General Compute secured a $400 million [1] chip-backed loan from Upper90 Capital Management to expand its AI inference cloud operations.
This transaction signals a pivotal shift in how the AI industry finances hardware. For the first time, a major debt facility is collateralized by specialized inference chips rather than the industry-standard Nvidia GPUs, suggesting a change in how investors perceive the value of AI hardware.
The loan is backed by SambaNova SN50 ASIC inference chips [2]. These chips are designed specifically for the inference stage of AI—the process of running a trained model to generate outputs—rather than the training stage where GPUs typically dominate.
General Compute, based in Silicon Valley, is utilizing the funds to scale its infrastructure. The move comes as inference workloads begin to represent a larger share of total AI compute demand [1]. By moving away from GPUs, the company is betting on the efficiency, and scalability of application-specific integrated circuits (ASICs).
Upper90 Capital Management, which has a history of financing GPU-heavy ventures, shifted its strategy for this deal. The firm said the SambaNova SN50 ASICs are higher-value and lower-volatility assets compared to the fluctuating market for Nvidia GPUs [1], [2].
The loan closed in the weeks preceding the announcement on Friday, July 17 [1]. This landmark deal establishes a new precedent for AI infrastructure financing, treating specialized inference hardware as a bankable asset class in the U.S. market.
“General Compute secured a $400 million chip-backed loan from Upper90 Capital Management.”
The transition from GPU-backed to ASIC-backed loans indicates that the AI market is moving from a 'training phase' to an 'inference phase.' As companies shift from building models to deploying them at scale, the financial industry is recognizing that specialized hardware for running AI may be a more stable and valuable asset than the general-purpose GPUs used to create them.
