Originally published on andrew.ooo
TL;DR: Mercor, founded in 2023 by three 21-year-old Thiel Fellows, has become the most capital-efficient AI startup ever measured. At $4.5 million in revenue per employee, they surpass Microsoft ($1.8M), Meta ($2.2M), and even Nvidia ($3.6M). The company hit a $10 billion valuation in October 2025 and is on track to reach $500M ARR faster than Cursor. Their secret? They pivoted from AI-powered hiring to becoming the essential human training layer for OpenAI, Anthropic, and Google DeepMind.
The Youngest Self-Made Billionaires in History
In October 2025, Forbes announced something unprecedented: three 22-year-olds had become the world's youngest self-made billionaires. Not through crypto speculation or inheriting a tech empire, but by building a company that solves one of AI's most critical bottlenecks.
Brendan Foody, Adarsh Hiremath, and Surya Midha—high school friends who competed together on the Bellarmine Speech and Debate Team in the Bay Area—had dropped out of college to pursue a simple idea. Just two years later, that idea became Mercor, a $10 billion company.
The Numbers That Defy Traditional SaaS
| Company | Revenue Per Employee | Notes |
|---|---|---|
| Mercor | $4.5M | Highest among AI startups |
| Cursor | $3.2M | Fastest to $1B ARR |
| Nvidia | $3.6M (FY 2025) | Chip monopoly |
| Meta | $2.2M (FY 2024) | 3.5 billion users |
| Microsoft | $1.8M (FY 2024) | Enterprise dominance |
What Does Mercor Actually Do?
Originally, Mercor was built to connect freelance programmers in India with companies in the United States. They developed an AI platform that could interview programmers and match them with hiring companies.
Then they saw a bigger opportunity.
OpenAI, Anthropic, Google DeepMind, and every major AI lab shared a common bottleneck: they needed human experts to train their models. Not just any humans—they needed scientists, doctors, lawyers, and engineers who could teach AI systems the subtleties that can't be captured in code.
Today's Business Model
- Recruitment: They identify and vet expert contractors
- Matching: AI algorithms pair these experts with AI labs
- Training: Experts perform reinforcement learning tasks
- Infrastructure: Mercor provides the software layer for managing feedback loops at scale
Current Scale
- 30,000+ expert contractors on their roster
- $1.5 million paid daily to contractors
- $85/hour average contractor compensation
- $500M ARR trajectory (faster than Cursor achieved it)
- Key clients: OpenAI, Anthropic, Google DeepMind
Why AI Labs Need Mercor
Modern AI models aren't just trained on data—they're trained on human preferences. When ChatGPT gives a helpful response, it's because thousands of hours of human feedback taught it what "helpful" means.
But as AI models get smarter, you need smarter humans to train them. Teaching an AI to reason like a lawyer or diagnose like a doctor requires actual lawyers and doctors.
The Revenue Per Employee Secret
Here's why Mercor generates $4.5M per employee:
Mercor's contractors are their product, not their employees.
When OpenAI pays Mercor for 1,000 hours of expert training time, that revenue counts against Mercor's ~50 core employees—not the hundreds of contractors who performed the work.
Key Takeaways for Founders
The Best Pivot is the Obvious One (In Hindsight) - Mercor's pivot from "AI hiring tool" to "AI training infrastructure" seems obvious now.
Marketplace Businesses Can Be Extremely Efficient - By keeping contractors off the payroll while building valuable infrastructure around them, Mercor achieves efficiency ratios that traditional SaaS can't match.
Timing the Market Matters - Mercor's growth accelerated when Scale AI's relationships fractured. Being positioned to capture that moment wasn't luck—it was preparation.
📖 Read the full article with detailed funding timeline, FAQ section, and more examples: andrew.ooo/posts/mercor-45m-revenue-per-employee-ai-training
Top comments (0)