Selling data to AI labs is like selling to the government: few buyers, brutal RFPs
mobav0 · x · 2026-10-05
calvinchen (Proximal) breaks down the data-for-labs business into two models:
- Lab-directed work: labs tell you what they need. It's commoditized; large players that can react at scale usually win the deals.
- Self-directed research: data has many open-ended research problems, so a strong research+engineering team can find angles even labs haven't explored. Much harder — requires talent at or above lab level — but far more valuable long term. Proximal is built for this path.
A quoted thread adds the ecosystem's texture: only a handful of buyers, so it resembles selling to the government — hustle to reach a researcher, pass vendor approval, set up a shared Slack channel, then face endless RFPs and ruthless competition against 20 vendors per project. Deliver well once and the business keeps coming, until the lab's research direction shifts or a better vendor replaces you.
More from Companies & People
- Prime Intellect's RL Training Job Listings Come With a Three-Book Reading List — Hesamation · 2026-10-05
- Schmidhuber joins Sakana AI as chief scientific advisor, free Tokyo symposium Oct 26 — hardmaru · 2026-10-05
- Investor Predicts Cursor Will Beat All Rivals Handily Within 12 Months — JOBhakdi · 2026-10-05
- AI Swarm Dynamics Hackathon returns Oct 3-4 with $3,000 prizes to build swarm-detection tools — lfschiavo · 2026-10-05
- Jason Slams Anthropic for Teaching Claude It's Sentient: "The Backstory of Blade Runner" — markjeffrey · 2026-10-05
- Sam Altman says world must 'accept some bad things happening' for AI's benefits — Polymarket · 2026-10-05