Mercor Series D: $750 Million at a $20 Billion Valuation for the AI Data Firm
Mercor raised $750 million in a Series D at a $20 billion valuation on November 10, betting that networks of human experts are the bottleneck for training better AI. The AI data gold rush keeps minting decacorns.

Mercor raised $750 million in a Series D on November 10, at a $20 billion valuation, and no lead investor was named. The San Francisco company connects AI developers with a network of more than 5 million human experts who train and evaluate models, and it is now expanding from serving frontier labs into custom evaluations and post-training for enterprises deploying AI agents on their own workflows.
The Mercor Series D is the biggest startup funding story of the week, and it says something blunt about where the money is going. Investors are no longer just betting on the companies that build AI models. They are betting on the companies that supply the human judgment those models are trained on.
Why human experts made the Mercor Series D worth $20 billion
Every frontier model is trained on data, and the easy data is gone. The public internet has been scraped, and the remaining gains come from high-quality, expert-produced work: doctors grading medical reasoning, lawyers checking legal arguments, engineers writing and reviewing code. That work cannot be scraped. It has to be commissioned, vetted, and paid for.
Mercor's pitch is that it built the marketplace for that work before anyone else took it seriously. Five million experts is a big network, and the company's move into enterprise post-training is the logical next step. Frontier labs were the first customers because they had the most money and the most urgent need. Enterprises deploying AI agents are the bigger market, because there are thousands of them and each one needs its models tuned to its own workflows.
The valuation tells you how seriously investors take this. Twenty billion dollars for a company that sells human expertise to AI labs is a bet that the data bottleneck is real, durable, and worth more than the models themselves. It is also a bet that the expert network is a moat. Anyone can spin up a labeling operation. Few can claim five million vetted experts.
The funding week in context
Mercor was not the only big raise on November 10. Path Robotics raised $300 million in a Series E at a $2.45 billion valuation to put its welding robots into more factories. Charta Health raised $55 million in a Series B to review medical charts with AI before claims are submitted. Index Industries came out of stealth with a $50 million seed to turn US housing into an investable asset.
The pattern across the week is consistent with what the last few funding rounds have shown. Fal raised $1.3 billion at a $16 billion valuation because running AI models is now infrastructure. Zipline raised $1.5 billion at $21.5 billion to make drone delivery nationwide. Nominal raised $240 million to make sense of hardware test data. The money is flowing to companies that solve the unglamorous problems around AI: inference, delivery, testing, and now, the humans who teach the models.
There is a risk in this story that nobody at the funding announcement will mention. If AI models get good enough to evaluate themselves, the market for human experts shrinks. Mercor's answer is that the frontier keeps moving: as models improve, the evaluation bar rises with them, and the experts who can judge the new frontier are always scarce. That is a reasonable argument, and it is also exactly the kind of argument that gets tested in a downturn.
For now, though, the market has spoken with its wallet. Three quarters of a billion dollars, a $20 billion valuation, and a business built on the idea that the smartest thing in AI is still a person. The gold rush has reached the people selling the picks, and the picks are people.


