Problem

AI developers, startups, graduate students, and research labs need substantial GPU capacity for model pre-training and other large-scale workloads, but access to reliable, vetted clusters is difficult and expensive. The deeper inefficiency is that contracted supercomputing capacity can sit unused while other buyers need it; without a liquid market, compute cannot be readily bought, sold, or reassigned. San Francisco Compute’s central use case is making affordable pre-training capacity available to organizations that cannot efficiently build or reserve large clusters themselves.

Product / Service

San Francisco Compute is a vertically integrated AI-compute provider: it builds data centers and GPU clusters, runs large-scale vetted infrastructure, and sells supercomputers through contracts that customers can sublease. Its central limit order book lets buyers trade compute by the hour and lets contract holders resell unused capacity, creating a real-time spot market for AI workloads rather than a purely fixed reservation model.

The model combines owned or managed infrastructure with marketplace liquidity and supplier partnerships. The intended benefit is better utilization, more flexible access, and lower costs; a related API built with Modular’s inference stack and SF Compute’s spot market was described as delivering up to 80% lower cost than typical alternatives.

Market

San Francisco Compute competes in the AI infrastructure and GPU-cloud market, with a differentiated emphasis on a real-time compute marketplace for buying, selling, and subleasing capacity. Identified competitors include Verda, INFINITIX, and CoreWeave, while the broader competitive set includes providers of on-demand and reserved GPU capacity. The company was founded in 2023.

The available evidence indicates substantial early traction rather than a clearly documented pre-revenue company: SF Compute raised $40 million in Series A equity financing at a reported $300 million valuation to expand its AI compute marketplace, and it has announced supplier and technology relationships including Soluna and Modular. The research does not provide revenue figures or confirm profitability, so its commercial scale beyond financing, partnerships, and marketplace expansion cannot be quantified from the available evidence.