Being good at a public leaderboard does not mean an AI agent can be trusted inside a specific company. That gap is now a business.
Mercor.io Corp., which supplies the human expertise used to teach and grade AI models, has banked $750M in a Series D that lifts its valuation to $20B. The deal includes a tender offer for employee shares; the company has not broken out how much is fresh capital.
From frontier labs to the enterprise floor
Until now the buyers of Mercor’s work have mostly been the big model labs. The company draws on more than 5 million domain experts and says its annualized revenue cleared $2B in June.
The new money points elsewhere. Co-chief executive Adarsh Hiremath argues a model can ace a benchmark and still stumble on the systems, policies and edge cases that decide how a business actually runs. Mercor’s answer is a bespoke benchmark for each client, then post-training against that company’s own standards.
Some of that runs on software from Deeptune, the New York startup Mercor bought in July, whose team spent two years recreating hundreds of enterprise applications for agents to practice on.
Nvidia and Salesforce write cheques
Mercor named no lead investor, but the round pulled corporate money from Nvidia, Salesforce and Zoom alongside returning backers Benchmark, Felicis and General Catalyst. Sapphire Ventures, StepStone, G Squared and funds tied to Apollo also joined.
The prior raise landed in October 2025: a $350M Series C led by Felicis at a $10B valuation. Doubling that mark in about a year puts Mercor among the faster-rising names in the data layer behind AI.
Co-chief executive Brendan Foody has said every job in the economy will eventually get its own agent. If that holds, someone has to certify the work.
