Mercor Raises $750M at $20B Valuation to Train AI on Real Jobs

Mercor closed a $750 million Series D on November 10, 2026, at a $20 billion valuation, money it intends to spend teaching AI agents to perform real occupational work rather than chat about it. The round included an employee tender and backing from NVIDIA, Salesforce and Zoom, alongside Benchmark, Felicis, General Catalyst, StepStone Group, Apollo-affiliated funds, Fundomo, G Squared, Glade Brook Capital, Sapphire Ventures and XTX Ventures.

For employers, the announcement touches a concrete people-process question: whether AI agents can be measured, trained and certified against a specific company's own workflows before deployment. Mercor's answer is a service built on custom evaluations, expert-generated training data and simulated work environments. The claim is bold; the evidence, as usual in this market, rests largely on the company's own reporting.

What does the funding actually buy?

Mercor says the capital will accelerate infrastructure and research for model evaluations and post-training — additional training applied to an existing model — and will fund more environments where agents can learn complex workflows. The company also opened New York and London offices in 2026, moving beyond San Francisco to sit closer to enterprise and laboratory customers and to recruit staff.

The vendor pitch deserves scrutiny on adoption terms. Mercor reports a network of more than five million experts paid to train frontier AI models. That figure is self-reported, without independent verification, and the company does not disclose how many are active at any given time.

One statistic stands out for its scale: Mercor claims the United States contains more than 800 distinct occupations with workflows against which AI models have yet to be evaluated. If accurate, that number frames a vast unfinished measurement project — and, conveniently for Mercor, a vast addressable market.

How does enterprise evaluation work?

Mercor has spent the past two years building infrastructure to train and evaluate frontier models on real-world work using human expertise. It now sells that capability to enterprises through evaluations designed to establish three things: what agents can do, what they cost and how their performance can improve.

The method distinguishes between two levels of knowledge — knowing how a task is generally performed, and knowing how a particular employer handles it under its own systems, policies and quality standards. Mercor's custom evaluations use an employer's actual workflows to measure models against company requirements and to identify where additional training or practice in realistic environments could help.

The company's enterprise playbook proceeds in steps:

• Start with a capable open model
• Build public evaluations
• Use post-training to reduce costs
• Then train models for the individual customer

Mercor describes post-training as requiring only a fraction of the resources needed at the frontier. Its frontier-lab and enterprise work reportedly share the same expert network, evaluation methods and post-training infrastructure, so lessons from one side inform the other. The efficiency claim is plausible; the company offers no cost figures to substantiate it.

What role does Deeptune play?

Earlier in 2026, Mercor acquired Deeptune, a reinforcement-learning environment company. The July 9, 2026 announcement outlined the deal and Deeptune's two years of work recreating hundreds of enterprise applications — from spreadsheets to Salesforce — as agent-training environments where agents can practice professional workflows.

Each environment contains three components:

• The software where the work occurs
• Tasks defining what the agent must accomplish
• Verifiers determining whether the agent succeeded

Mercor's experts create the tasks and verifiers, translating human judgment into learning signals. The July announcement says Mercor's APEX system measures model performance on economically valuable tasks across thousands of real-world workflows, while Deeptune supplies the software layer that turns expertise into training environments. The acquisition price was not disclosed.

What does Mercor's own spending reveal?

The company's internal numbers offer an unusual data point for HR planners. Mercor reports spending three times as much on inference — the computing used to run models — as on employee salaries. It says the spending pays for itself by making employees better at their jobs.

Co-founder and chief executive Brendan Foody said that if the ratio holds as AI spreads through the economy, it points to something closer to double-digit growth in gross domestic product over the next five years, nearly three times the rate he described as customary.

That projection deserves caution. It extrapolates one company's spending pattern to an entire economy, and Mercor is hardly a typical employer. A startup whose product is AI training will naturally spend more on inference than a manufacturer or a bank would.

What should HR leaders take from this?

The funding supports a single enterprise proposition: build the environment, supply the experts and measure whether an agent is ready. For Mercor, the practical test is reliable performance inside a customer's own workflows, where general task knowledge alone is insufficient.

For people-process decisions, the relevant questions are capability questions, not product questions. Can an outside vendor credibly evaluate an AI agent against your internal standards? Who inside the organization owns that evaluation? And what happens to the expert workforce — Mercor's five million contractors included — whose judgment becomes the training signal? The company's answers rest on self-reported numbers; buyers will want to test them against their own workflows before committing.

If Mercor's model proves out, the next five years could see evaluation of AI against company-specific work become as routine as background checks — and the 800 unevaluated occupations will not stay unevaluated for long.