Mercor reportedly spends 3 times as much on AI inference as on employee salaries

Brendan Foody says Mercor spends three times as much on LLM inference—running language models—as on employee salaries, and that the returns support headcount rather than replace it. In his September 13, 2026 projection, he predicts companies will spend as much on inference within five years as they currently spend compensating knowledge workers, which he estimates at roughly $40 trillion annually. If wages remain roughly constant and companies absorb that spending profitably, he argues, the economy would need about $40 trillion in additional annual final economic output. Producing that much more annual GDP in year five than a roughly 3% growth baseline would imply about 9% average annual GDP growth over those five years, he says.