Tech Brew Ride Home (Portfolio Profile) SF Compute
Nov 22, 2023
Alex Gajewski, Co-founder of San Francisco Compute, shares his journey from music production to revolutionizing AI with short-term GPU rentals. He discusses the challenges of launching their computing cluster and how containerization became essential for scaling AI startups. Gajewski also delves into community-driven innovations in AI and the emerging marketplace for model weights, emphasizing the need for accessible resources in developing cutting-edge technology. Tune in for valuable insights on the AI landscape and future opportunities!
AI Snips
Chapters
Transcript
Episode notes
First Cluster Launch Challenges
- SF Compute's first cluster launch was challenging due to unforeseen hypervisor issues with the new H100 GPUs.
- The team rewrote their entire infrastructure in 37 hours, switching from VMs to containers, and successfully launched.
Flexible Compute for AI Startups
- Traditional cloud providers often require long-term contracts and large upfront payments, creating a barrier for AI startups.
- SF Compute offers flexible, short-term bursts of compute, enabling startups to test and scale models more efficiently.
Inspiration for SF Compute
- Alex Gajewski's previous startup, Metaphor Systems, a generative search engine, faced challenges scaling its model due to compute limitations.
- This experience, along with difficulties finding short-term compute for a music model, led to the idea for SF Compute.