SF Compute CEO discusses a market for programmable GPU clusters
Original title: The Programmable Data Center — Evan Conrad, SF Compute
Evan Conrad says SF Compute now operates much of a cloud so traded GPU contracts settle as usable compute.
A Latent Space interview with SF Compute CEO Evan Conrad, recorded at AI Engineer New York, covers how the company moved from subleasing a cluster to operating infrastructure that delivers traded GPU hours. Conrad describes matching orders that differ by GPU generation, networking, software, and whether workloads can be interrupted. He argues a market operator should stay independent of large hardware or inference businesses, and cites on-the-fly InfiniBand fabric subdivision as one programmable-data-center feature.
- The interview contrasts physical delivery of compute with cash settlement against an index.
- Conrad says NVIDIA would not be independent enough to run the exchange.
- Topics include spot and forward pricing and batch versus live inference leases.
- Show notes mention a $245 million compute sale and Abel, SF Compute’s settlement ledger.
- Organizations
- SF ComputeNVIDIA
From the original article
15 min read at latent.spaceFrom GPU trading to physical settlement, independent markets, and programmable compute
latent.space
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.