AI training data is becoming a billion-dollar business as labs buy workflow data and RL environments
创业邦 · wechat · 2026-07-25
This long WeChat article argues that AI training data has become one of the most underestimated businesses in the current AI wave.
Why the market is exploding
- Deedy’s map of 28 training-data companies reportedly shows about $8.5B in annual revenue and nearly $100B in total valuation.
- The piece says the market is growing fast because data is no longer a one-off input; it is now a recurring fuel for faster model iteration.
- As models improve, labs need more targeted, high-value data more frequently, especially for coding and agentic workflows.
From labeling to deeper services
- Early text labeling paid around $0.02 per item; expert labeling now commands much higher hourly rates.
- Companies like Scale AI are described as moving from pure labor outsourcing toward evaluation, task design, and engineering services.
- The article says this shift lifts margins and gives data companies more pricing power.
Two new trends
- Type1 data: real-world workflow data is becoming more valuable than hand-designed tasks.
- RL environments: interactive training environments are replacing static datasets as a key purchasing target.
Four major players diverge
- Scale AI is moving downstream into enterprise AI applications.
- Surge stays boutique and focuses on high-end RLHF and safety work.
- Mercor is expanding an expert network and bought Deeptune to enter RL environments.
- Handshake is positioning itself as a data pipeline between workers, companies, and AI labs.
The article’s conclusion: simple data labeling is being compressed into the thinnest part of the value chain, while the real money moves to workflow capture, expert production, and environment infrastructure.
More from Companies & People
- X drama: Anthropic researchers accused of spying on academic customers and racing them to results — basedjensen · 2026-09-11
- Investor argues Palantir-Nvidia partnership should slash Anthropic's IPO valuation — pdamodaran · 2026-09-11
- PyTorch Day Korea 2026 launches first offline conf, CFP closes Sept 13 — PyTorch · 2026-09-11
- Class action accuses Anthropic of overselling Claude subscriptions with deceptive usage multipliers — The Decoder · 2026-09-11
- AI post-training and evals jobs pay up to $850K, with median offers at $210K–$325K — FinanceYF5 · 2026-09-11
- 89% of firms use AI, only 6% see significant ROI — the busywork illusion — mikeflache · 2026-09-11