gpu-clusters
(4)
Evan Conrad Is Teaching Supercomputers How to Have an Exit
A year-long GPU lease nearly emptied a two-person startup's bank account. Evan Conrad turned the mistake into SF Compute - an order book where AI companies can buy serious computing power and sell back the hours they no longer need.
TensorPool
TensorPool (YC W25) is a San Francisco startup that turns cloud GPUs into a git-style command line. Instead of wrestling with SSH sessions, provider dashboards, and idle-instance billing, machine learning engineers describe a training job in a config file and run 'tp job push' - TensorPool spins up a multi-node cluster across partner clouds, streams logs, saves outputs, and shuts everything down, billing only for runtime. It advertises GPU access at roughly half the cost of major cloud providers, with H100s starting around $1.99/hr. Founded by three Stanford dormmates, the company was accepted into Y Combinator's Winter 2025 batch.
Spectro Cloud
Spectro Cloud is a San Jose-based enterprise software company that builds Palette, a Kubernetes management platform used to run, secure and scale containerized workloads, AI infrastructure and GPU clusters across cloud, data center, bare metal and edge environments. Founded in 2019 by Cisco/Cliqr veterans, it counts GE HealthCare, T-Mobile, Nokia and the U.S. Air Force and Navy among its customers and closed a $75M Series C in November 2024 led by Growth Equity at Goldman Sachs.
Mansour Karam
Mansour Karam is the founder and CEO of Aria Networks, the AI-native networking company that raised $125M to rethink how GPU clusters talk to each other. A Stanford-trained electrical engineer with a Lebanese-American background, he helped build Arista Networks from a handful of employees to a $4B IPO, then founded Apstra - the company that coined 'intent-based networking' - and sold it to Juniper in 2021. A piano player who names companies after musical terms, Karam has spent 25 years at the bleeding edge of how data moves, and now bets that the network - not the GPU - is the real bottleneck in the AI era.