Mercor raises $750M at $20B valuation to take AI agent evaluation to enterprises

Mercor.io Corp. said it has closed $750 million in new funding that values the AI training data company at $20 billion, as it moves to sell enterprises a way to measure whether AI agents can be trusted with real work.

Mercor supplies the human expertise used to train and evaluate AI models, and its network includes more than 5 million domain experts. Most of that work has gone to frontier AI labs until now. The new money is meant to extend the business into enterprises across industries, which Mercor contends need their own means of judging whether an AI agent can be relied on for a given job.

Co-Chief Executive Brendan Foody said every job in the economy will require its own AI agents, and businesses will need a way to train and evaluate those agents on their own work. He said Mercor spent the past two years building that infrastructure with the labs, a business that has grown quickly. In a July post on X, Foody said annualized revenue had passed $2 billion in June.

Co-Chief Executive Adarsh Hiremath said a model can score well on a public benchmark and still fail on the systems, policies and edge cases that define how a particular business operates. For each enterprise customer, Mercor builds a custom benchmark that tests models against that company's own standards, with post-training starting from there. Part of that post-training runs on software from Deeptune Inc., which Mercor bought in July. Deeptune's team had spent the two years before the acquisition recreating hundreds of enterprise applications for AI agents to practice on, and Mercor said the new money will pay for research and evaluation infrastructure, including more of those environments.

The previous round came in October 2025, when Felicis led a $350 million Series C that valued Mercor at $10 billion. According to Foody's blog post on the new round, it included a tender offer for employee shares, and the company has not said how the $750 million splits between that and fresh capital.

Sapphire Ventures president, partner and co-founder Jai Das said frontier labs have been the main buyers of expert-driven training data so far. He expects that to change as open models become good enough for enterprises to build on, and said Mercor has built something difficult to replicate after several years of work with the labs.

More company and startup news from TechManNews.