Mercor raises $750M at a $20 billion valuation to sell enterprises the AI-training playbook it built for frontier labs

Mercor, the San Francisco company that pays a network of more than five million experts to train and evaluate AI models, said on Nov. 10, 2026, that it raised a $750 million Series D at a $20 billion valuation to expand its evaluation and post-training business so companies in any industry can test and train AI agents on their own work.

Mercor got rich selling human expertise to the labs racing to build frontier models; now it's betting every other company will want its own evals and its own post-trained model. Twenty billion dollars says investors believe the "AI needs humans to grade it" business has legs. The release skips two things worth remembering: how much of the $750M is employees cashing out, and the breach that cost it Meta's work earlier this year.

Mercor, a San Francisco company founded in 2023, said it raised a $750 million Series D at a $20 billion valuation. A Series D is a later private funding round, after a company already has a business. The valuation is the price this round puts on the whole company. Mercor pays a network of more than five million experts to train and evaluate AI models. An expert, here, is a person paid for professional knowledge. The announcement is dated Nov. 10, 2026. Those lines are Mercor's.

Investors named in the round are Benchmark, Felicis, Fundomo, funds managed by affiliates of Apollo, G Squared, General Catalyst, Glade Brook Capital, NVIDIA, Salesforce, Sapphire Ventures, StepStone Group, XTX Ventures, and Zoom. The release does not name one firm as the lead. Those names are Mercor's.

The round included an employee tender, co-founder and co-CEO Brendan Foody wrote on the company blog. A tender, here, is a chance for employees to sell some of their shares as part of the round. Mercor did not say how much of the $750 million went to that tender. The Next Web is the one that flags the missing split. The $20 billion price is double the $10 billion valuation from Mercor's $350 million round in October 2025. That comparison is The Next Web's.

The money is meant to speed up evaluation, which the company calls evals, and the post-training business. An eval is a test that scores how a model does on a job. Post-training is the extra practice after a general model already exists. Mercor says it will build custom benchmarks from each company's real workflows, then post-train models on that company's work, often starting from an open model a customer can build on. It also wants more realistic environments where agents can practice. An agent is software that can take a chain of steps, not only answer one question. A benchmark is a shared test. Those lines are Mercor's.

Deeptune, which Mercor acquired earlier this year, spent two years recreating hundreds of enterprise applications as practice environments for agents. The new work also builds on APEX, Mercor's public benchmark series for measuring how AI does economically valuable work. Those lines are in the Nov. 10 release.

The main buyers so far have been frontier AI labs, the companies training the largest models. Mercor says the same kind of work is now being bought by big tech, developers of open models, application-layer companies, and the Fortune 500. It opened offices in New York and London this year. Those lines are Mercor's.

Foody, co-founder and co-CEO, is quoted: "Every job in the economy will need its own AI agents, and every company will need a way to train and evaluate them on its own work." Adarsh Hiremath, the other co-founder and co-CEO, is quoted: "A model can score well on a public benchmark and still fail on the systems, policies, and edge cases that define how a specific business runs." Jai Das, president, partner, and co-founder at Sapphire Ventures, is quoted in the release: "For the last few years, frontier labs have been the primary buyers of expert-driven training data, which is about to change. As open models become good enough to build on, every enterprise will want to evaluate and train agents on its own work." Those quotations are theirs.

On his blog, Foody wrote that Mercor already spends three times as much on inference as on employee salaries, and that the spend pays for itself by making the team better at their jobs. Inference is the cost of running a model to get an answer. He argues that if that ratio holds as AI spreads through the rest of the economy, it points to something closer to double-digit GDP growth over the next five years, nearly three times the rate he says people are used to. GDP is the size of an economy's output in a year. That growth line is his argument. It is not an outside forecast, and it is not a result Mercor measured across the economy.

The Next Web added context the company left out. In April 2026, a breach at Mercor led Meta to freeze its AI data work with the company. Mercor published an incident update in June. Neither the Nov. 10 release nor Foody's blog post mentions it. Those lines are The Next Web's.

The picture is a headshot of Brendan Foody against a gray background. The credit is Mercor, from the Series D post.

In plain terms, Mercor raised $750 million at a $20 billion valuation, double where The Next Web puts it in October 2025, to sell other companies the testing and training service it built for frontier labs. Employees can sell some shares in the round. The company did not say how much of the money that is. The release also skips the breach that cost it Meta's work earlier this year.