In the spring of 2023, while his Georgetown classmates were cramming for finals, Brendan Foody was busy testing a different theory. “I just didn’t go to finals,” he told Fortune. Instead, he dropped out to build what would become the fastest-scaling AI startup of the decade.
Fast forward to March 2026: Foody, along with his high school debate teammates Adarsh Hiremath and Surya Midha, have landed on the Forbes Billionaires List at age 22—breaking Mark Zuckerberg’s long-standing record of becoming a billionaire at 23.
Their company, Mercor, hit a $10 billion valuation after raising $350 million in Series C funding. The investors? Felicis, Benchmark, General Catalyst, and Robinhood Ventures. The business model? Teaching AI what only humans know.
From $500/Week Gig to $10 Billion Empire
The Mercor origin story reads like startup folklore. Three Bay Area kids who met through high school debate competitions at Bellarmine Prep. They clicked at age 10 over forensics tournaments. All three became Thiel Fellows, receiving Peter Thiel’s $200,000 grant to skip college and build instead.
At a hackathon in São Paulo, they stumbled onto a simple arbitrage: connect companies with skilled engineers abroad, handle the logistics, take a cut. Their first client paid $500/week for a developer; Mercor paid the engineer 70% and kept the rest.
Within nine months, they hit $1 million in annual revenue run rate. But that was just the warm-up.
The Pivot That Changed Everything
Here’s where it gets interesting. Mercor started as an AI-powered hiring platform—basically matching companies with remote talent faster than LinkedIn ever could. Useful, but not revolutionary.
Then the AI labs came calling.
OpenAI, Anthropic, Google DeepMind—they all had the same problem. Their models could write code and solve math problems, but they struggled with judgment. Knowing what should be done, not just what could be done. Understanding nuance, taste, tradeoffs.
You can’t teach an AI model good judgment by scraping the internet. You need humans—specifically, humans who already have good judgment. Doctors. Lawyers. Investment bankers. Consultants. People who make high-stakes decisions for a living.
Mercor pivoted from “helping companies hire engineers” to “helping AI companies hire the humans who can train their models to think like professionals.” They now have 30,000+ domain experts on their roster, earning an average of $85/hour, with Mercor paying out more than $1.5 million daily to contractors.
The APEX Framework
Mercor’s secret weapon is APEX—the AI Productivity Index. While other benchmarks test AI on abstract puzzles and graduate-level math, APEX tests models on 200 real-world professional tasks. Can the model draft a competent legal brief? Write a financial memo that wouldn’t get you fired? Analyze a medical chart without hallucinating diagnosis codes?
To build it, they assembled an advisory board that sounds like a TED speaker lineup: former Treasury Secretary Larry Summers, ex-McKinsey managing partner Dominic Barton, legal scholar Cass Sunstein, cardiologist Eric Topol.
As Mercor puts it: “It’s great to have 10,000 PhDs in your pocket—it’s even better to have a model that can reliably do your taxes.”
The Scale AI Windfall
Mercor’s growth also benefited from one of 2025’s biggest AI industry shake-ups. When OpenAI and Google DeepMind reportedly cut ties with data-labeling giant Scale AI after Meta invested $14 billion and poached its CEO, they needed alternatives fast.
Mercor was there. They told investors they’re on track to hit $500 million ARR faster than Cursor (the AI coding tool that became the poster child for hypergrowth). That’s not a typo. $500 million in annual recurring revenue, in about three years.
The Philosophy of Human-AI Labor
Foody has a counterintuitive take on AI job displacement. “Everyone’s been focused on what models can do,” he says. “But the real opportunity is teaching them what only humans know—judgment, nuance, and taste.”
His argument: AI won’t eliminate labor, it will reallocate it. As software automates repetitive white-collar work, humans move up the value chain to teaching machines how to reason, decide, and create. Every evaluator who grades an AI output helps refine the model. Every rubric becomes training data.
“We’ll automate maybe two-thirds of knowledge work,” Foody predicts. “And that’ll be incredible, because it lets us do things like cure cancer and go to Mars.”
Is that techno-optimism bordering on delusion? Maybe. But he also has $10 billion in validation suggesting the market agrees.
What This Means
Three things stand out:
1. The Data Labeling Market Is Massive We talk about AI model training like it’s purely algorithmic. In reality, there’s a shadow economy of human evaluators, experts, and contractors feeding data and feedback into these systems. Mercor found a way to own the supply chain.
2. Founder Age Is Compressing Zuckerberg was 23 when he hit billionaire status. The Mercor trio did it at 22. The next generation might do it at 20. When the leverage is software and the moat is network effects, time-to-scale keeps shrinking.
3. The “Human-in-the-Loop” Era Is Here We spent years asking “when will AI replace humans?” The better question: “how will humans and AI work together?” Mercor’s bet is that even the most advanced models need human judgment to handle economically valuable work. If they’re right, “AI training” becomes a massive new job category.
Foody hasn’t taken a day off in three years. Even at dinner with his parents, he’s thinking about work. “People burn out when they work hard on things that don’t feel compounding,” he says. “I see the ROI of my time every day.”
That’s either inspiring or concerning, depending on your relationship with hustle culture. But at 22, with $2 billion in net worth and a business that might define how humans and AI collaborate for the next decade, he’s earned the right to be a little obsessive.
The Zuckerberg record stood for nearly two decades. In the age of AI, records don’t last that long.