Mercor, a San Francisco startup that pays experts to train and test AI models, has raised $750M in a Series D round at a $20B valuation.
That is twice the $10B valuation Mercor reached when it raised $350M in October 2025. Named investors include Benchmark, Felicis, Fundomo, General Catalyst, Nvidia, Salesforce, Sapphire Ventures and Zoom.
The business is easy to describe and harder to replace. A model can absorb vast amounts of text on its own, but judging whether an answer is correct, rather than merely fluent, in a specialized field takes someone who knows the subject. Mercor supplies those people, and AI developers use their work to train models and to test them afterward. It competes for that role with data-labeling companies such as Scale AI, which also sells human-supplied training data to AI developers.
The investor list is worth a second look. Nvidia, Salesforce and Zoom all sell products that depend on AI demand continuing to grow. Backing a supplier of training data is a bet on the supply chain behind the models, not only on one startup.
The open question is how long that supply chain needs this much human input. If models get better at generating and checking their own training material, the demand for paid expert review could shrink. A $20B price assumes it does not.