TokenBudgeting: Our Conversations with Enterprises on Token Spend

Key Points:

• Enterprise AI token consumption surged early in 2026 due to "tokenmaxxing," where companies like Meta encouraged high token use to boost productivity, but this led to unsustainable spending and subsequent budget caps.
• Most enterprises have now shifted to token budgeting with monthly AI usage limits varying widely from $250 to several thousand dollars per employee, reflecting different company policies and roles.
• Despite media reports of excessive token spending, data shows that only a small percentage of high-spending customers drive most revenue, while the majority of enterprises spend modestly on AI tools, primarily in engineering and data science.
• Employees are adapting to budget limits by downgrading AI models, turning off premium features, and using free tools like Microsoft 365 Copilot to conserve tokens and stretch their allowances.
• AI spending is increasingly viewed as a productivity lever tied to project revenue, with companies balancing token budgets against expected output, indicating budgeting and controlled token use will remain standard in enterprise AI adoption.