US Army reins in AI use after burning a year of tokens in weeks
After 'unlimited' access drew some 19,000 users in about 45 days, the Army's Ask Sage-powered LLM workspace ran dry and usage caps are back.
The US Army has reinstated usage limits on its enterprise generative-AI platform after soldiers and civilian staff exhausted the service's annual token allotment within weeks, according to reporting by Wired picked up by Ars Technica on July 22. Members of the Army Combat Capabilities Development Command were informed by email that the pool of tokens funded by the Army CIO had run out and that caps were returning.
Unlimited, briefly
The Army Enterprise LLM Workspace runs on Ask Sage, a brokerage platform that gives government users access to commercial frontier models — including Google's Gemini, Meta's Llama, and OpenAI's models — under a five-year contract worth roughly $49 million. In May the Army CIO announced unlimited access for the workforce, and adoption exploded: roughly 19,000 users came aboard in about 45 days. By mid-June, consumption had blown through the annual allocation, forcing the service to walk back the unlimited promise and restore per-user limits while it renegotiates capacity.
The episode is an unusually clean natural experiment in AI demand. Given a free hand, a large workforce did not treat generative AI as a novelty; it consumed a year of budgeted inference in roughly a month and a half.
Why it matters: token economics do not behave like the seat licenses that enterprise IT budgets were built around, and the Army just demonstrated the failure mode at institutional scale. For the Pentagon, the incident is simultaneously an embarrassment and a validation — the demand signal for AI tools inside government is real and enormous, but procurement models lag far behind usage-based pricing. Every large organization rolling out 'unlimited' internal AI access is now on notice: measure consumption first, promise second. Vendors, meanwhile, will read this as evidence that government AI contracts are about to get much bigger.