the first deal to put special chips as collateral – chips designed to run AI models that have already been trained faster and more efficiently, than more expensive chips used to build the models in the first place.

The funding is the latest sign that markets are responding to concerns about the value of AI tools and tokens by turning to infrastructure that uses open source models more cheaply than new LLMs from frontier labs.

Founded by CEO Finn Puklowski, General Compute raised $15 million in seed funding in May to build a neocloud for inference around silicon from SambaNova, a chipmaker backed by Intel. (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 consideration. They are energy efficient and do not require expensive water cooling systems, meaning they can be installed much faster than GPUs in a large variety of data centers. General Compute says the new chips will provide 16 times faster response than GPU-based clouds.

The challenge is to get more chips, especially if you are a new company.

The founder of Upper90 and CEO Billy Libby, a former Goldman Sachs statistician, had a playbook for this: In 2021, his company financed the purchase of a GPU by Crusoe, a power-oriented data center startup, which he believed was the first loan against the number of advanced chips.

Traditional lenders avoided such deals at the time because of the risks and uncertainties surrounding GPU depreciation. But since CoreWeave made chip-backed loans a business model and then the basis of a blockbuster IPO, this type of financing has become commonplace.

“When we funded Nvidia GPUs as the first group to do that, the market wasn’t doing well,” Libby told TechCrunch. “We can put something together as an early adopter, and be compensated for the risk.”

Now that GPUs are relatively well understood and perhaps overbought, 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 looking for a player last year that was considered,” Libby said. “Everyone doesn’t need a supercomputer, but they do need inference and AI.”

That thesis has been gaining momentum, with companies that provide access to open models, such as OpenRouter and Fireworks, expanding new cycles on a large scale. New models like Kimi’s K3, as recently as this week, have proven to compete with recent releases from Anthropic and OpenAI in coding benchmarks. And new chip makers like Groq and Cerebras have found interest in investors and public markets alike.

General Compute’s ability to access chips outside of Nvidia’s ecosystem is important for the same reason. TensorWave, another AI infrastructure company, is making a similar bet in partnership with AMD. As alternatives to Nvidia emerge, computer suppliers not locked into Nvidia’s deals may have an opportunity to provide a more cost-effective solution.

“There are a lot of chips that are starting to scale that are amazing [total cost of ownership]or that can work much faster than Nvidia, but there are not too many buyers for them,” said Puklowski. Like, this is the first sign of financial self-organization and the separation of Nvidia’s independent monopoly.”