RL environment firms earn just ~5% of the $200B AI labs burn — could it be a $2T market?
joecole · x · 2026-10-03
- Jonah B argues demand for RL environments and training data persists as long as there are things humans know that AI doesn't. OpenAI and Anthropic alone burn a combined $140B/year; add Google, Meta and xAI and it's roughly $200B. Yet per a July Menlo Ventures market map (deedydas), companies selling training data and RL environments bring in only $8.5B combined — about 5% of that spend.
- He extrapolates: if demand for intelligence is uncapped, by analogy to Walmart + Amazon's combined $1.4T revenue, these firms could plausibly reach $2T/year.
- The quote-tweet raises a structural concern: expert-generated-data businesses may churn through their own supply — once domain experts train a model, they're no longer needed, forcing constant recruitment of new expert pools, unlike marketplaces like Uber where drivers stay reusable.
More from Venture
- Bending Spoons: 50+ Acquisitions and the $1.38B Vimeo Deal Behind the AI-Driven Roll-Up Machine — pdamodaran · 2026-10-03
- 60% of the Latest YC Batch Are Using Supabase, Says Garry Tan's Retweet — garrytan · 2026-10-03
- SMB spends 2 days and $250 building an AI quoting tool that saves 20 hrs/week — MatthewChang · 2026-10-03
- Supertake adds crypto to its personal financial agent, supporting 19 major coins — MurrLincoln · 2026-10-03
- MSFT, NVDA, AAPL and META accounted for 200%+ of S&P 500's Q3 gains, per Citadel — menhguin · 2026-10-03
- The FSD math: $1,200/year subscription avoids $6,000 of crash harm, author argues — jamesdouma · 2026-10-03