SF Compute helps companies afford larger GPU clusters by letting them resell GPU hours to other teams. Customers can reserve enough GPUs for their biggest workloads, keep the hours they need, and sell unused hours to help reduce the reservation cost.
Those resold hours become short-term reservations for other teams, giving them access to large clusters for individual jobs without a long-term commitment.
Standard Intelligence builds frontier models that learn from video to carry out computer tasks. They resell GPU hours outside of peak usage periods, which allows them to purchase larger, interconnected clusters to serve their most demanding workloads.
“We procure compute calibrated to our peak load rather than the average. Because our cluster on SF Compute is supported by a resale market with high liquidity, the cost to hold excess capacity is just the small delta from price fluctuations, rather than the entire value of the underlying contract. This means we get larger scaleups while having lower baseline spend than traditional clouds.”
Galen MeadCEOResale from customers like Standard Intelligence makes more GPU capacity available for teams that need shorter commitments.
Roboflow, a computer vision startup, buys short-term GPU reservations on SF Compute. Their teams can run larger jobs and reserve just the time each job needs.
“SF Compute lets us take on bigger workloads with shorter commitments. Through the market, we can access larger clusters for the time we need them and match our reservations to the jobs we want to run.”
Isaac RobinsonResearch LeadIn a traditional anchor tenant model, one customer makes a long-term commitment to a cluster at a fixed price. The provider earns the agreed contract revenue, regardless of how much of that capacity the customer uses.
With SF Compute, that same commitment can generate an additional source of revenue. When the anchor tenant resells GPU hours to other teams, the provider earns a share of the resale fees on top of the original contract revenue.
SF Compute lets providers offer long-term anchor contracts, short-term burst contracts, or both. Shorter reservations can command higher hourly rates while costing customers less overall. Both contract types let customers resell GPU hours, with providers earning a share of resale fees on top of contract revenue.
Need compute for your next project? Talk to our team.
