The human in the loop.
By Mara Quinn | November 11, 2026
The most valuable input to frontier AI right now is not a chip. It is a cardiologist reviewing a diagnosis, a securities lawyer flagging an error, a software engineer catching a subtle bug. Mercor has built the market for that labor, and on Tuesday investors valued it at $20 billion.
The San Francisco company announced a $750 million Series D on November 10, just over a year after closing a $350 million Series C in October 2025 that valued it at $10 billion. Doubling a ten-figure valuation in twelve months takes either extraordinary growth or extraordinary hype. Mercor has delivered the former. According to a Sourcery interview with CEO Brendan Foody, the company now pays out more than $5 million per day to the experts in its network and has grown 5X on a mid-nine-figure revenue base since September 2025.
From Hiring Tool to AI Training Infrastructure
Mercor began in 2023 as an AI-powered hiring platform, built by co-founders Brendan Foody, Adarsh Hiremath, and Surya Midha. The original thesis was using AI to match workers with jobs. The pivot was recognizing that the most urgent unmet demand was for evaluated AI: models that could be tested, corrected, and refined against expert human judgment.
The company has since built a network of more than five million paid experts. They are not writing training data in the abstract. They are doing actual professional work: reviewing outputs, flagging errors, ranking responses, and in doing so producing the signal that makes frontier models better at professional tasks. The labor is paid. The judgment is specific. The result is a dataset no synthetic process can replicate.
The Deeptune Acquisition Changes the Enterprise Pitch
Earlier in 2026, Mercor acquired Deeptune. The acquisition was about what Deeptune spent two years building: hundreds of simulated enterprise applications where AI agents can practice professional workflows before deployment. Technical capabilities also mattered.
That infrastructure changes what Mercor can sell. For enterprises evaluating whether an AI agent can handle their accounts-payable process, their contract review queue, or their IT help desk, Mercor now offers a testing environment that mirrors actual work conditions. Agents fail or succeed against tasks that reflect real enterprise software, not toy benchmarks. According to Mercor's Series D announcement, the company is now selling agent evaluation directly to enterprise clients, backed by this Deeptune infrastructure.
The pitch is well-timed. AI agents are moving from demos to production, and the gap between demo performance and production reliability is exactly where enterprise IT teams lose confidence. Mercor sells a way to close that gap before go-live.
800 Occupations, Most of Them Still Unmeasured
The growth numbers are large. The opportunity Mercor is describing is larger.
The United States has more than 800 distinct occupations containing workflows against which models have yet to be evaluated, according to Mercor's announcement. A model may handle general code review competently, but no one has yet tested it against the documentation standards of a nuclear plant operator or the billing codes a hospital coder uses daily. Mercor argues that closing this evaluation gap is itself a substantial business, one that requires real experts, paid to do real work, at scale.
That argument is now backed by a roster of investors spanning venture capital, growth equity, and strategic money. The Series D includes Benchmark, Felicis, Fundomo, General Catalyst, Glade Brook Capital, Sapphire Ventures, StepStone Group, and XTX Ventures, plus funds managed by affiliates of Apollo and participation from G Squared. NVIDIA, Salesforce, and Zoom joined as strategic backers. Jack Dorsey and Peter Thiel are also investors, according to the company's announcement.
New Offices, Sustained Investment
Mercor opened offices in New York and London this year. The company did not detail team sizes at either location, but the expansion points to a push to build proximity to enterprise buyers and the professional networks that feed its expert marketplace.
The Series D proceeds will accelerate investment in evaluation and post-training infrastructure and research, per the company's announcement. Both areas require sustained capital rather than a single deployment.
What the Valuation Signals
The doubling of Mercor's valuation in twelve months says something about more than one company. It reflects where the bottleneck in AI development currently sits. Compute is expensive but scalable. Expert human judgment is expensive and scarce. The companies building systematic access to that judgment, across the full breadth of professional workflows, are capturing real value.
Mercor's bet is that five million experts is a start, not a ceiling. Most of the work of evaluating AI against human professional life still lies ahead.
