20VC Exclusive: Mercor Raises $100M at a $2BN Valuation: Scaling to $70M in ARR in 24 Months | 9-9-6: 9AM-9PM - 6 Days Per Week: The Most Intense Culture in Silicon Valley | The Future of Programming, Models and Data with Adarsh Hiremath

Mercor's Young Founders Turn AI's Data Bottleneck Into a Hiring Marketplace

Mercor argues that sourcing human expertise for AI labs is a talent-matching problem, and a hiring marketplace is the natural way to solve it. Models keep improving but still leave a long tail of tasks that need experts to finish, so data stays the bottleneck. That demand, plus automated hiring and marketplace network effects, drove growth to $70M ARR in 24 months.

Takeaways

• Human data and talent assessment are now one problem. Crowdsourced labelling once sufficed for AI training. Labs now need domain experts, and finding them is a talent-matching task. Mercor places experts with labs the way it places staff at any company, and net retention is above 100%.
• Expert humans are the bottleneck for smarter models. Harry cites a Grok contact who says synthetic data often beats low-quality human data. Adarsh agrees synthetic data helps but says experts unlock the next level, since AI gets a task 60 to 80 percent done and humans finish.
• Foundation models will be few, applications many. Adarsh expects only a couple of companies to build foundation models, with others building on top and many models serving different use cases. The market is shifting to reinforcement learning, and data is the bottleneck.
• Mercor grew to $70M ARR with a lean machine. The company has no sales team beyond its founders and a fully automated hiring process from listing to paid work. Take rate is set case by case, above 30% for some customers, with quality mattering more than price.
• Culture is harder to scale than software. The 9-9-6 schedule was floated mainly so people would not work Sundays. Mercor selects for care about the mission because it can teach many skills but not care. Rapid growth acts as a perpetual stress test.
• Programming matters more as software commoditises. Adarsh says programming rises in importance at a higher level of abstraction while coding agents commoditise software. Network-effect businesses like Meta and Airbnb will thrive, and the next SaaS wave will replace whole services end to end.
• The $100M round is built for a long game. Mercor raised $100M at a $2B valuation, with Harry calling 5% dilution phenomenal. Adarsh says the company will not deploy it immediately and wants a balance sheet matching the long-term goal of a unified global labour market.

How Mercor started

1. Adarsh, Brendan and Suriya start a software dev shop and recruit strong engineers from India.
2. They realise the people matter more than the software.
3. Manual recruiting leads them to automate the candidate side, then the company side, forming a marketplace.