AI Training Startup Mercor Raises $750 Million at $20 Billion Valuation

Mercor, a platform that pays human experts to create training data and evaluation tests for AI models, announced a $750 million Series D round on November 10 that values the company at $20 billion. The round doubles the roughly $10 billion valuation Reuters reported after Mercor's $350 million Series C in October 2025 [2].

The round and the investors

Mercor's press release names 13 investors: 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. No lead investor is identified. The company says the round "included an employee tender," meaning part of the $750 million bought shares from staff rather than funding operations. Mercor does not disclose the split [2].

Revenue claims: run-rate, not revenue

The press release and Mercor's blog post contain no revenue figure. The $3 billion number comes from co-founder Brendan Foody's X post at 15:52 UTC on announcement day, where he wrote that Mercor "scaled from $1M to $3B in run-rate revenue in 27 months" and called it the fastest any company has reached that mark—a claim with no independent verification [2].

Earlier, in a May 8 engineering blog post, Mercor said it had "crossed $1B in annualized revenue run rate earlier this year" and went from nothing to a $500 million run-rate in 17 months. That same post said the company paid "over $2 million every day to more than 30,000 weekly active contractors" [2].

Molly O'Shea, who interviewed Foody for the announcement, summarized "$5M+ in daily expert payouts" in her notes—an interviewer's figure, not a company filing [2].

Three cautions on the $3 billion run-rate:

• Run-rate is not revenue. It annualizes a recent period; neither source says which period.
• Nothing is audited. Mercor is private; these figures appear in no regulatory filing.
• Most goes back out. At $5 million daily, expert payouts would total roughly $1.8 billion annually—about 60% of a $3 billion run-rate, by simple arithmetic on two figures from different sources [2].

The $20 billion valuation implies roughly 6.7 times the claimed run-rate.

What Mercor actually sells

Mercor describes "more than five million experts in our network getting paid to train frontier AI models." That is a cumulative network size, not weekly active workers. In May, Mercor's own figure for weekly active contractors was a little over 30,000 [2].

The company began as an AI-powered hiring platform, backed by Jack Dorsey and Peter Thiel per Reuters. Today its experts write tasks, grading rubrics, and reference answers that show whether a model can do a real job. Mercor publishes a public benchmark series called APEX. Earlier this year it acquired Deeptune, whose team "spent two years recreating hundreds of enterprise applications" as practice environments for agents [2].

The pitch: every company tests its own agents

Co-CEOs Foody and Adarsh Hiremath split the argument in the release. Foody focuses on demand: "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." Hiremath argues public benchmarks are insufficient: "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" [2].

Sapphire Ventures' Jai Das supplies the market view: frontier labs "have been the primary buyers of expert-driven training data, which is about to change" [2].

Foody's blog post adds a fact about Mercor's own costs: "We already spend 3x as much on inference as on employee salaries." He then projects that ratio across the economy to forecast double-digit GDP growth—the first part is a single-company datapoint, the second a extrapolation [2].

What to watch

• Whether Mercor publishes audited revenue, not only run-rate. O'Shea's interview notes list an IPO timeline among topics discussed.
• The customer mix: the pitch depends on enterprises, not only frontier labs. Foody's chart shows foundation-model customers flat since 2025, with enterprises driving the rise to nearly 120 total customers.
• Competition. O'Shea's summary names Scale AI as a direct rival [2].