General Compute, an AI inference cloud startup, has obtained a $400 million loan from Upper90, a technology investment firm. The transaction is notable because it uses inference-specific chips as collateral — processors designed for running already-trained AI models quickly and efficiently, rather than the typically more expensive GPUs used to train models.
The financing reflects a broader market shift: investors and customers are reacting to concerns over the cost of AI tools and tokens by favoring infrastructure that can run open-source models more cheaply than the newest frontier large language models (LLMs).
The company and its technology
General Compute was founded by Finn Puklowski. In May the company raised a $15 million seed round to build an inference neocloud based on silicon from SambaNova, a chipmaker supported by Intel. The term "neocloud" here denotes infrastructure purpose-built for AI workloads, as opposed to the general-purpose services offered by hyperscalers like Amazon Web Services or Microsoft Azure.
General Compute’s SN50 chips are tailored for inference: they are power-efficient and do not require costly water-cooling systems, which enables faster deployment across a wider variety of data centers compared with GPUs. The company says the SN50 chips deliver 16x faster inference than GPU-based clouds.
Challenges and financing context
Sourcing large quantities of such chips is a key challenge, especially for a newly established company. Upper90 co-founder and CEO Billy Libby — formerly a quantitative trader at Goldman Sachs — has prior experience structuring this kind of financing: in 2021 his firm financed GPU purchases for Crusoe, an energy-focused data center startup, a deal Libby regards as among the first loans secured against advanced chips.
Traditional lenders had previously avoided these types of deals because of the depreciation risk and market uncertainties surrounding GPUs. But as firms like CoreWeave turned chip-backed lending into a business model and eventually into a high-profile IPO, chip-backed financing has become more mainstream.
Market implications and strategic importance
Now that GPU financing is better understood, Upper90 is shifting toward inference-chips to capture the next phase of AI demand. Libby argues that many users do not need supercomputers; they need cost-effective inference and AI access.
Other market participants are arriving at similar conclusions: companies that provide access to open-source models, such as OpenRouter and Fireworks, have raised new rounds at substantial valuations. New models — for example Kimi’s K3 — have recently shown competitive performance on coding benchmarks against releases from Anthropic and OpenAI. New chipmakers like Groq and Cerebras have also drawn interest from potential acquirers and public markets.
General Compute’s ability to use chips outside the Nvidia ecosystem is significant for the same reasons. TensorWave is making a similar strategic bet with a partnership involving AMD. As more Nvidia alternatives scale, compute providers not locked into Nvidia agreements may have an edge in offering lower-cost inference.
Finn Puklowski said there are multiple chips beginning to scale that offer strong total cost of ownership or can operate faster than Nvidia in certain scenarios, but demand from buyers remains limited. By partnering with Upper90, he argues, the deal signals more than capital for buying compute; it is an early sign of capital organizing around alternative hardware and the fragmentation of Nvidia’s dominant market position.
Summary
The General Compute–Upper90 loan illustrates a shift in AI infrastructure financing toward inference efficiency and non-Nvidia hardware, driven by demand for cheaper access to open-source models and by evolving lender comfort with chip-backed collateral.
