Skip to content

AI data market

Mercor AI data business model: separating expert services from licensing

By SourceX Editorial · Reviewed by Noah Loul ·

9 sections

Short answer

Mercor's AI data business model is mostly an expert services marketplace: labs buy experts' hours, Mercor says it pays experts over $4 million a day, and contractors sign terms protecting the client's IP. That revenue is sold once. A second, less documented program licenses enterprise workflow data that can sell again. Mercor's figures are run-rates, daily pay and a $10 billion valuation, not audited revenue, so classify each line before comparing.

Value drivers

Mercor AI data business

This page covers mercor AI data business model: separating expert services from licensing.Illustrative

Key takeaways

  • Mercor's disclosures describe labs buying expert hours, with output protected for the client, which is services revenue sold once.
  • Its public figures are a $10 billion valuation, unaudited run-rates, daily pay to experts and pool size; none is audited revenue, and several carry no date.
  • Mercor's undated claim that it pays out over $2 million a day to contractors measures payments to workers, not revenue; annualized by our arithmetic, not Mercor's, it is about $730 million, the same order of magnitude as the $1 billion run-rate in the same post.
  • A five-question reuse test separates Mercor's marketplace from its enterprise data program, which licenses data that can be sold again.
  • The test investors should run is one number: the share of revenue that did not require paying an expert for new work.

What does Mercor's business model reveal about expert data demand?

Mercor's own disclosures describe mostly an expert services marketplace. AI labs pay for time from domain experts, Mercor says it pays those experts over $4 million a day, and contractors sign an agreement that protects the client's intellectual property and data. That is demand for made-to-order human judgment priced by the hour, not for a dataset that can be licensed again.

The distinction decides how an investor should read every Mercor number. A services marketplace grows with hours worked and passes a large share of revenue to the people doing the work; a licensing business grows with the number of licenses sold on an asset it already holds. Mercor also runs a program that licenses enterprise workflow data, so the real question is the mix between the two, which Mercor does not publish.

Useful evidence would be audited revenue split by line, pay to experts as a share of revenue, and the number of times any delivered dataset is licensed. The table below shows what is public as of 2026-10-26 and keeps the rest visibly unknown; the reuse test then classifies each line, and the alternative explanation tests that classification.

Mercor's disclosures, dated, with the blanks left blank

Seven fields cover what Mercor publishes about itself. Every figure is the company's own claim, and the date column shows how often a figure appears without a date in the excerpt verified for this article.

Two patterns stand out. The figures use at least four different measures, valuation, annualized run-rate, daily pay to experts and pool size, and none of them is audited revenue. The same measure also changes across pages: daily pay to experts appears as over $2 million, more than $3 million and over $4 million. That is consistent with growth, but the figures cannot be dated from the excerpts, so no growth rate can be computed from them.

Mercor's disclosures, dated, with the blanks left blank
FieldWhat Mercor saysWhere, and date shownMetric typeUnknown
Funding and valuation$350 million Series C led by Felicis at a $10 billion valuation, five times the Series B valuationSeries C post, October 27, 2025Price paid for equityRevenue at the time of the round
Revenue run-rateZero to a $500 million run rate in 17 months; later crossed $1 billion annualizedEngineering blog; no date in the excerptAnnualized run-rate, unauditedAudited revenue, period, run-rate definition
Pay to expertsOver $2 million a day, more than $3 million a day, and over $4 million a dayThree Mercor pages; no dates in the excerptsDaily payments to experts and contractors, as Mercor describes them; not revenuePeriod of each figure; share of revenue
Expert poolMore than 5 million domain experts; more than 30,000 weekly active contractorsNewsroom; engineering blogRegistered pool versus active workersHow expert and active are defined
ProfitabilityDescribes itself as a profitable Series C companyCareers pageSelf-descriptionMargin, period, accounting basis
Enterprise data licensingPays enterprises for workflow data, based on data volume, tools connected and data depthData pageProgram descriptionRevenue, deal count, reuse of licensed data
EvaluationAPEX benchmark: 200 expert-written cases in four professions, later extended to 400arXiv, September 2025Benchmark outputCommercial terms and revenue

Run-rate, pay to experts and valuation measure different things

Each Mercor figure answers a different question, and combining them needs dates and definitions that the pages do not give.

A run-rate annualizes a recent period's revenue. Mercor's engineering blog says it went from zero to a $500 million run rate in 17 months and later crossed $1 billion in annualized revenue run rate. A run-rate is not audited, its period is not stated, and a strong month annualized can overstate a year.

Pay to experts is closer to a cost line than to revenue. The same engineering post, which shows no date in the verified excerpt, says Mercor pays out over $2 million every day to more than 30,000 weekly active contractors. That is a payout figure, money paid to the people doing the work, and the post states revenue separately, as a run-rate. Annualizing the payouts is our arithmetic, not a Mercor figure: $2 million a day for 365 days is about $730 million a year, the same order of magnitude as the $1 billion run-rate stated in that post. Both company figures are floors and neither carries a date, so the share of revenue paid to experts cannot be computed from them. The arithmetic shows only that pay to experts is large relative to revenue, which is the signature of a labor marketplace rather than of a license with low marginal cost.

Pool size and activity measure different things too. Mercor's newsroom cites more than 5 million domain experts, while the engineering post counts more than 30,000 weekly active contractors. If both held at the same time, weekly active contractors would be under 1 percent of the pool; the pages carry no dates in the excerpts, so treat that as a question to ask, not a finding.

A valuation is a price. Mercor's Series C post of October 27, 2025 says Felicis led a $350 million round valuing the company at $10 billion, five times its Series B valuation. It records what investors paid on that date, not how fast revenue grew.

The reuse test: is the revenue sold once or licensed again?

Expert services revenue is sold once; licensing revenue can be sold again. Five questions sort any line of an AI data business into one side or the other, and Mercor's two disclosed lines land on opposite sides.

Ownership comes first. Mercor's contractor glossary describes the Confidential Information and Inventions Assignment Agreement, signed during onboarding with the worker agreement and offer letter, as a required document that protects the client's IP and data. Output made under those terms is the client's, which is what a lab buying made-to-order work wants, and it is why the same hour cannot be sold twice.

Reddit's filings show the other side. Its 2024 annual report says substantially all of the contract value behind its licensing revenue comes from two partners, so access to one body of platform data earns revenue from more than one licensee. Mercor's enterprise data program sits nearer this side: it pays enterprises for workflow data, with compensation based on data volume, the tools connected and the depth of the data. Whether a dataset delivered through the program is licensed to more than one buyer is not disclosed.

The test does not say which model is better. A services marketplace can grow quickly when demand is strong and compounds through recruiting and vetting; a licensing business can earn more than once on an asset but depends on rights and repeat buyers. What the test settles is which comparisons are fair: Mercor's marketplace belongs beside other expert-labor platforms, and its data program beside other licensors.

The reuse test: is the revenue sold once or licensed again?
QuestionExpert services: Mercor's marketplaceReusable licensing: Reddit's filingsMercor's enterprise data program
Who owns the output?The client; contractors sign an agreement protecting the client's IP and dataThe licensor keeps the data and grants rightsThe enterprise licenses its own records; terms not public
Can the same unit be sold again?No: made to one client's specificationYes: licensing contract value from two partnersNot disclosed
What grows revenue?More expert hoursMore licensees and renewalsMore enterprise suppliers and buyers; volumes not disclosed
Where is the marginal cost?Pay for each added hour of expert workLow for each added licensee once data is preparedExtraction and anonymization, which Mercor says it covers
What is the defensible asset?Recruiting, vetting and managing experts at speedRights to data others cannot reachSupplier relationships and rights; not yet evidenced publicly
A fork diagram: an expert hour flows to one client-owned deliverable, while a licensed dataset flows to two licensees and a possible next one, with Mercor's mix between the two marked not disclosed.
The reuse test as a picture: services revenue is sold once, licensing revenue can be sold again. Structure, not Mercor data.

What the demand signal says: labs are buying judgment by the hour

Mercor's disclosures say AI developers are buying expert judgment by the hour in specific professions, and that the work increasingly targets evaluation of professional tasks.

Its first AI Productivity Index (APEX), released in September 2025, has 200 test cases across investment banking, management consulting, law and primary medical care, with prompts and grading rubrics written by experts; an extended version doubled the held-out set to 400 cases. The cases were written by experts in those professions, which points to demand concentrated in high-wage knowledge work where model errors are costly.

Scale AI's account of its own 2025 points the same way: it reports that its data business delivered more than 150,000 hours of data. When two of the most cited companies in the category measure data work in hours, the category is selling labor as well as assets, and investors should model it that way.

The limit is that these are seller-side signals. They show what Mercor and Scale chose to build and report, not what labs spend in total or how that spending divides between expert hours, licenses and in-house work.

The strongest alternative: a data company inside a services business

The strongest alternative is that Mercor is building reusable assets inside a services business, and the evidence supports it in part.

APEX is a reusable asset: a benchmark built once from expert work and published for repeated use. The enterprise data program is licensing by design. If either grows into a large share of revenue, Mercor's economics would move toward a licensor's, with revenue that does not need a new expert hour for each dollar earned.

Against that, the figures Mercor emphasizes, daily pay to experts and weekly active contractors, are labor metrics, and the contractor terms route output to clients. Until Mercor reports revenue by line, the defensible reading is a services marketplace with two asset-building lines of unknown size. An investor can test that reading with one number: the share of revenue that did not require paying an expert for new work.

What could make this reading of Mercor wrong

Five risks could make this reading wrong or make Mercor's model less durable than its figures suggest.

At a glance

What could make this reading of Mercor wrong

  • Undated, unaudited metrics: run-rate and daily pay figures appear without dates in the verified excerpts, so growth cannot be computed from them and should not be.
  • Buyer concentration: frontier labs are few, and Mercor does not disclose its largest customers' share of revenue.
  • Automation of expert work: Scale says its own evaluations evolved from expert-centered assessment to a mix of domain expertise and automated assessment; more automated grading would change demand for expert hours.
  • Rights in the data program: licensing enterprise workflow data depends on each supplier's rights in content from customers, employees and platforms, which a program description cannot establish.
  • Valuation as evidence: a $10 billion price records investor expectations on one date and says nothing about revenue quality.

Questions to ask Mercor, any expert-data company and SourceX

Ask for numbers that sort revenue into services and licensing, with dates and definitions. Those questions apply to SourceX as well.

SourceX sits on the licensing side of the reuse test. It helps companies license business records they select to AI developers as datasets, and for those companies it takes on the sourcing and qualification, the rights review and the contracting as well as data preparation, delivery and payment. The licensing side shows in the contract: a company signs only after approving scope and terms, nothing is delivered before that signature, and the signed agreement is what permits SourceX to use the de-identified dataset and to sublicense and distribute it, within its terms. SourceX publishes no revenue, run-rate or deal metrics and shares company and commercial metrics with investors directly. It is raising from accredited investors under Rule 506(c); investors can request materials and put the questions below to it.

Checklist · 6 items

  1. What was audited revenue by line for the last fiscal year, and how is run-rate defined?
  2. What share of revenue is paid to experts, and how has it moved?
  3. What share of revenue required no new expert hours?
  4. How many times has a dataset delivered through the data program been licensed, and to how many buyers?
  5. What share of revenue comes from the largest customer and from the five largest?
  6. Who owns each output, and which outputs may be reused or resold?

0 of 6 ticked. Tick each item as you go.

Methods note: what the Mercor evidence table and reuse test rest on

The evidence table and the five-question reuse test are editorial frameworks built for this article on 2026-10-26. The evidence set is Mercor's own pages on mercor.com (the Series C post, mission page, newsroom, careers page, data page and two engineering posts), its contractor documentation glossary, the APEX paper on arXiv, Reddit's 2024 annual report, and Scale AI's look back at 2025 and SEAL leaderboards post.

Excluded: valuations and revenue figures for Mercor reported only by the press, later funding reports not confirmed on Mercor's own domain, job postings and third-party estimates. Every source was verified through search excerpts on its own domain and should be opened by a reviewer before release; several Mercor pages showed no date in the excerpt, and the table says so.

Not measured: Mercor's audited revenue, revenue by line, the share paid to experts, customer concentration or reuse of any delivered dataset. The figure of about $730 million is our annualization of Mercor's undated claim that it pays contractors over $2 million a day; it is not a figure Mercor reports, and it is not revenue. No company is ranked.

Frequently asked questions

How does Mercor make money?

Mercor's pages describe two lines: a marketplace in which AI labs pay for work by domain experts, whom Mercor says it pays over $4 million a day, and a program that pays enterprises for workflow data and licenses it for AI research. Mercor does not publish how revenue splits between them, though it describes itself as profitable.

What is Mercor's revenue run rate?

Mercor's engineering blog says it went from zero to a $500 million run rate in 17 months and later crossed $1 billion in annualized revenue run rate. A run-rate annualizes a recent period and is not audited, and the excerpt verified here shows no date, so it should not be read as annual revenue.

Is Mercor a data company or a staffing company?

On its own disclosures, mostly an expert services marketplace with two asset-building lines: its APEX benchmarks and its enterprise data licensing program. The labor metrics it emphasizes, daily pay to experts and weekly active contractors, point to services; its revenue mix by line is not public.

Who owns the work Mercor's experts produce?

Mercor's contractor glossary describes a required Confidential Information and Inventions Assignment Agreement that protects the client's intellectual property and data. Contractors can read the full agreement before signing, and that text, not a summary, governs ownership.

How should investors compare Mercor with Scale AI or listed data companies?

Only on the same metric and period. Mercor reports run-rate, daily pay to experts and a valuation; Scale AI reports bookings, growth rates and contract ceilings; listed peers file audited revenue. Until one metric is public for all of them, record the comparison as not possible rather than ranking them.

Sources

  • Mercor's Series C post (October 27, 2025) says Felicis led a $350 million round that values Mercor at $10 billion, five times its Series B valuation. Source
  • Mercor's mission page says domain experts on its platform are paid over $4 million per day to train frontier AI models. Source
  • Mercor's newsroom cites more than 5 million domain experts across hundreds of professional fields. Source
  • A Mercor engineering post says the company went from $0 to a $500 million run rate in 17 months, crossed $1 billion in annualized revenue run rate, and pays out over $2 million every day to more than 30,000 weekly active contractors. Source
  • A later Mercor engineering post says the company processes more than $3 million a day in contractor payments. Source
  • Mercor's careers page describes the company as a profitable Series C company. Source
  • Mercor's data page describes paying enterprises for real-world workflow data supplied for frontier AI research, with compensation based on data volume, the tools connected and the depth of the data, and says Mercor covers extraction and anonymization costs. Source
  • Mercor's contractor glossary describes the Confidential Information and Inventions Assignment Agreement, signed at onboarding with the worker agreement and offer letter, as a required document that protects the client's IP and data. Source
  • The AI Productivity Index (APEX), from Mercor (September 2025), has 200 test cases across investment banking, management consulting, law and primary medical care with expert-written prompts and rubrics; an extended version doubles the held-out set to 400 cases. Source
  • Reddit's 2024 annual report says substantially all of the contract value associated with its licensing revenue is derived from two partners. Source
  • Scale AI says its data business delivered more than 150,000 hours of data in 2025. Source
  • Scale AI says its SEAL evaluations initially centered on human expert evaluations and evolved to combine domain expertise with automated assessments. Source

Request SourceX investor materials

SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.

See if your company qualifies