Mercor sells the one input frontier AI labs cannot download: expert human judgment. It was founded in 2023 by Brendan Foody, Adarsh Hiremath, and Surya Midha, then college students at Georgetown and Harvard who knew each other from the Bay Area debate circuit. All three later took Thiel Fellowships. At 22, they became the youngest self-made billionaires on record.
What it does. The last generation of AI training data was cheap and generic — contractors labeling images at low hourly rates. That work is largely finished. Models have read the public internet, and the bottleneck has moved to teaching them skilled professional work. Judging whether a model wrote a competent investment memo requires someone who has written one; a $30-an-hour generalist cannot do it. So Mercor recruits doctors, lawyers, bankers, and senior engineers, screens them through a 20-minute AI video interview, and matches them to the labs that need them. Its network runs to roughly 300,000 vetted professionals. It pays out over $2 million a day to them, at an average above $85 an hour, and keeps a percentage of each placement.
Why this might be interesting. The case rests on three things.
The first is a supply shock the company did not create but moved into faster than anyone. In June 2025 Meta paid $14.3 billion for a 49% stake in Scale AI and hired its founder to run a superintelligence lab. Within days Google — Scale’s largest customer — cut ties, and Microsoft, OpenAI, and xAI followed. No lab wants to hand unreleased training data to a vendor half-owned by a competitor. Scale laid off 14% of staff a month later. One word explained every defection: neutrality. Roughly a billion dollars a year of frontier-lab spending went looking for a new home, and Mercor was one of a handful of independents standing there when it did.
The second is that this is not a staffing business being priced like infrastructure by mistake. Labs are spending on the order of a billion dollars each annually on human data. Mercor has moved beyond selling hours: it built APEX, a public benchmark measuring how well AI models perform real professional work across law, banking, consulting, and medicine, and it sells environments for training AI agents. It acquired Deeptune in July 2026 to go further in that direction. Owning the yardstick the industry is measured by, and the environments it trains in, is a quieter kind of leverage than headcount.
The third is simply the curve. Mercor says it crossed $2 billion in gross annualized revenue in June 2026, doubling four months after passing $1 billion. It has reported being profitable on a cash-flow basis — rare at this growth rate. Whether that curve reflects durable position or the peak of a spending cycle is the whole question.
Funding. A $100 million Series B at $2 billion in February 2025, then a $350 million Series C at $10 billion in October 2025, led by Felicis with Benchmark, General Catalyst, and Robinhood Ventures. Roughly $490 million raised in total. In July 2026, Bloomberg and TechCrunch reported the company is in early discussions to raise at roughly $20 billion, having told investors it holds at least one term sheet at that price. Forbes puts the round at $500 million and expects it to close this month. Terms could change. The round may not happen.
The current opportunity. We have an SPV in Mercor. Fee terms are negotiable based on size. If you would like to look at it, reply and we can share what we have.
Worth keeping in mind. Start with the revenue figure, because it is the most misread number in the sector. The $2 billion is gross — the total the labs pay, most of which passes straight through to the contractors. Mercor’s take rate is somewhere between 20% and 35% depending on whose estimate you use. Net revenue is a fraction of the headline, and a buyer who prices this at “10x revenue” is not looking at the same business the sellers are. There is a live industry argument about exactly this, and Foody himself has been one of its loudest voices: in June he publicly accused Sequoia of structuring rounds so that inflated headline valuations reach employees and angels. He is now reportedly raising at a doubled headline number against a gross revenue figure. Draw your own conclusion, but ask the question.
Then the competition, which is thicker than the narrative suggests. Surge AI bootstrapped past $1 billion in revenue with roughly 110 employees and no outside capital, and has reportedly been seeking $15 billion or more. Turing, Micro1, Handshake, and Uber have all moved into the same work. Scale sued Mercor for trade-secret misappropriation in September 2025. And every one of Mercor’s customers is a company with the resources to build this in-house — the pressure to do so rises with the invoice.
Finally, the year Mercor has had. It disclosed a data breach in March 2026 exposing customer and contractor data, after which Meta paused its work indefinitely. Contract workers have filed lawsuits. Forbes has separately reported an employee fired for embezzlement and suspected North Korean operatives on the platform using stolen credentials. A doubling of the valuation in nine months, at a company that just lost a marquee customer over security, is the tension a buyer has to price.