Mercor Seeks Funding at a $20 Billion Valuation — With a Pitch Promising $69 Billion in Revenue by 2030
The expert-data company Mercor booked $614 million in the first half of 2026, 91 percent of it from companies building foundation models. But behind the growth figures and talk of a $20 billion valuation lie two vulnerabilities: extreme customer concentration — and a security incident that led Meta to suspend its partnership.
The concrete numbers
Mercor, a company that supplies training data produced by subject-matter experts, recorded an annualized recurring revenue (ARR) of over $2 billion as of June 2026. In the first half of 2026 as a whole, the company reached $614 million in revenue, and 91 percent of that revenue came from companies developing foundation models — that is, the frontier model labs themselves (BigGo Finance, finance.biggo.com).
At the same time, the company is reportedly in talks for a new funding round at a $20 billion valuation, according to the same source. It is worth emphasizing what has actually been verified here: all the figures come from a single secondary account by BigGo Finance, which in turn cites The Information and interviews conducted by the investor Felicis. No primary document — no press release, financial statement, or direct statement from the company — exists in the available source material. The figures should therefore be read as the company's own account, relayed through one channel.
What Mercor actually does
Mercor began as a campus recruiting business, but today it supplies training data to OpenAI, Anthropic, and other companies building frontier models. The data is produced by contractors with professional expertise — lawyers, doctors, and programmers — who create, evaluate, and quality-check the content the models are trained on.
The scale of this operation was quantified in a company disclosure in May 2026: Mercor reported paying contractors more than $14 million per week, with over 30,000 active contractors each week. The company also claims to have an expert network of "nearly five million" people — but that, as the source itself points out, is a company estimate that cannot be read as the number of active or exclusive workers. The 30,000 weekly active contractors are thus the most concrete measure of how much of the network is actually working at any given time.
The economics of the model are easy to sketch: the frontier labs pay for expert-produced data, Mercor takes a margin, and payouts to contractors are the largest cost item. According to fundraising materials The Information has reviewed, Mercor has told prospective investors that gross margin is now 33 percent and is expected to improve to 56 percent. The same materials reportedly include a projection that total revenue will approach $69 billion by 2030. These are highly ambitious pitch figures from a company seeking funding — not established forecasts — and they should be treated as such.
Where the expert-data boom comes from
The figures say something about how frontier labs' data spending has shifted. As models have consumed most of the publicly available text on the internet, competitive advantage in model training has increasingly come down to data that is not freely available: domain knowledge, professional judgment, high-quality code, and evaluations performed by people who actually master the field. This is the market Mercor serves, and it explains why 91 percent of revenue comes from a relatively small number of foundation model companies with large training budgets.
The vulnerabilities: customer concentration and data security
That customer concentration is one of the two main vulnerabilities. When 91 percent of revenue comes from foundation model companies, the business depends entirely on a few large customers continuing to buy. The source material also notes that Mercor has business relationships with Chinese companies, including Tencent and Alibaba — Chinese model companies accounted for roughly 2 percent of revenue in the second quarter, according to BigGo Finance. That is a small share, but the connection is potentially politically fraught in a market where the largest customers are American.
The second vulnerability came clearly into view in March 2026. According to BigGo Finance, a security incident tied to the attack on the open-source dependency LiteLLM led Meta to suspend its partnership with Mercor. There are no details about the incident itself or Meta's reasoning in the available source material — the scope of the incident, what was exposed, and why Meta chose to suspend are not documented here. But the core is clear: a single security incident was enough for a major customer to pull away, illustrating how fragile such a supplier relationship is when the data being handled is competitively critical to customers.
From student project to data middleman
The company's own history is itself an indication of how fast the market has moved. Mercor was founded in 2023 by Brendan Foody, Adarsh Hiremath, and Surya Midha. According to interviews conducted by the investor Felicis and reported by BigGo Finance, Foody skipped his college final exams, and the three left their studies to become Thiel Fellows in 2024. Three years after founding, the source thus cites a company with $614 million in half-year revenue and talks about a $20 billion valuation.
What remains to be verified
Several central elements of the story rest exclusively on the company's own account, relayed through a single secondary source:
- The $20 billion valuation is described as ongoing talks, not a completed round. No details about date, investor, or round structure have been confirmed.
- The projection of $69 billion by 2030 and the margin improvement from 33 to 56 percent come from fundraising materials obtained by The Information — that is, sales material aimed at investors, not audited forecasts.
- The details of the LiteLLM incident and Meta's suspension have not been made public in the available material, and the consequences for Mercor's revenue base remain unclear.
- The expert network of nearly five million is a marketing figure; the actual workforce is better measured by the 30,000+ weekly active contractors.
What does emerge clearly, however, is the shape of the market: frontier labs are spending enormous sums on data created by people with specialized expertise, the money is flowing to companies that can organize such expert networks at scale — and suppliers in this market carry a double risk. They are at once highly dependent on a few customers and highly exposed to security incidents, because whatever leaks or gets compromised in such an operation is precisely what customers compete on.