General Compute secures $400 million loan backed by inference chips
As the market for GPU-backed financing matures, Upper90 pivots to support General Compute’s deployment of SambaNova hardware, betting on cost-effective open-source model execution.
General Compute, an artificial intelligence inference cloud startup founded by Finn Puklowski, has secured a $400 million loan from tech investment firm Upper90. The transaction marks a significant development in the sector, reported as the first deal to utilise inference-specific chips as collateral. This represents a distinct pivot from the previous market dominance of graphics processing units, which have long served as the primary asset class for AI infrastructure financing.
The funds will be deployed to install SambaNova SN50 chips, hardware designed to run pre-trained AI models efficiently without the need for expensive water-cooling systems. General Compute claims these new chips deliver inference speeds 16 times faster than GPU-based clouds. The company previously raised a $15 million seed round in May to build an inference neocloud around silicon from SambaNova, an Intel-backed chipmaker, distinguishing its purpose-built infrastructure from the general-purpose offerings of traditional hyperscalers.
Upper90, led by co-founder Billy Libby, is shifting its focus towards inference infrastructure as the market for GPU-backed financing matures. Libby, a former Goldman Sachs quantitative trader, previously financed GPU purchases by Crusoe in 2021, a deal believed to be the first loan against the value of advanced chips. At the time, traditional lenders avoided such transactions due to risks surrounding hardware depreciation, but the model gained traction after CoreWeave commercialised chip-backed loans and pursued an initial public offering.
The financing reflects a broader market response to concerns regarding the high costs of AI tools and tokens. Upper90 is targeting infrastructure that runs open-source models more cheaply than the latest large language models from frontier labs. Libby noted that while everyone does not require a supercomputer, there is a substantial need for inference capabilities. This thesis is supported by recent funding rounds for companies providing access to open models, such as OpenRouter and Fireworks, which have raised capital at significant valuations.
General Compute’s ability to access chips outside the Nvidia ecosystem is viewed as a strategic advantage. Puklowski stated that the partnership with Upper90 signals capital organising itself around the fragmentation of Nvidia’s dominance. Other firms, such as TensorWave, are making similar bets on partnerships with AMD, suggesting that compute providers not locked into Nvidia deals may hold an advantage in providing cost-efficient inference services.
