SF Compute enables programmable GPU compute markets with physical settlement
SF Compute, led by CEO Evan Conrad, is building a programmable marketplace for GPU compute that goes beyond simple spot pricing. The company treats GPU capacity like a commodity, using physical settlement – actually delivering the hardware resources rather than a cash‑only contract – and matching orders with detailed constraints such as interruption tolerance, storage backup, and specific GPU…
Key points
- SF Compute has placed about 5% of its GPU capacity on a programmable order book, achieving high liquidity for that slice.
- The firm uses physical settlement, delivering actual GPU compute and matching constraints such as interruption tolerance, storage backup, and GPU generation.
- Customers can lease compute from an hour up to three years; spot B300 rates are $5‑$6/hr, while resale inference can earn $25‑$40/hr.
Only about 5% of SF Compute’s total GPU capacity is currently on its order book, but that slice enjoys high liquidity. Customers can buy compute for as short as an hour or lock in capacity for up to three years. Spot prices for B300 GPUs run about $5‑$6 per hour, while reselling the same capacity for inference can generate $25‑$40 per hour, giving the firm an average $25 profit per $100 of capacity sold back.
The platform also handles forward contracts, batch versus live inference leases, and price curves that reflect hardware release cycles. Conrad notes that larger hardware owners may lack the neutrality needed to run a fair exchange, positioning SF Compute as an independent market operator for AI workloads.
The Programmable Data Center — Evan Conrad, SF Compute
Latent Space · 20 October 2026
Loading the full article…
This text was published by Latent Space and written by Latent.Space. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Enterprise & Industry
All →- Reservoir opens agricultural robotics incubator for Washington growers · 1 src
- US restarts three nuclear reactors to supply power for AI data centers · 2 src
- Smartsheet launches MCP Server for AI agents to manage work data on AWS · 1 src
- Octave Reliance builds secure multi-tenant AI foundation on Amazon Bedrock · 1 src
- BriefCatch expands into unified legal writing suite with verification tools · 1 src
Comments
via GitHub Discussions