AI / Funding

🔥 Mercor Raised $750 Million to Teach AI Every Job on Earth

Mercor closed a $750M Series D at a $20B valuation on 10 Nov 2026, doubling in a year by selling AI labs expert human data.

Mercor Raised $750 Million to Teach AI Every Job on Earth — Tech Arcade
Photo: Valentin Lacoste / Unsplash

The hottest company in the AI boom doesn’t train a model, build a chip, or ship a robot. It books contractors. And investors just decided that business is worth twenty billion dollars.

Let’s get the number on the table first: $750 million. That’s the size of the Series D that Mercor closed on 10 November, at a $20 billion valuation (Reuters, via AI Weekly). The round doubled the $10 billion price tag Mercor carried after its October 2025 Series C, which itself was five times the $2 billion it was worth in February of that year (TechCrunch).

Nvidia and Salesforce wrote cheques. So did Benchmark, General Catalyst, and funds tied to Apollo (TechStartups). Three years after it was founded, a company that started life as an AI recruiting tool is now the picks-and-shovels supplier to the labs building frontier models.

Here is the thesis: Mercor sells the one thing you still cannot synthesise, which is human expertise, and the AI labs are paying almost anything to get it.

🧠 Why This Matters

Everyone obsesses over the models. Almost nobody watches the data that feeds them, and that is exactly where the money has quietly moved.

Frontier labs have scraped most of the public internet already. What they need now is specialised, expert-graded data: a radiologist marking up scans, a litigator annotating case law, a quant checking a model’s maths. That work cannot be crawled. It has to be done by actual professionals, one task at a time.

Mercor is the marketplace that routes that work. It claims more than five million professionals across medicine, law, finance, engineering, and science on its books (TechStartups). When OpenAI, Anthropic, or Nvidia needs a thousand experts to grade model outputs by Friday, Mercor is who they call. Nvidia used Mercor-sourced data to train its two most recent Nemotron models (AI Weekly).

That is why a labour-matching business is being priced like a chip designer. The compute layer has Nvidia. The data layer, increasingly, has Mercor.

📊 Deep Dive

The growth curve is the whole story, so look at it directly:

  • Valuation, Feb 2025: $2 billion (TechCrunch)
  • Valuation, Oct 2025: $10 billion, on a $350 million Series C (TechStartups)
  • Valuation, Nov 2026: $20 billion, on a $750 million Series D (AI Weekly)
  • Gross revenue, H1 2026: $614 million (AI Weekly)
  • Annualised gross run rate, June 2026: past $2 billion, doubled in four months (Dealroom)

Now the part that makes seasoned investors raise an eyebrow: the people running this are barely old enough to rent a car. Co-founders Brendan Foody, Adarsh Hiremath, and Surya Midha were college dropouts and former high-school debate teammates. At the October 2025 round, aged 22, they were reported as the world’s youngest self-made billionaires (TheNextWeb).

This latest funding accelerates our investment in the infrastructure and research for evals and post-training. — Mercor, company blog

The product has also crept upmarket. Mercor started matching contractors to jobs. It now builds evaluation and post-training infrastructure, the scaffolding labs use to measure and fine-tune models, which is a far stickier place to sit than a staffing agency.

⚠️ The Catch

A $20 billion valuation on roughly $2 billion of gross billings deserves a hard look, because gross is not what Mercor keeps.

Contractors take an estimated 60 to 70 percent of billings, per reporting from The Information. That puts Mercor’s net revenue somewhere around $600 to $800 million, which means you are paying something like 25 to 33 times net revenue for the equity (TheNextWeb). Rich, even by 2026 standards.

Then there is concentration. A handful of AI labs drive the bulk of the business, and AI Weekly reports several Mercor staff now work almost entirely on the Nvidia account (AI Weekly). Lose one whale and the run rate wobbles.

And the company has already taken real damage. In March 2026, a supply-chain attack on the open-source library LiteLLM exposed up to four terabytes of Mercor data. Meta, a major customer, paused all work indefinitely, class-action suits followed, and the hacking group Lapsus$ claimed the breach (TheNextWeb). Mercor called the impact “very limited.” OpenAI and Anthropic stayed.

There is also a governance flag. When Mercor acquired AI-agent-testing startup Deeptune in 2026, it emerged that Foody had personally been an angel investor in Deeptune months earlier. He told Fortune the angel stake was “in a lot of ways the main motivation” for the deal (TheNextWeb).

🎯 What Happens Next

Watch three things.

Margins. If Mercor can push more clients onto its evals and post-training products, it keeps more of each dollar and the valuation starts to make sense. If it stays a pass-through for contractor pay, the multiple looks stretched.

Customer spread. New logos outside the top labs would ease the concentration risk. Expect Mercor to chase enterprise and government buyers who also want expert-graded data.

The automation irony. Mercor’s own pitch is that AI will reshape every job. The question nobody has answered is what happens to a human-expertise marketplace when the models get good enough to grade themselves.

🧩 Bigger Picture

Mercor’s rise lands exactly as the old model of cheap, anonymous online labour fades. Amazon has let Mechanical Turk, the crowd-work platform that defined the 2010s, drift into the background (Runtime). The work that replaced it is not cheaper and anonymous. It is expensive and credentialled.

That tells you where the value sits in this phase of AI. Raw data is commoditised. Expert judgement is the scarce input, and whoever aggregates the experts collects the toll. For now, that is three dropouts in San Francisco and a five-million-strong roster of doctors, lawyers, and engineers teaching machines to do the jobs they trained for.

The models may be artificial. The intelligence they are learning from is still, for the moment, very much human, and it just got priced at $20 billion.


Sources

❓ Quick answers

How much did Mercor raise and at what valuation?

Mercor raised a $750 million Series D at a $20 billion valuation, announced 10 November 2026 (per Reuters via AI Weekly), double its $10 billion October 2025 mark.

What does Mercor actually do?

It runs a marketplace of more than five million vetted professionals (doctors, lawyers, engineers) who grade and generate expert data to train and evaluate AI models for clients like OpenAI, Anthropic and Nvidia.

Why is a $20 billion valuation controversial?

The $2 billion run rate is gross billings; contractors keep an estimated 60-70%, so net revenue is roughly $600-800M, implying about 25-33x net. A March 2026 data breach also paused Meta's work and drew class-action suits.