Army AI demand runs into token limits after a usage surge

The U.S. Army’s push to make generative AI part of daily work has already run into a practical constraint: token supply. WIRED reports that members of the Army’s Combat Capabilities Development Command were told to limit Ask Sage use after an Army CIO token pool was exhausted by mid-June.

Ask Sage is a multimodal AI platform that can route users to models including Gemini, Llama, and ChatGPT. According to emails viewed by WIRED, employees had monthly token allotments and were encouraged to use the system if they had access but were not active.

The report says the Army renewed usage at current levels, but it was unclear whether the pool would be renewed after October 1. A token is a unit of model input or output, so heavy use can turn enthusiasm into infrastructure and budget pressure quickly.

The episode does not show that the Defense Department is backing away from AI. It shows the opposite problem: when millions of workers are encouraged to use AI, capacity planning becomes a central part of deployment.