Mercor Revenue: How It Grew From $1M to a $2B Run Rate

Mercor started as a marketplace that matched software engineers in India with startups in the United States.

That was not the business that made it huge.

The real growth started when Mercor realized that AI labs needed thousands of skilled people to help train and evaluate their models. Mercor had already built software to find, interview, and match talent. It turned that recruiting system into infrastructure for supplying experts to AI companies.

The result was unusual even by AI startup standards.

Mercor went from roughly $1 million in annualized revenue to about $500 million in 17 months. It crossed a $1 billion annualized run rate in early 2026. By June 2026, The Information reported that Mercor had reached more than $2 billion in gross annualized revenue.

That last number needs an important caveat: it is not $2 billion of SaaS ARR. Mercor bills customers for contractor work and then pays a large share of that money to the contractors. Reporting suggests roughly two-thirds of gross revenue goes to them.

The growth story is still remarkable.

Mercor grew because it did five things well:

It started with a narrow marketplace where the founders could manually find supply and demand.
It automated the slowest parts of recruiting with AI.
It followed customer pull from normal hiring into AI training and evaluation.
It built a referral loop that brought more experts onto the platform.
It expanded deeply inside a small number of AI labs with enormous budgets and urgent demand.

This is how it happened.

Mercor's growth in one chart

Date	Milestone
January 2023	Brendan Foody, Adarsh Hiremath, and Surya Midha found Mercor
2023	Mercor starts matching software engineers outside the U.S. with American startups
January 2024	Mercor announces a $3.6M seed round and says it has 100,000 candidate profiles across 25 countries
September 2024	Mercor raises roughly $30M at a $250M valuation; candidate pool reaches 300,000+
February 2025	Mercor reaches about $75M in annualized revenue and raises $100M at a $2B valuation
September 2025	Brendan Foody says Mercor went from $1M to $500M in revenue run rate in 17 months
October 2025	Mercor raises $350M at a $10B valuation
Early 2026	Mercor says it crosses $1B in annualized revenue run rate
June 2026	The Information reports more than $2B in gross annualized revenue
September 2026	Mercor says it has 5M+ expert profiles, 400+ employees, and pays its expert network about $4M per day

1. Mercor started with Indian software engineers

Mercor was founded in January 2023 by Brendan Foody, Adarsh Hiremath, and Surya Midha.

The three had met years earlier at Bellarmine College Preparatory in San Jose. Foody and Midha later went to Georgetown. Hiremath went to Harvard.

The first version of Mercor was simple.

American startups needed engineers. Skilled engineers outside the United States wanted access to better-paying American work. Mercor connected the two sides and took a fee.

According to a later profile by investor Felicis, the idea took shape during a hackathon in São Paulo. The founders had already hired engineers from India for their own projects. They then started helping friends do the same.

Foody has said they partnered with the coding club at IIT Kharagpur.

The early operation looked more like a scrappy staffing business than an AI company.

Candidates were found manually. Interviews were done manually. Matches were tracked in tools such as WhatsApp and Google Sheets. The first customers were friends and seed-stage startups.

But it worked.

Before raising institutional funding, Mercor said it had already reached seven figures in annualized revenue.

That gave the founders an important signal: companies wanted access to global talent.

It also exposed the first bottleneck.

Three founders could not manually interview every candidate.

2. Mercor automated the recruiting work

The next step was to turn the manual process into software.

Mercor built crawlers that pulled information from resumes, GitHub profiles, portfolio websites, and other public sources.

Then it built an AI interviewer.

A candidate could join a video call with an AI agent, answer questions about their experience, and complete one assessment that Mercor could reuse when matching them to future jobs.

Employers could search the talent pool with natural-language queries, watch candidate interviews, and hire people through the platform.

In January 2024, Mercor announced a $3.6 million seed round led by General Catalyst and described the product as a fully automated hiring platform.

At the time, Mercor said it had already built a pool of 100,000 candidates across 25 countries.

By September 2024, Reuters reported that the candidate pool had grown beyond 300,000.

This software mattered for more than efficiency.

It made each candidate reusable.

Instead of starting from zero whenever a company opened a role, Mercor could search a growing database of people it had already assessed.

Mercor later said that more than half of offers were going to people who had not applied to the specific job they were offered.

That turned recruiting from a series of one-off searches into a searchable network.

3. A meeting with xAI showed Mercor a much bigger market

The most important change happened before Mercor became well known.

In August 2023, one of Mercor's customers introduced the founders to the co-founders of xAI.

Foody later said the introduction happened because Mercor had access to strong Indian engineers who were good at math and coding.

The xAI team was interested in the quality of those people.

Mercor did not start working with xAI immediately. But the meeting gave the founders a clue.

Frontier AI companies were going to need large numbers of highly skilled humans.

At the time, much of the AI data industry was built around lower-skilled labeling and annotation. As models improved, the hard problems started moving toward areas where the person creating or evaluating the data needed real expertise.

A model doing advanced math needed feedback from someone who understood advanced math.

A coding agent needed software engineers.

A model doing legal work needed lawyers.

A finance model needed people who understood finance.

Mercor already had a system designed to find and assess people with those skills.

The company had accidentally built useful infrastructure for a much bigger market.

4. Mercor moved from recruiting employees to supplying experts for AI training

Mercor initially worked with AI labs through Scale AI.

Felicis says Mercor recruited more than 1,000 engineers for Scale projects.

That was another important signal.

The demand was not for a few full-time hires. AI companies needed large groups of people who could work on model-training projects for days, weeks, or months.

Mercor eventually started working directly with AI labs.

By February 2025, TechCrunch reported that most of Mercor's roughly $75 million annualized revenue was already coming from AI labs. Mercor said it was working with the world's top five AI labs, including OpenAI.

This changed the economics of the business.

A normal startup might hire a few engineers through Mercor.

A frontier AI company might need hundreds or thousands of experts across many fields.

And the work kept changing.

One project might require mathematicians. The next might require software engineers. Then lawyers, doctors, investment bankers, consultants, writers, or researchers.

Mercor's customer did not need one employee.

It needed a system that could repeatedly find the right human expertise.

That is a much larger problem.

5. Mercor found customers with urgent, almost unlimited demand

One reason Mercor grew so quickly was the type of customer it found.

Frontier AI labs were spending billions of dollars trying to improve their models.

If better training data could improve a model even slightly, that improvement could be worth a lot.

Foody has argued that these companies were willing to spend heavily on anything that improved model capabilities.

That created unusually strong demand.

It also created unusually large expansion inside existing accounts.

In a 2025 interview with Lenny Rachitsky, Foody said Mercor had never lost a customer and had net revenue retention above 1,600%.

Those are company claims, not independently audited figures. But they help explain the shape of Mercor's growth.

Mercor did not need to sign thousands of normal SaaS customers.

A small number of AI labs could keep spending more as their need for expert data grew.

The Information reported that about 91% of Mercor's gross revenue in the first half of 2026 came from foundation-model companies.

That concentration is a risk.

It is also one of the main reasons Mercor could grow so fast.

6. Short AI projects forced Mercor to automate everything

AI labs did not behave like normal employers.

Their needs changed with model-training cycles.

A lab might need a few hundred experts almost immediately. It might run the project for several weeks, pause it, then return with a different task.

Mercor has said customers made requests such as finding 300 qualified people in two days.

A traditional staffing company would struggle to do that repeatedly.

Mercor had to automate more of the workflow:

finding candidates
assessing them
matching them to projects
onboarding them
tracking work
paying contractors
collecting performance data
rematching good workers to new projects

This became one of Mercor's biggest advantages.

The system got better because the company was not only collecting resumes and interview results. It was also seeing how people performed on real projects.

In its 2025 "master plan," Mercor said AI labs could send performance data back within days. A normal employer might take months to learn whether a hire was good.

That shortened feedback loop helped Mercor improve matching much faster.

The basic loop became:

assess experts → place them on projects → observe performance → improve matching → place them again

7. Referrals solved the supply problem

Demand from AI labs was only useful if Mercor could find enough qualified people.

Referrals became a major growth channel.

As Mercor expanded from software engineering into law, medicine, finance, science, consulting, writing, and other fields, it needed a constant supply of new experts.

Existing experts helped bring them in.

In October 2025, Mercor said more than half of new experts came through referrals.

By March 2026, Mercor said it had more than 1.9 million referrals and more than four million vetted expert profiles.

Its current referral program uses an earnings-sharing model. A referrer earns 20% of what a referred expert earns, up to a cap.

That creates a useful marketplace loop:

more AI projects → more experts earn money → experts refer other experts → Mercor can fill more projects → AI labs send more work

Unlike broad paid acquisition, the referral cost is tied to successful work.

The program also helps Mercor reach specialists who may never search for an "AI training job" themselves.

8. Mercor focused on higher-skilled work while AI models improved

Mercor also benefited from a shift inside the AI industry.

Early model training depended heavily on large amounts of basic labeling.

As models became more capable, labs needed harder examples and better evaluations.

The value moved toward experts who could judge whether a model was doing real professional work correctly.

That played directly into Mercor's strength.

In one interview, Foody described an early project where Mercor found 25 Olympiad medalists in 24 hours.

Later, the work expanded into professional domains such as software engineering, law, finance, medicine, consulting, and research.

This matters because expert work is harder to commoditize than basic labeling.

Mercor is not only selling hours of labor.

It is selling access to people who can define what a correct answer, good decision, or successful workflow looks like.

9. The Meta–Scale deal gave Mercor another boost

Mercor was already growing quickly when Meta made a major investment in Scale AI in 2025 and hired Scale CEO Alexandr Wang.

The deal created a problem for Scale.

Scale worked with AI labs that directly competed with Meta. Once Meta became closely tied to Scale, some customers had a reason to look for alternatives.

TechCrunch reported that leading AI labs including OpenAI and Google DeepMind cut ties with Scale after the deal.

Mercor was well placed to benefit.

By then it already had relationships with major labs, a large expert network, and infrastructure for running training projects.

Foody later said Mercor's business grew sharply after the Scale deal.

But the timing matters.

The Scale disruption did not create Mercor's business.

Mercor had already reached a nine-figure revenue run rate before it happened.

The deal accelerated a company that already had product-market fit.

10. Mercor moved up the stack from labor to evals and training environments

The next stage of Mercor's growth is less about finding people.

It is about packaging what those people know.

Mercor has expanded into model evaluations, benchmarks, datasets, and reinforcement-learning environments.

Its APEX benchmark family tests whether AI systems can complete economically useful work in areas such as software engineering, accounting, investment banking, law, consulting, and medicine.

In 2026, Mercor also moved further into reinforcement-learning infrastructure.

It acquired Sepal AI, a company working on training data, benchmarks, and RL environments.

In July 2026, Mercor announced plans to acquire Deeptune, which builds simulated versions of tools and enterprise software that AI agents can practice using.

This is a logical extension of the original business.

First, Mercor found experts.

Then it used those experts to create training data.

Now it can use experts to design tasks, judge model performance, and help build the environments where agents learn.

The company is moving from supplying labor to supplying more of the system used to train AI.

How fast did Mercor actually grow?

The numbers are easy to misunderstand because Mercor is not a normal SaaS company.

Here is the clearest version of the reported growth:

Mercor started 2024 with under $1 million in annual revenue, according to the company.
By February 2025, TechCrunch reported roughly $75 million in annualized revenue.
By September 2025, Foody said Mercor had gone from $1 million to $500 million in revenue run rate in 17 months.
Mercor says it crossed $1 billion in annualized revenue run rate in early 2026.
The Information reported that Mercor reached more than $2 billion in gross annualized revenue in June 2026.
The Information also reported $614 million in gross revenue for the first half of 2026.

The word gross is important.

Mercor bills customers and then pays the experts doing the work.

The Information reported that roughly 60% to 70% of gross revenue is paid to contractors. Its later reporting said contractor payouts were about two-thirds of gross revenue and that Mercor's gross margin was 33% in the second quarter of 2026.

So a $2 billion gross run rate does not mean Mercor keeps $2 billion.

It also does not mean Mercor generated $2 billion over the previous 12 months.

It means the most recent revenue pace, annualized, was above $2 billion before contractor payouts.

That is still extraordinary growth. It is simply a different business model from a software company with $2 billion in subscription ARR.

Mercor's growth loop

Mercor's growth can be reduced to one loop:

AI labs need better data and evaluations.
Mercor finds skilled experts.
AI interviews and assessments make those experts searchable.
Experts work on short AI projects.
Mercor learns from their performance.
Good experts get matched to more work.
Experts refer other experts.
A larger expert network lets Mercor serve more projects and new domains.
Labs spend more because Mercor can fill more of their needs.

Each side strengthens the other.

More demand attracts more experts.

More experts let Mercor serve more demand.

Performance data makes matching better.

And large customers can expand far faster than a normal SaaS account.

What Mercor did differently

It followed the customer pull

Mercor did not stay attached to its original idea of recruiting software engineers for startups.

The founders noticed that AI labs had a much more urgent problem and moved toward it.

That decision changed the size of the company.

It automated before hiring armies of recruiters

The original manual process showed Mercor what needed to be automated.

AI interviews, reusable assessments, search, onboarding, and payments let the company process far more people than the founding team could handle manually.

It found a market where speed mattered

Finding 300 people in two days sounds unreasonable in normal recruiting.

In frontier AI, it can be a real customer request.

Mercor built for that speed.

It made supply reusable

A candidate was not useful for only one open role.

Once Mercor assessed someone, that person could stay in the network and be considered for future projects.

This made the supply side more valuable over time.

It built a referral loop instead of relying only on paid acquisition

When experts make money, they have a reason to invite other qualified experts.

Mercor says referrals became one of its largest sourcing channels.

It landed a few customers that could expand enormously

Mercor's growth is heavily concentrated in frontier AI labs.

That creates risk, but it also explains the speed.

When a customer has billions to spend and a rapidly growing need for expert data, expansion can happen much faster than it does in normal B2B software.

What could slow Mercor down?

Mercor's growth has been exceptional, but there are real risks.

The first is customer concentration.

The Information reported that foundation-model companies generated about 91% of Mercor's gross revenue in the first half of 2026. If a few large labs reduce spending, switch vendors, bring the work in-house, or use more synthetic data, Mercor could feel it quickly.

The second is supply quality.

Five million profiles are useful only if Mercor can identify which people are genuinely good, available, and willing to do the work.

The third is that AI training itself keeps changing.

The work may move from human-written examples toward simulated environments, automated verifiers, synthetic data, or new training methods.

Mercor is already responding by expanding into benchmarks and RL environments.

The fourth is operational risk.

Managing tens of thousands of contractors, large customer projects, payments, private data, and fast-changing workloads is much harder than running a normal software product.

Mercor's future depends on whether its systems can keep improving as fast as its revenue has.

The biggest lesson from Mercor's growth

Mercor did not grow because it found one clever marketing channel.

It grew because it kept moving toward stronger demand.

The first wedge was global recruiting.

The next was automated assessment.

Then came expert labor for AI training.

Then evaluations.

Now Mercor is moving into datasets, benchmarks, and reinforcement-learning environments.

The common thread is the same: find skilled people, understand what they are good at, and deploy that expertise where it is most valuable.

The company's original recruiting product turned out to be useful infrastructure for an entirely different market.

That is the main lesson from Mercor.

The founders did not predict every step.

They built something useful, watched where demand became unusually strong, and followed it.

Borrowing Mercor's referral loop

Mercor's supply grew because experts brought in other experts, and that starts with being useful where those people already talk. Okara's Reddit Agent finds those conversations in your niche and drafts replies, and the LinkedIn Agent keeps you visible to the professionals you want to reach.

Frequently asked questions

How did Mercor grow so fast?

Mercor started by matching software engineers outside the United States with American startups. It automated candidate sourcing and interviews, creating a reusable talent pool. The major growth inflection came when AI labs began using Mercor to find experts for model training and evaluation. Large AI customers expanded spending quickly, while referrals helped Mercor grow its expert network.

What was Mercor's main growth channel?

There was no single channel. Early customers came largely through founder networks and word of mouth. On the supply side, referrals became one of Mercor's biggest channels. On the demand side, growth came from landing frontier AI labs and expanding deeply inside those accounts.

What was Mercor's biggest pivot?

Mercor moved from helping startups hire software engineers to supplying experts who train and evaluate AI models. That shift turned a recruiting marketplace into an AI data and evaluation business.

Did Mercor reach $2 billion in ARR?

Not in the normal SaaS sense. The Information reported that Mercor reached more than $2 billion in gross annualized revenue in June 2026. The figure annualizes a recent revenue period and includes money later paid to contractors. Reporting suggests roughly two-thirds of gross revenue goes to contractor payouts.

Who are Mercor's customers?

Mercor says it works with the top five AI labs and six of the Magnificent Seven technology companies. Public reporting has named customers or relationships including OpenAI, Anthropic, Google DeepMind, Meta, Nvidia, and other AI companies.

How large is Mercor today?

As of September 2026, Mercor's newsroom says the company has more than five million domain-expert profiles, more than 400 employees, and pays roughly $4 million per day to its expert network. Its latest confirmed funding round valued the company at $10 billion.

What is Mercor building next?

Mercor is moving beyond expert staffing into datasets, model evaluations, benchmarks, and reinforcement-learning environments. Its APEX benchmarks measure how well AI systems perform real professional tasks, while acquisitions such as Deeptune extend the company into simulated software environments for training AI agents.

What is Mercor's valuation?

Mercor raised $350 million at a $10 billion valuation in October 2025. That was up from a $2 billion valuation in February 2025 and $250 million in September 2024. In July 2026, Bloomberg reported that Mercor was in talks to raise at a $20 billion valuation. That round had not been confirmed as of September 2026.