AI Training Data Startups Reach $8.5 Billion in Revenue, Valued at Approximately $100 Billion
In July 2026, over 50 startups sold training data and RL environments to large AI labs, driving behind-the-scenes progress in AI.
These companies generated total revenues of approximately $8.5 billion, with a valuation of around $100 billion, over 75% of which is concentrated in four companies: Scale, Surge, Mercor, and Handshake.
In market mechanisms, large AI labs have become the main buyers, benefiting data suppliers, with funds flowing from general computing power to high-quality training data and RL environment supply, expanding the data labeling and synthetic data ecosystem.
Source: Public Information
ABAB AI Insight
Companies like Scale AI initially entered the market through data labeling services, followed quickly by Surge, Mercor, and Handshake, forming an oligopoly dominated by a few leading firms, continuing the capital path of data as a core production factor for AI.
On the capital path, these companies provide high-quality data and RL environments to labs like OpenAI and Anthropic, shifting resources from human labeling to synthetic data and automated platforms, motivated by capturing the bottleneck profits of AI training data, leading to rapid valuation growth.
Similar to the early concentration of cloud computing infrastructure providers, the current AI data supply is transitioning from fragmentation to concentration among leading firms.
Essentially, this represents capital concentration: training data has become a scarce input for AI advancement, with a few companies capturing most of the value through scale and technological advantages, reshaping the pricing power and entry barriers of the data layer in the AI industry chain.
ABAB News · Law of Cognition
Data is not oil; it is the oxygen of the AI era, and those who sell oxygen will get rich first.
75% of the valuation is held by four companies; the oligopoly era has arrived.
The stronger the model, the more valuable the backend data supply chain.