General Compute, an AI inference cloud startup, has secured a $400 million loan from tech investment firm Upper90. This could be the first transaction to offer a dedicated inference chip as collateral. That means chips built to run already trained AI models quickly and efficiently, rather than expensive chips used to build models in the first place.
The funding is the latest signal that the market is responding to concerns about AI tools and token pricing by focusing on infrastructure that runs open source models cheaper than Frontier Labs’ latest LLM.
General Compute, founded by CEO Finn Puklowski, raised $15 million in a seed round in May to build a silicon-centric inference neocloud for Intel-backed chipmaker SambaNova. (Neocloud is purpose-built for AI workloads, unlike the general-purpose infrastructure provided by traditional hyperscalers like AWS and Azure.)
The company’s SN50 chip is designed for inference. They are more power efficient and do not require expensive water-cooling systems, so they can be deployed in different data centers more quickly than GPUs. According to General Compute, the new chip offers 16 times faster inference than GPU-based clouds.
The challenge, especially for startups, is getting these chips in large quantities.
Billy Libby, co-founder and CEO of Upper90 and a former Goldman Sachs quantitative trader, had a strategy for this. In 2021, his company funded a GPU purchase by energy-focused data center startup Crusoe, which he believes was the first loan to take advantage of the value of advanced chips.
At the time, traditional financiers avoided such transactions due to the risks and uncertainties surrounding GPU depreciation. But this type of financing has become commonplace since CoreWeave made chip-backed loans a business model and the basis for a blockbuster IPO.
“When we were the first group to fund Nvidia GPUs, the market was inefficient,” Libby told TechCrunch. “We were able to actually build something as an early participant and get some compensation for that risk.”
Now that GPUs are relatively well understood and perhaps overbought, Upper90 is looking to companies like General Compute to ride the next wave of the AI boom. “We think the open source model is going to be important, and last year we went looking for players in inference,” Libby said. “Not everyone needs a supercomputer, but they do need reasoning and AI.”
That idea is growing stronger as companies like OpenRouter and Fireworks that provide access to open models raise new rounds at huge valuations. New models like Kim’s K3, which launched just this week, have proven competitive in coding benchmarks with the latest releases from Anthropic and OpenAI. And emerging chipmakers like Groq and Cerebras are attracting interest from acquirers and the public markets alike.
For the same reason, it’s also important that General Compute has access to chips outside the Nvidia ecosystem. TensorWave, another AI infrastructure company, is making a similar bet on its partnership with AMD. As more alternatives to Nvidia emerge, compute providers that are not locked into contracts with Nvidia may have an advantage in providing cost-effective inference.
“There are a lot of chips that have incredible (total cost of ownership) and run much faster than Nvidia that are starting to scale, but there aren’t a lot of buyers for them,” Puclovski said. “By working with Upper90, this is more than just ‘cool startup got money to buy computing.’ Similarly, this is the first sign that capital itself is organizing and that NVIDIA’s monopoly control is fragmenting. ”
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
