Mercor Raises 750 Million Dollars at a 20 Billion Valuation to Sell AI Evals to Enterprises

Mercor, the San Francisco company that pays domain experts to produce and grade training data for artificial intelligence models, has spent three years selling its network to the laboratories building frontier systems. On November 10, 2026, it moved to sell the same machinery to everyone else. The company announced a $750 million Series D at a $20 billion valuation, a number that places it among the most valuable private AI companies and resets expectations for the data, evaluation and post-training layer of the stack.

The pitch is narrower than the valuation suggests and, in some ways, more ambitious. Mercor does not build models. It builds the measurement and refinement layer around them: custom benchmarks written with a customer's own experts, and post-training work that tunes a model to a specific company's systems, policies and quality bar. The company argues that public benchmarks, however saturated, say little about whether an agent will survive contact with a real workflow inside a real organization.

That argument has become easier to sell. Over the past two years the largest labs have spent heavily on human expertise, hiring doctors, lawyers, accountants and engineers through networks like Mercor's to write questions, grade answers and catch failures that automated scoring misses. Dealroom reported on August 20, 2026 that Mercor expected gross revenue to total $2 billion for the full year, and that the company had raised roughly $520 million before this round, reaching a $10 billion valuation in October 2025.

The new round therefore doubles the October 2025 mark in about thirteen months, a pace that has become familiar in AI infrastructure but rare for a company whose product is, at bottom, organized human labor. Mercor says the money will accelerate evaluation and post-training work for enterprises, expand its bench of experts and support offices it opened this year in New York and London alongside its San Francisco headquarters.

Key Facts

Business Wire reported on November 10, 2026 that Mercor raised $750 million at a $20 billion valuation, describing the company as one helping organizations train and evaluate AI using human expertise. The release listed fourteen investors: Benchmark, Felicis, Fundomo, funds managed by affiliates of Apollo, G Squared, General Catalyst, Glade Brook Capital, NVIDIA, Salesforce, Sapphire Ventures, StepStone Group, XTX Ventures and Zoom. It did not name a single lead investor.

Co-founder and chief executive Brendan Foody wrote in a company blog post dated November 10, 2026 that the round included an employee tender alongside the investor participation, a detail absent from the press release. The distinction matters: a tender lets existing staff sell shares back to the company or to incoming buyers, which means not all $750 million is fresh capital for the business. Mercor did not disclose the split.

The company's numbers are large for its age. Founded in 2023, Mercor says its network now exceeds five million experts and serves frontier AI labs, big tech, AI startups and Fortune 500 companies. It employs more than 400 people. Foody wrote that the company pays out more than $4 million to experts per day, up from the $2 million per day it disclosed in 2025, and that Mercor already spends three times as much on inference as on employee salaries.

CNBC reported on November 10, 2026 that Foody appeared on Squawk Box that morning to discuss the round and the valuation, in a segment running 6 minutes and 29 seconds. The network's description framed the conversation around the enterprise business rather than the frontier lab data work that built Mercor's earlier growth, and it placed the financing in the main business news lineup rather than only in trade press.

Mercor also disclosed that it acquired Deeptune earlier this year. That team spent two years recreating hundreds of enterprise applications as environments where agents can practice multi-step workflows, and the company says the acquisition underpins its post-training offering. The approach extends APEX, Mercor's public benchmark series, into private versions written for individual customers.

Analysis

The doubling from $10 billion to $20 billion in roughly a year is the headline, but the more interesting detail is the mix of investors. Nvidia and Salesforce sit on the cap table alongside crossover and growth funds, and Nvidia's presence is not passive. Dealroom reported on August 20, 2026 that Nvidia had discussed backing Mercor at a $20 billion valuation, with General Catalyst in talks to lead, and that Nvidia had prioritized its Nemotron family of open models. Prior reporting has described Nvidia as a customer that pays Mercor for expert data used in those models. An investor that is also a customer is a durable signal in a market where data contracts are often short.

What this really means is that evaluations, not raw data volume, are becoming the product. A benchmark that a customer co-writes with its own experts is sticky in a way that a pile of labeled examples is not: it encodes institutional knowledge, it has to be refreshed as workflows change, and it produces the failure cases that post-training then fixes. Foody's argument, that a model can score well on a public benchmark and still fail on the systems, policies and edge cases that define how a specific business runs, is the commercial thesis in one sentence. Adarsh Hiremath, the other co-chief executive, has made the same point about custom benchmarks being the starting point for post-training rather than an afterthought.

The bigger picture here is that the AI data business is rotating from a handful of frontier labs toward the much larger population of enterprises that have budget, proprietary process and no reliable way to tell whether their agents work. Sapphire Ventures president and co-founder Jai Das said frontier labs had been the primary buyers of expert-driven training data, which is about to change. If that rotation happens, Mercor's five million experts become less a staffing pool for labs and more a distributed measurement instrument for the Fortune 500.

The money is real, but so is the risk buried in the comparison. Dealroom reported on August 20, 2026 that Mercor paid roughly 60% to 70% of gross revenue to contractors as of last year, which means the headline $2 billion gross figure translates into a much smaller net number, with net revenue for the first half of 2026 estimated at approximately $180 million to $250 million. The company swung to a small operating profit the previous year and told investors it expected operating profit of $226 million in 2026 and $1.7 billion in 2027. Those projections, not the valuation, are the thing to watch.

Why It Matters

The funding lands in a week when Nvidia was already dominating technology coverage for less flattering reasons. CNBC's own description of the Foody interview paired the round with questions about a scandal surrounding Nvidia chief executive Jensen Huang. Mercor's ability to raise at double its last mark while its most prominent investor was under a news cycle of its own says something about how badly the market wants exposure to the evaluation layer.

It matters for labor as much as for software. Mercor's economics depend on converting professional expertise into structured tasks at a payout rate that still leaves a margin, and on inference costs that Foody says already exceed salaries by three times. If that ratio holds broadly, the constraint on this business is not demand but the price of compute and the supply of genuinely expert reviewers, both of which are tightening.

It matters for buyers too. Enterprises that adopt custom evals are effectively writing down what good means for their domain, an artifact that outlives any single model vendor and gives them negotiating leverage when they switch. That is the sense in which Mercor's language about enterprises owning their intelligence is more than marketing: the benchmark, not the weights, is the asset that transfers.

Next Up

Expect the enterprise push to be measured in named customers rather than valuation marks. Mercor has not disclosed how much of the round is primary capital, who led it, or what revenue the enterprise segment contributes today, and those gaps will be filled by the company's own disclosures or by reporting over the next two quarters. The Deeptune environments and the APEX benchmark line are the products to watch for evidence that custom evals convert into recurring post-training contracts.

The competitive question is whether frontier labs keep buying expert data at the current rate if their own models improve at grading, and whether Nvidia's stake buys preference in Nemotron work. Dealroom reported on August 20, 2026 that Nvidia invested $18.6 billion in private companies and infrastructure funds in the quarter ending April 2026, more than it invested throughout 2025, and noted that the chipmaker still relies heavily on synthetic data despite rising spending on human generated datasets. For a company selling human expertise, that gap is the entire opportunity.