GPU lender pivots to inference chips in $400M deal

General Compute, founded by CEO Finn Puklowski, raised a $15 million（約24億円） seed round in May to build an inference neocloud—infrastructure purpose-built for AI workloads, unlike the general-purpose systems offered by traditional hyperscalers like AWS or Azure. The company has now secured a $400 million（約640億円） loan from Upper90, a tech investment firm, collateralized by SambaNova's SN50 inference chips.

The chips are designed specifically to run already-trained AI models quickly and efficiently. According to General Compute, they deliver 16 times faster inference than GPU-based clouds and are power-efficient enough to operate without expensive water-cooling systems, allowing deployment across a much wider range of data centers. The SN50 chips come from SambaNova, an Intel-backed chipmaker, giving General Compute access to silicon outside the Nvidia ecosystem.

Upper90's financing decision carries strategic weight. The firm's co-founder and CEO, Billy Libby, a former Goldman Sachs quantitative trader, previously financed GPU purchases by Crusoe Energy in 2021—a move he believes was the first loan against the value of advanced chips. At that time, traditional lenders avoided such deals due to uncertainty around GPU depreciation. The market shifted dramatically after CoreWeave built a business model around chip-backed loans and then went public with a blockbuster IPO, making this form of financing routine. "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."

With GPUs now mature and potentially over-supplied, Upper90 is targeting the next opportunity: open-source model inference. "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." This thesis aligns with broader market trends. Companies providing access to open models, like OpenRouter and Fireworks, have raised new rounds at substantial valuations. New models such as Kimi's K3 have demonstrated competitive performance against recent releases from Anthropic and OpenAI on coding benchmarks. Alternative chipmakers like Groq and Cerebras have attracted interest from acquirers and public markets. General Compute's Puklowski framed the Upper90 deal as historic: "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." Other infrastructure companies, notably TensorWave, are pursuing similar strategies with alternative suppliers like AMD, suggesting a broader shift toward non-Nvidia ecosystems as cost concerns reshape the AI infrastructure market.

The $400 million（約640億円） loan to General Compute represents a deliberate pivot by Upper90 away from GPU financing toward inference-specific hardware. Upper90's CEO Billy Libby, a former Goldman Sachs quantitative trader, previously pioneered chip-backed lending in 2021 when he financed GPU purchases for Crusoe Energy—a deal that faced skepticism from traditional lenders worried about GPU depreciation and market uncertainty. That strategy proved prescient: CoreWeave later built a business model around chip-backed loans and achieved a blockbuster IPO, normalizing this form of financing across the industry.

Now that GPU markets are mature and potentially oversupplied, Upper90 is positioning itself ahead of the next wave by backing companies focused on inference at a time when the open-source AI model ecosystem is growing stronger. Companies like OpenRouter and Fireworks are raising at significant valuations, and new models like Kimi's K3 are demonstrating competitive performance against releases from Anthropic and OpenAI. General Compute's bet on SambaNova silicon—rather than Nvidia—matters because it sidesteps dependence on a single dominant supplier. Competitors like TensorWave are making similar moves with AMD partnerships, suggesting a broader fragmentation of Nvidia's market position as customers seek cost-efficient alternatives.