Three Bay Area high school friends started Mercor in 2023. By October 2025, the company hit a $10 billion valuation on a $350 million Series C, making all three founders, then 22 years old, the youngest self-made billionaires. By early 2026, the company was profitable on a free-cash-flow basis.
The product: a technology-enabled talent marketplace that connects skilled professionals with AI companies. The workers train AI models. They teach chatbots to think more like humans by "sharing knowledge, experience, and context that can't be captured in code alone." Mercor sits in the middle, taking a cut.
That's the clean version. The full picture includes a data breach, a trade-secrets lawsuit from their largest competitor, and structural questions about what happens to the worker data that flows through the platform.
How the money flows
Mercor manages over 30,000 contractors who are collectively paid over $1.5 million per day. The revenue model is a percentage-based recruiting fee, typically around 30%, for direct talent placements. If a contractor earns $100, Mercor keeps roughly $30.
The math at scale: $1.5 million per day in contractor payouts implies roughly $550 million per year in payroll, plus $235 million in Mercor's recruiting fees, for total platform volume approaching $800 million annually.
The investor list reads like the who's who of venture capital. The Series C was led by Felicis with participation from Benchmark, General Catalyst, and Robinhood Ventures. The Series B, just months earlier, had valued the company at $2 billion. The 5x jump to $10 billion in a single round is unusual even by Silicon Valley standards.
What the workers actually do
Mercor's contractors fall into several categories:
RLHF trainers. Reinforcement Learning from Human Feedback is the process of rating AI outputs so the model learns which responses humans prefer. This is the core of what makes ChatGPT, Claude, and Gemini feel conversational rather than robotic. The workers compare model outputs, rank them, explain why one is better, and the model adjusts.
Domain experts. Doctors, lawyers, engineers, financial analysts who provide expert feedback on AI outputs in their specialty. The model can generate a medical summary, but only a doctor can say whether the summary would mislead a patient.
Coders. Software engineers who review, correct, and improve AI-generated code. They verify that the model's code compiles, handles edge cases, and follows best practices. This is the human labor behind the coding benchmarks.
Data annotators. Workers who label, categorize, and clean training data. The unglamorous foundation that every model depends on.
The data web
Here's where it gets complicated.
Mercor sees everything. Every contractor's resume, work history, skill assessments, performance ratings, hourly output, quality scores, and payment history flows through the platform. Every interaction between a contractor and an AI model is logged. The platform knows which contractors are best at which tasks, how fast they work, what their error rates look like, and how much they're willing to work for.
That data has multiple uses:
- Matching. Mercor uses it to route the right contractor to the right task. This is the stated purpose.
- Enterprise products. In March 2026, Mercor launched Mercor Enterprise AI, which packages their tooling into workflow capture, agent specification, quality guardrails, and continual learning. The data from contractor interactions trains these enterprise tools.
- Benchmarking. Mercor and Cognition co-launched APEX-SWE in March 2026, a software engineering benchmark. Benchmark creation requires large volumes of human-verified coding data, which Mercor generates through its contractor workforce.
The question that nobody in the investor presentations asks clearly: when a contractor trains an AI model through Mercor, who owns the interaction data? The contractor generated it. The AI company commissioned it. Mercor's platform captured it. And Mercor is building enterprise products and benchmarks from it.
The cracks
The data breach
In April 2026, Mercor confirmed a data breach tied to the LiteLLM supply-chain attack. A third-party forensics investigation is underway. The details are still emerging, but a supply-chain compromise means the breach didn't come through Mercor's own systems. It came through a dependency in their stack.
For a company that handles the resumes, performance data, and payment information of 30,000 contractors, plus the proprietary training data of their AI company clients, a breach is material. The specific data exposed hasn't been fully disclosed as of this writing.
The Scale AI lawsuit
Scale AI, the largest AI data labeling company, filed suit against Mercor and a former Scale employee named Eugene Ling. The allegation: Ling downloaded over 100 proprietary customer strategy documents to a personal drive before joining Mercor. If the allegations are true, Mercor's growth strategy may have been informed by a competitor's internal playbook.
The lawsuit is ongoing. Mercor has not publicly detailed their defense. The case is worth watching because it touches on the trade-secrets dynamics that are common in the AI data industry, where the competitive advantage is often in the customer relationships and operational playbooks rather than the technology itself.
The contractor economics
Mercor's 30% fee is standard for talent marketplaces. But the workers training AI models are generating value that compounds far beyond the hours they work. A single hour of expert RLHF feedback improves a model that serves millions of users. The worker gets paid for the hour. The model improvement persists indefinitely.
This isn't unique to Mercor. Every AI data company operates on the same asymmetry. But Mercor's $10 billion valuation, built on the back of $550 million in annual contractor payroll, makes the gap between worker compensation and platform value unusually visible.
What this means if you're a contractor
If you're considering working through Mercor or a similar platform:
- Read the contractor agreement carefully. Understand what rights you're granting over your interaction data and work product.
- Understand that your performance data is being used to build platform products and benchmarks, not just to match you with tasks.
- Monitor the breach disclosure. If you've worked through Mercor, your personal and professional data may have been exposed.
- Track the Scale AI lawsuit. The outcome may clarify how trade secrets and competitive practices work in this industry.
The pay is real and the work is flexible. But the data dynamics deserve the same scrutiny you'd give any platform that monetizes your output at a multiple of what they pay you.
If the broader question of how to value your own labor and build independent income interests you, The W-2 Trap covers why trading hours for dollars, whether through a salary or a gig platform, has structural limits that don't scale. Search "The W-2 Trap" on Amazon Kindle.
Related reading
- Babel Audio pays up to $50/hr to read aloud — another AI data platform, different model, similar worker dynamics
- Build an AI job agent and an inference engine that sizes itself — automating the job search pipeline
- Escape the rat race: the complete guide — the bigger picture on labor economics and exit strategies
- Every time an AI model update broke something — the models these contractors train sometimes regress anyway
Fact-check notes and sources
- $10 billion valuation, $350M Series C (October 2025), founded 2023: techcrunch.com and cnbc.com.
- 30,000 contractors, $1.5M/day payroll: mercor.com/blog/series-c and techfundingnews.com.
- 30% recruiting fee model: sacra.com/c/mercor.
- Profitable on FCF basis, $6M profit H1 2025: sacra.com/c/mercor.
- Investors (Felicis, Benchmark, General Catalyst, Robinhood Ventures): techcrunch.com.
- LiteLLM supply-chain data breach (April 2026): wikipedia.org/wiki/Mercor and industry reporting.
- Scale AI lawsuit (Eugene Ling, 100+ proprietary documents): wikipedia.org/wiki/Mercor.
- APEX-SWE benchmark co-launch with Cognition (March 2026), Mercor Enterprise AI launch: mercor.com/blog/series-c.
This post is informational, not legal, employment, or financial advice. Mentions of Mercor, Scale AI, Cognition, and all investors are nominative fair use. No affiliation is implied.
