Mercor Books $614M in H1 2026, Leaning on OpenAI, Anthropic

The interesting number in this week's Mercor reporting isn't the top line, it's the mix. The Information reported, citing internal financial documents, that the AI training-data startup booked around $614 million in gross revenue in the first half of 2026, with the bulk of that spend coming from a handful of frontier AI labs.

Mercor's public story lately has been velocity. Co-founder and CEO Brendan Foody said the company crossed a $2 billion annualized gross revenue run rate in June 2026, roughly four months after hitting $1 billion. Publicly disclosed customers include OpenAI, Anthropic, Meta, and Google, and Mercor claims to work with all of the top five AI labs and six of the Magnificent Seven tech companies. Zoom out on the business model and the exposure is starker: Mercor is essentially a marketplace paying more than 30,000 domain experts to write answers, grade model responses, and score outputs for post-training, with contractors keeping roughly 60 to 70% of the 'gross revenue' number that gets quoted.

Why this matters if you are not a data-labeling investor: this is the second-order economy of the model race. Every dollar OpenAI and Anthropic raise flows partly into judging their own models better, and Mercor's numbers are the clearest public readout of just how big that spend has become. It also sharpens the debate over whether the next leg of frontier progress is being carried by model-side scaling or by the humans grading the outputs.

The honest caveat is that this is reporting on internal documents rather than audited filings, and neither the exact customer-by-customer split nor the net margin under that gross number is public. The historical warning sign is familiar though. Appen was worth $4.3 billion in 2020 when five clients delivered around 80% of its revenue, and today it trades at a small fraction of that. Mercor is riding an up-cycle that looks nothing like Appen's, but the shape of the customer list rhymes.

The upside case for operators watching this is that Mercor's growth is a signal to build the next Mercor for domains the biggest labs haven't optimised for yet, from safety red-teaming to specialised medical and legal evaluation. Whoever gets there without needing OpenAI or Anthropic to be the majority of the P&L is probably the more interesting business to back.