Data Marketplaces & Intermediaries

Mercor turned a hiring platform into the market that trains AI — and that is the bridge from interviews to training data

Source: ValueAdd VC · Oct 1, 2026

Mercor is the clearest example yet of a shift we have tracked from many angles: the company started life as an AI-hiring platform — matching candidates to roles — and has turned into something larger and more valuable, an expert-data marketplace that sells human reasoning to the labs building frontier models. The framing its own backers now use is blunt: it is 'the labor market that trains' AI. Instead of placing people into jobs, it routes vetted experts into the work of producing the data models learn from.

The mechanics are what make it relevant to anyone sitting on a corpus of real human performance. Mercor runs a network of more than thirty thousand screened domain experts — doctors, lawyers, engineers, PhDs — who generate the reasoning traces, grading rubrics, and evaluations that post-training now depends on. Its customers are not resellers; they are the frontier labs themselves, OpenAI and Google DeepMind among them. By mid-2026 the platform was moving roughly two billion dollars a year in gross volume, having doubled in four months, and was reported to be in talks for a twenty-billion-dollar valuation. It has also started acquiring its way toward reinforcement-learning environments — simulated settings where AI agents practice real tasks — so that the data it sells is not just annotation but lived problem-solving.

The reason this is the most important story on our radar is the convergence it makes visible. Mercor sits on top of the exact seam our own thesis runs along: the business of evaluating how skilled people perform and the business of selling how skilled people perform have become the same business. A platform born to screen and interview talent is now the channel through which that talent's judgment reaches a model's weights. For an owner of a structured, consented, multimodal record of real technical interviews — people actually reasoning through hard problems on video, in audio, in live code — Mercor's trajectory is both a validation and a map. The scarce input is authentic human performance, the buyers are the labs, and the market clearing that data is being built in public, at a twenty-billion-dollar valuation, right now.

Key Points

• Mercor began as an AI-hiring platform and has repositioned into an expert-data marketplace — the labor market that supplies frontier labs with the human reasoning they train on
• It routes work to 30,000+ vetted domain experts — doctors, lawyers, engineers, PhDs — who produce reasoning traces, rubrics, and evaluations for model post-training
• Gross volume reached roughly $2B annualized by mid-2026 (doubling in four months), with the company reported to be in talks for a $20B valuation
• Its customers are the frontier labs themselves — OpenAI and Google DeepMind among them — the same buyers that need scarce, high-quality human data
• It has acquired its way toward reinforcement-learning environments, buying a company that builds simulated settings where AI agents practice real work
• The single most important read for us: the same business that evaluates how people perform is now the business that sells how people perform — interview/assessment and training data are converging into one market