The US Army burned through a year's supply of AI tokens: a lesson for any business with an AI budget

The US Army ran into an unexpected constraint: its annual supply of AI tokens — the units in which large-model usage is billed — was spent long before the year ended. The military had to impose AI usage limits while additional quota is procured.

The story is instructive not for its scale but for how typical it is. An organization rolled AI out broadly, staff got a taste for it — and consumption grew multiples faster than forecast. Unlike per-user licenses, tokens burn in proportion to intensity: long documents, big exports, "redo it again" loops — all of it multiplies the bill.

The takeaways apply to a company of any size. Track consumption by department and scenario — without measurement you can neither budget nor find waste. Set limits and alerts, as you would for cloud infrastructure. And remember that expensive scenarios often come from cheap carelessness — like the habit of sending a model an entire document when one section is needed. AI is now a line item in operating costs, and it deserves to be managed as soberly as telecoms or electricity.