Rising AI token usage has led companies to ration access to advanced models as costs for automated agents climb.
Key facts
- •AI usage is measured in tokens, with one token representing approximately three-quarters of an English word.
- •Companies have used trillions to quadrillions of tokens in recent months for automated tasks.
- •High costs associated with agentic AI have led many companies to begin rationing model access.
- •Some businesses are moving toward cheaper, freely available AI models of Chinese origin.
- •The cost of virtual workers may exceed that of human employees depending on the specific task.
Major companies are increasingly rationing the use of advanced AI models following a surge in consumption during 2026. While the cost per token has declined, the massive volume of tokens used for automated tasks has resulted in significant expenses for businesses.
Token Usage and Productivity
AI systems process information in units called tokens, with one token roughly equaling three-quarters of an English word. To maximize productivity, many top companies implemented "token leaderboards" to track employee usage of advanced models.
Automation Costs and Market Shifts
Usage has reached trillions and sometimes quadrillions of tokens in recent months, primarily driven by "agentic use," where AI agents perform tasks automatically. Due to the resulting high costs, some companies are now rationing access to these models. Additionally, some firms are shifting toward cheaper AI models derived from Chinese sources that are available for free on the market.
Advertisement
This article was independently rewritten by ManyPress editorial AI from reporting originally published by BBC Technology.
