General Compute, an AI inference cloud startup, has secured a $400 million loan from Upper90, a tech investment firm. The deal may be the first to use inference-specific chips—built to run already-trained AI models efficiently—as collateral, rather than the more expensive GPUs used for training. The financing signals growing market demand for cheaper AI infrastructure that runs open-source models, as concerns over the cost of frontier AI tools intensify.

Founded by CEO Finn Puklowski, General Compute raised a $15 million seed round in May and built its inference neocloud around SambaNova‘s SN50 chips, which are power-efficient and don’t require water cooling. That allows faster deployment across diverse data centers. General Compute claims the SN50 chips deliver 16 times faster inference than GPU-based clouds. Upper90, which previously financed GPU purchases by data center startup Crusoe in 2021, is now turning to inference chips as the GPU market matures. Upper90 CEO Billy Libby noted that GPUs are now well understood, while inference and open-source models represent the next wave.

The deal underscores the fragmentation of Nvidia’s dominance. As open-model providers like OpenRouter and Fireworks raise large rounds, and alternative chipmakers like Groq and Cerebras attract attention, compute providers not locked into Nvidia deals may have an advantage for cost-efficient inference. General Compute‘s Puklowski framed the loan as the first signal of capital organizing around chip alternatives, moving beyond the ‘cool startup gets money to buy compute’ narrative.