Rising costs of the token consumption race

Companies are increasingly tracking employee AI usage through “tokenmaxxing,” with examples like Databricks’ engineer spending over $7,000 in tokens and Harvey’s legal-tech startup rising to 12 trillion tokens monthly. The surge has led firms such as Uber, Meta, and Microsoft to impose caps and policies after exhausting AI budgets within months, while model choice—top-tier models costing five to ten times more per million tokens than optimized ones—emerges as a key cost-control lever.