More AI Token Consumption = Better Results? Study Says High Usage Doesn't Guarantee Higher Output

The tech industry is entering a new phase of rational AI spending. Data from enterprise management platform Jellyfish shows that companies consuming the most AI tokens aren't necessarily the winners.

A study found that the top 10% of Claude Code users consume 10 times more tokens than the median ordinary developer, but their output is only twice as much. AI tokens are basic units for model operations and billing basis. Jellyfish's director notes that uncontrolled token consumption isn't a sustainable strategy, and CFOs have started strictly controlling such expenses.

Experts suggest enterprises focus on outcome-based metrics (like cost per code merge request) instead of total token count. The optimal path is to let more engineers enter a moderate and reasonable usage range, improving efficiency while avoiding unnecessary cost waste.