Key facts
- Mercor's CEO, Brendan Foody, stated the company spends more on AI tokens than employee salaries.
- Mercor is a startup that helps companies train AI models.
- The company uses AI agents for functions like project management, recruiting, and candidate evaluation.
- Foody predicts that within five years, the average enterprise will spend more on compute than on headcount.
- This comes amid broader debate about AI spending translating into business returns.
Brendan Foody, CEO of Mercor, a startup valued at $10 billion that assists companies like OpenAI and Anthropic in training AI models, has revealed that the company's expenditure on AI tokens for its internal agents now surpasses its spending on employee salaries. Foody made these remarks during an appearance on the "20VC" podcast. He indicated that Mercor utilizes AI agents across various operational functions, including project management, recruitment, accounting, fraud detection, and candidate evaluation, having conducted over 5 million AI-assisted interviews. Foody anticipates that this trend of AI costs escalating beyond labor expenses will become a widespread phenomenon within the next five years, suggesting that the average enterprise will eventually spend more on compute power than on its human workforce. His comments emerge at a time when executives are debating the tangible business returns generated by increasing investments in artificial intelligence. For instance, Uber COO Andrew Macdonald recently expressed that he has not yet observed a clear correlation between escalating AI expenditures and proportional productivity improvements. Foody attributes this to a Jevons paradox-like effect, where the decreasing costs and rapidly advancing capabilities of AI lead to significantly higher consumption. He believes that the cost of AI inference and compute will ultimately exceed the cost of human labor, although humans will retain importance in tasks that AI cannot perform.