Best NeocloudsAn independent guide to GPU clouds

Compare GPU clouds / SF Compute

SF Compute

Order-book market for GPU cluster time, with resale

Marketplace

SF Compute sells blocks of H100 and H200 cluster time on an order book: buyers set a node count, a time window and a maximum price per node-hour, and can resell reserved time they do not use. Clusters run on InfiniBand with managed Slurm. eWEEK reports that it manages several thousand GPUs and takes about 10% per transaction.

Best for: Short, flexible blocks of multi-node H100 or H200 time.

No third-party rating

  • – Publishes prices
  • – Serverless inference
  • – AMD GPUs
  • – European data centers
Visit SF Compute

Facts checked November 9, 2026 against the sources listed below. SF Compute does not pay to appear here.

1 Legal nameSan Francisco Compute Company
2 HeadquartersSan Francisco, CA, USA
3 FoundedNot disclosed
4 Funding or listingPrivate. $40M Series A led by DCVC and Wing Venture Capital (Nov 2025).
5 How you rentMarketplace, On-demand, Reserved clusters, Bare metal
6 Minimum commitmentNone stated; time blocks bought on an order book (docs show 1-hour orders) on 8-GPU nodes
7 InterconnectInfiniBand, with programmatic Elastic InfiniBand Partitions (public preview)
8 OrchestrationManaged Slurm, sf CLI, REST API
9 GPUs offeredH100 SXM, H200
10 RegionsNot disclosed
11 CertificationsNot disclosed
12 Customers and partnersClaims Standard Intelligence and Roboflow reach larger clusters through its resale market (Sep 2026).

Published prices

SF Compute does not publish on-demand prices. Ask for a quote.

Worth knowing before you sign

  • Prices are visible only after sign-in, and the order-book API needs a token.
  • Not rated in SemiAnalysis ClusterMAX 3.0.
  • No security certifications are stated.

Go to SF Compute's site

Sources

  1. SF Compute
  2. SF Compute order book (docs)
  3. SF Compute instance SKUs (docs)
  4. eWEEK: $40M Series A

Work at SF Compute? Send corrections with a source to editors@bestneoclouds.com. We fix verified errors within five business days.

Compare with

  • Prime Intellect GPU compute marketplace with its own clusters and post-training tools
  • Vast.ai Open GPU marketplace with live host prices
  • Shadeform GPU aggregator across about 18 clouds at the clouds’ own prices
  • Hyperbolic GPU marketplace of supplier capacity, plus multi-cloud GPU software (Forge)
  • Mithril Bid-based GPU spot market, formerly Foundry