News
DealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory鈥檚 NVIDIA & AMD Enterprise ExpansionDealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory鈥檚 NVIDIA & AMD Enterprise Expansion
EveryNeocloud
SF Compute

SF Compute

GPU cloud with a resale market for reserved capacity

SF Compute is a neocloud provider based in San Francisco that builds and operates GPU supercomputers. It allows customers to sublease unused reserved capacity through an internal marketplace, reducing financial risk from long-term contracts.

Share on WhatsApp

  • Allows subleasing of reserved GPU capacity
  • Offers elastic InfiniBand partitions
  • Formerly an AI lab that pivoted to infrastructure
  • Customers include Standard Intelligence and Roboflow
  • Command-line tool for buying and reselling compute

What it is

SF Compute operates as a neocloud that builds data centres and GPU clusters, then leases them to customers. Unlike traditional providers that lock users into rigid long-term contracts, SF Compute allows clients to sublease their reserved capacity if they no longer need it. This model aims to prevent the financial risks of overbuying or underutilising expensive hardware. The company offers virtual machines, bare-metal access, and managed Slurm scheduling. It functions both as a technical operator of supercomputers and as a marketplace for compute resources.

How it came about

The organisation began as an AI laboratory that purchased a large GPU cluster. Facing the risk of failure due to high costs, the team was forced to sublease the hardware to survive. This experience led to a pivot from research to infrastructure provision. The founders created a compute market with physical settlement to address the industry-wide problem of stranded assets. The team includes former leaders from Lambda, Crusoe, Digital Ocean, AWS, and Hut8. They claim to have previously deployed hundreds of thousands of GPUs and eight gigawatts of data centre capacity.

The compute

The platform provides elastic InfiniBand partitions, allowing users to create isolated networks for training jobs programmatically. Customers can connect workers within a job while keeping separate tasks isolated. The system supports scaling instances within the same partition as workloads grow. Users interact with the service via a command-line tool called sf, which avoids complex dashboards. This interface enables scripting and automation for buying or reselling capacity. The company manages the underlying hardware, including GPU clusters and associated CPU fleets, ensuring technical control over the infrastructure.

Where the GPUs are

SF Compute is based in San Francisco. The company builds its own data centres and supercomputers rather than relying solely on third-party hosting. It serves as a technical partner for financially motivated parties who wish to place GPU clusters on their balance sheets. SF Compute operates these assets on behalf of the owners, retaining full technical control. This hybrid approach allows for rapid scaling similar to a market while maintaining the operational standards of a cloud provider. Specific data centre locations beyond the headquarters are not detailed in the available information.

Customers and deals

The provider serves ambitious teams seeking large-scale compute without catastrophic balance sheet risk. Recent users accessing larger GPU clusters through the resale market include Standard Intelligence and Roboflow. The company also acts as a technical partner for entities looking for economic returns from GPU ownership. These partners build clusters on their own balance sheets, while SF Compute handles construction, operation, and leasing. This arrangement allows cluster owners to achieve levered returns by utilising the subleasing feature.

Money and ownership

SF Compute operates on a model where customers reserve compute for terms such as three months. If capacity is unused, it can be resold, turning idle GPUs into revenue rather than sunk costs. The company claims its hybrid model outperforms brokered contracts by eliminating the need for networks of counterparties. By offering subleasing, it aims to increase returns for cluster owners. The business structure allows it to scale quickly by serving financially motivated parties who provide the capital for hardware while SF Compute provides the technical expertise.

Its place in AI

The company positions itself as a solution to the desperate gamble often required when buying from neocloud vendors. It invented the first compute market with physical settlement to derisk customer investments. By allowing subleasing, it addresses the fear of being forced to bet the farm on fixed infrastructure. The team describes itself as having high talent density, with experience from major tech and infrastructure firms. It aims to shape the AI industry by providing safe, flexible access to supercomputing resources for those who come after.

Questions people ask about SF Compute

How does SF Compute differ from other GPU clouds?

SF Compute allows customers to sublease their reserved capacity if they do not use it. This prevents the financial risk of being locked into long-term contracts for unused hardware, turning idle GPUs into revenue instead of sunk costs.

What type of networking does the platform support?

The platform offers elastic InfiniBand partitions. Users can programmatically create isolated networks for specific training jobs, connect workers, and add instances to the same partition as their workload grows.

Who are some of the customers using SF Compute?

Recent customers accessing larger GPU clusters through the resale market include Standard Intelligence and Roboflow. The company also partners with financially motivated entities that own the hardware but rely on SF Compute for operations.

How do users interact with the SF Compute platform?

Users install a command-line tool called sf. This allows them to log in, buy, or resell GPU capacity directly from the terminal. The design avoids complex dashboards and supports scripting and automation.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on SF Compute yet, so the facts on this page come from the sources listed above: sfcompute.com/about, sfcompute.com/. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryNeocloud wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the provider of the day, new pages on EveryNeocloud, and the conferences and anniversaries of the week ahead.