The $400 Million Inference Bet
General Compute just pulled off something that sounds like financial alchemy: a $400 million loan from Upper90, backed entirely by inference-specific chips. It might be the first deal of its kind to use inference chips as collateral — hardware designed to run already-trained AI models fast and lean, rather than the ridiculously expensive GPUs used to build them in the first place.
The founders are CEO Finn Puklowski and CTO Jason Goodison. They raised a $15 million seed round in May 2026 to build what they call an inference neocloud, running on silicon from SambaNova, an Intel-backed chipmaker. Neoclouds aren't your typical hyperscaler infrastructure. They're purpose-built for AI workloads, unlike the general-purpose compute offered by AWS or Azure.
This isn't just another startup raising venture cash. It's a signal that the market is starting to respond to the growing pain of AI tool and token costs by pivoting toward infrastructure that runs open source models far more cheaply than the newest frontier LLMs from the big labs.
What Makes Inference Chips Different
General Compute's SN50 chips — made by SambaNova — are engineered for one thing: inference. They're power-efficient. They don't require expensive water-cooling systems. That means they can be deployed much faster than GPUs across a wider variety of data centers, which matters when you're trying to scale.
The company claims its SN50 chips deliver 16 times faster inference than GPU-based clouds. That's not a marginal improvement. That's a fundamental shift in how you think about deploying AI at scale.
The challenge, as Puklowski puts it, is getting your hands on a lot of these chips, especially when you're a brand-new company. Most buyers are locked into Nvidia deals. The secondary market for inference chips is thin. That's exactly why this Upper90 financing matters.
Upper90's Playbook: From GPUs to Inference
Upper90 co-founder and CEO Billy Libby brings a quant trader's playbook to this problem. A former Goldman Sachs quantitative trader, Libby's firm financed GPU purchases by Crusoe — the energy-focused data center startup — back in 2021. He believes that was the first loan ever made against the value of advanced AI chips.
Traditional lenders walked away from such deals at the time. The risks were enormous. The uncertainties around GPU depreciation were real. But CoreWeave eventually turned chips-backed loans into a full business model, then used it as the foundation for a blockbuster IPO. Suddenly, this kind of financing wasn't a curiosity. It was mainstream.
"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 its attention to companies like General Compute. The thesis is straightforward: "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."
Open Source Models Driving the Shift
The broader market is already voting with its wallet. Companies providing access to open models — OpenRouter and Fireworks, among others — have raised new funding rounds at huge valuations. The writing's on the wall: open source isn't catching up to frontier models. In some cases, it's already competing with them.
Kimi's K3, for instance, has proven it can compete with the latest releases from Anthropic and OpenAI on coding benchmarks. That's significant. It means you don't need to pay premium prices for frontier LLM access if your workload is well-suited to open models. And open models run cheaper on specialized inference hardware.
New chipmakers like Groq and Cerebras have also drawn interest from acquirers and public markets alike. The infrastructure landscape is fragmenting, and General Compute is positioning itself squarely in that opening.
A Fragmenting AI Infrastructure Market
General Compute's ability to access chips outside of Nvidia's ecosystem matters for the same reason that TensorWave — another AI infrastructure company — is making a similar bet on a partnership with AMD. Compute providers that aren't locked into Nvidia deals may have a structural 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."
That last line is worth sitting with. For years, Nvidia's stranglehold on AI infrastructure has been unshakable. This deal suggests that's starting to change. Capital is organizing around inference-specific hardware. Open source models are gaining ground. And the market is beginning to price in alternatives.
It's not clear how fast this shift will accelerate. But at $400 million, General Compute has proven that the market is willing to bet big on the inference future. The question now is whether other players will follow.
