SF Compute provides marketplace-style access to large-scale NVIDIA GPU compute, offering hourly and time-block purchases of vetted GPU clusters, a mechanism to resell unused time, and operational surfaces including a web dashboard, CLI, and API.
Published Sep 26, 2026
Mainstream B2B review platforms (G2, TrustRadius, Gartner Peer Insights) did not surface clearly accessible, product-specific review pages for SF Compute during this research pass; independent third-party review coverage is limited.
SF Compute describes a post-training case study applying reinforcement learning techniques to a reasoning model for chess and links to a whitepaper.
Critical evaluation:
Get a comprehensive analysis of SF Compute including market position, competitive landscape, adoption trends, and peer benchmarks.
Download full report →From SF Compute public documentation (not exhaustive):
The SF Compute site includes vendor-curated testimonials that reference pricing, availability, and support, for example:
Caution: these are curated and vendor-published testimonials and should be treated as vendor-sourced signals rather than independent verification.
Vendor-published LSI materials include a claim that LSI is significantly cheaper than other options for many open-source models and references a workstream with a research lab that generated substantial synthetic data and realized material cost savings compared to a competitor.
Critical evaluation:
A third-party partner announced a commercial agreement with SF Compute to provide an initial allocation of H100-class capacity with plans to scale later.
Critical evaluation:
Best fit tends to be teams that can use large GPU capacity efficiently and have engineering maturity:
SF Compute (also branded as “The San Francisco Compute Company”) provides access to large-scale GPU compute (notably NVIDIA H100/H200 today, with newer hardware referenced in messaging) via a marketplace-style model. The platform offers flexible access to large clusters without long-term commitments, including the ability to purchase time-blocks of compute and a mechanism to resell unused time.
Operationally, SF Compute operates vetted GPU clusters and supports workloads with a virtualization stack, documentation/CLI, and support; the company presents its economic model as a market rather than a fixed price list. In self-serve flows, customers primarily receive VMs; bare metal, Kubernetes, and Slurm are mentioned as available via contacting the company, and some documentation indicates Kubernetes as the default for certain purchase flows.
Key product surfaces:
Positioning: a GPU compute marketplace and flexible cluster provider for training and large-scale inference workloads. The offering is positioned against:
Target buyers (inferred from site and messaging):