The Economics of Superstar AI Researchers

The superstar effect explains the pay gap among AI researchers. Even when the difference between the gold and silver medalist in a 100m race is tiny, the rewards diverge sharply — and the same logic applies to AI.

Two key conditions drive this.

• One person's work must reach a massive market.
• The quantity of many people must not easily substitute for the quality of one.

AI satisfies both conditions. Research results can be reflected simultaneously in a product used by hundreds of millions of people, like ChatGPT, and frontier labs face compute constraints that make it hard to scale up the number of experiments infinitely to replace top talent. As a result, even a very small difference in ability can translate into a huge difference in value, potentially creating pay gaps of 10x or even 100x.

The closer competition gets to 'all in or nothing,' the wider the gap grows. In a winner-take-all landscape where trillions of dollars in rewards are at stake, ultra-high recruitment battles can emerge — such as Meta's (reported) $100 million-level package.

However, a pay gap does not necessarily mean a gap in capability. Factors such as a researcher's accumulated private experimental knowledge, their management role in leading a team, and the shift in duties for top-tier researchers who move closer to being 'managers' — like Noam Brown — can all be reflected in compensation, and benchmarks such as RE-bench still cannot fully measure the value of actual large-scale research projects. As AI usage spreads more widely and deeply going forward, the superstar effect is likely to grow even stronger in research environments that increasingly resemble 'managing an army of Claudes.'