General Compute Lands $400M Loan for Inference Chips, Signaling Shift in AI Infrastructure Financing

General Compute, an AI inference cloud startup, has secured a $400 million loan from Upper90, a tech investment firm, in what is believed to be the first major financing deal to use inference-specific chips as collateral. The SN50 chips, designed by Intel-backed SambaNova, are purpose-built for running already-trained AI models efficiently, offering General Compute a path to compete with GPU-heavy rivals on cost and performance.

The loan structure mirrors the early GPU-backed deals that fueled CoreWeave's rise, but with a crucial difference: the underlying hardware is optimized for inference, not training. Upper90, which earlier financed GPU purchases for Crusoe, is betting that inference—the process of using AI models in production—will become the dominant workload as open-source models proliferate. “Everyone doesn’t need a supercomputer, but they do need inference and AI,” said Upper90 co-founder and CEO Billy Libby.

General Compute CEO Finn Puklowski claims the SN50 chips deliver 16x faster inference than GPU-based clouds, thanks to power efficiency and the elimination of expensive water-cooling. The startup raised a $15 million seed round in May and plans to deploy the new capital to scale its neocloud infrastructure rapidly. The deal also underscores a growing appetite for non-Nvidia alternatives: TensorWave recently made a similar bet with AMD, and inference-focused chipmakers like Groq and Cerebras have drawn acquisition interest.

As more chips from SambaNova, AMD, and others enter the market, capital markets are beginning to organize around them, threatening Nvidia's near-monopoly in AI compute. “This is the first signal of capital organizing itself and the fragmenting of Nvidia’s monopolistic dominance,” Puklowski said.