Mercor raises $750 million for AI evals at $20 billion valuation

Mercor said on Tuesday it had raised $750 million in a Series D round. The round values the company at $20 billion. Mercor sells access to named specialists. Those specialists grade and write the data AI labs use to train their models for real jobs.

Who put the money in

Mercor's announcement lists Benchmark, Felicis and Fundomo as investors in the round. Funds tied to Apollo, G Squared, General Catalyst and Glade Brook Capital also took part. So did NVIDIA, Salesforce, Sapphire Ventures, StepStone Group, XTX Ventures and Zoom. Existing employees also sold shares in a tender.

The round is Mercor's fourth. Its Series C in October 2025 valued the company at $10 billion. The valuation has doubled in about a year.

The co-founder and chief executive, Brendan Foody, wrote the announcement. He said the money would go into "evals and post-training". It will also go into "building more environments where agents can learn complex workflows". Earlier in 2026, Mercor bought Deeptune. That team had spent two years rebuilding enterprise apps as safe places for an agent to practise.

What Mercor actually sells

The product is a network of more than five million people. Mercor says they are real workers in real jobs. A lab or a company training an AI pays Mercor to find the right specialists. Those people grade the model's answers. They write the hard test cases. Or they run the agent against a copy of their own working tool. The point is to make the model good at a job no public dataset covers.

Foody said "AI is only as useful as the human expertise behind it". He said the demand for specialist knowledge "compounds" as models get better. He said there are "more than 800 distinct occupations" in the United States alone. Each one is a place where a model has not yet been measured.

The part worth reading twice

The announcement makes one striking claim about Mercor's own work. Foody wrote that the company "spends 3x as much on inference as on employee salaries". He said this "pays for itself" by making the team better at their jobs. Mercor gives no dollar figure behind this. It is the company's own statement.

Mercor has opened offices in New York and London alongside its San Francisco headquarters. It did not name a customer in this announcement. It did not state a revenue figure. A trade newsletter, Latent Space, said on the same day that Mercor's annualised revenue is now around $3 billion. Latent Space cited a private briefing. Mercor did not confirm the figure.

For a team building with AI, the signal is simple. Benchmark and Felicis are paying a doubled valuation. They are buying one bet. The bet is that human specialist work makes an agent good at a real job.

Source

• Mercor
• Latent Space