Deal Sheet
- Company:
- Mercor, which pays experts to train and test AI models
- Co-founders:
- Brendan Foody, Adarsh Hiremath, and Surya Midha
- Funding:
- $750M Series D at a $20B valuation (November 2026), up from $10B (October 2025)
- Investors:
- Benchmark, Felicis, General Catalyst, Nvidia, Salesforce, Zoom, funds managed by affiliates of Apollo, and others
- Growth:
- Run-rate revenue from $1M to $3B in 27 months (company figure); 5M+ experts in its network
- The Shift:
- From training data for a few AI labs to tests and retraining for AI agents at any company
Mercor, a startup that pays doctors, lawyers, and bankers to train AI, raised $750 million at a $20 billion valuation on November 10. A year earlier it was worth $10 billion, and in 2023 Y Combinator turned it down. Until now, its biggest buyers were the few labs that build frontier AI models. Now it wants to sell to every company that puts AI to work.
Mercor's valuation at each priced round. Graphic: EO.
For Two Years, Mercor's Biggest Buyers Were a Handful of AI Labs
Mercor's work looks a lot like tutoring. Its network counts more than 5 million professionals, who write hard questions for AI models, grade the answers, and write better ones. Frontier AI labs pay for that work and use it to make their models smarter.
The biggest buyers are few, so losing one hurts. In June 2025, Meta paid $14.3 billion for 49% of Scale AI, then the best-known company in the field. Within a week, OpenAI said it was phasing out its work with Scale, and Google reportedly planned to cut ties too.
Mercor's own investors describe the same short list. "For the last few years, frontier labs have been the primary buyers of expert-driven training data, which is about to change," Jai Das, President and Partner at Sapphire Ventures, said in the funding announcement.
Mercor Wants to Grade AI for Every Company That Uses It
Brendan FoodyCo-Founder and Co-CEO, Mercor
The money goes to evals and post-training, meaning tests that measure how well an AI model does a job, and extra training to fix what the tests find.
Mercor co-founders Adarsh Hiremath, Brendan Foody, and Surya Midha (left to right).Photo: ULY Inc. for Mercor.
Businesses are starting to hand AI agents, software that carries out tasks on its own, real work like drafting a deal memo or reviewing a contract. The labs build general models. They don't know any one company's systems, policies, or exceptions. Mercor's plan is to bring in experts who know the job, have them write a test from the company's own work, score the agent, and train it on what it got wrong. For two years Mercor was a tutor for a few labs. Now it wants to be the exam board for AI at work.
“Every company deploying AI needs to know whether it works, and the answer is different for each one.”Adarsh HiremathCo-Founder and Co-CEO, Mercor
Mercor has already shown what such an exam looks like. In January it published APEX-Agents, a set of tasks that investment banking analysts, management consultants, and corporate lawyers wrote from their own jobs. AI agents work through them with real files and tools, and Mercor scores each new model on a public leaderboard.
Mercor's APEX-Agents leaderboard, as of November 2026.Source: Mercor.
Earlier this year Mercor bought Deeptune, whose team spent two years recreating hundreds of enterprise applications for agents to practice in. Two of the investors in this round, Salesforce and Zoom, sell software to businesses, not to AI labs.
The $20B Assumes Customers Mercor Hasn't Named
Mercor now describes its customers as "frontier AI labs, big tech, AI startups, and Fortune 500 companies." It hasn't named any of those Fortune 500 companies.
Mercor's own chart counts about 120 customers, and enterprises now outnumber the labs. But it counts customers, not dollars. About 10 of them are foundation-model labs, the companies that build the largest AI models, and Mercor hasn't said how much of its revenue comes from outside them.
Mercor's customer count by type, 2023 to 2026, as shared by its CEO.Source: Brendan Foody on X.
That share is the number to watch. If it stays small, Mercor still depends on a few labs, the position that made Meta's deal so costly for Scale AI in 2025. If banks, law firms, and consulting firms become a large share of what Mercor earns, Mercor becomes the grader for AI at work.
