source: techcrunch ai: why the first gpu financiers are turning to inference chips in a $400 million deal
level: business
general compute, an ai inference cloud startup, secured a $400 million loan from tech investment firm upper90. the deal uses inference-specific chips as collateral, a first for the industry. these chips run already trained models efficiently, unlike the pricier gpus used for training. the startup raised a $15 million seed round in may to build a cloud around sambanova's sn50 chips, which are power-efficient and avoid costly water cooling. this allows faster deployment across more data centers, with claimed inference speeds 16 times faster than gpu clouds.
upper90 previously pioneered gpu-backed loans in 2021 with crusoe, when traditional lenders avoided such deals due to depreciation risks. as coreweave turned chip-backed loans into a business model and a major ipo, the practice became common. now, with gpus well understood and possibly overbought, upper90 is betting on inference chips. ceo billy libby sees open source models as key, noting that not everyone needs a supercomputer but they do need inference. this aligns with growing interest in open model providers like openrouter and fireworks, and new competitive models like kimi's k3.
the deal highlights a fragmenting market beyond nvidia's dominance. general compute's access to non-nvidia chips mirrors moves by others like tensorwave with amd. as more chip alternatives scale, providers not tied to nvidia may offer cheaper inference. ceo finn puklowski called the financing a signal of capital organizing against nvidia's monopolistic hold, not just a startup buying compute. this shift could reshape ai infrastructure economics, favoring cost-efficient inference for the broader market.
why it matters: this signals a practical shift toward cheaper, specialized hardware for running ai models, which could lower costs and reduce reliance on nvidia for ai inference workloads.
source: techcrunch ai: why the first gpu financiers are turning to inference chips in a $400 million deal