US Army reinstates AI limits after troops burn through year's token supply in weeks

US Army Depletes Annual AI Token Supply in Weeks

The US Army has been forced to reinstate limits on AI token usage after exhausting its annual supply of tokens barely a month into what was promised as an unlimited access program 1

1

. In May 2026, the Army CIO announced unlimited tokens for its workforce, but by mid-June, internal emails to members of the Army's Combat Capabilities Development Command (DEVCOM) confirmed the token pool had run dry 1

1

. The development highlights the challenges with generative AI token consumption that even well-funded government agencies face when adopting AI at scale.

According to documents reviewed by WIRED, the Army had access to 100,000,000 tokens as part of an annual subscription to an enterprise pack for Ask Sage, a multimodal generative AI platform that powers the Army's enterprise LLM workspace 1

1

. Ask Sage allows users to run different large language models including Alphabet's Gemini, Meta's Llama, and OpenAI's ChatGPT 1

1

. The platform is accredited for Controlled Unclassified Information and has been deployed across the Department of Defense's Chief Digital and AI Office for acquisitions [1](https://arstechnica.com/ai/2026/07/us-army-faces-ai-use-limits-after-exhausting-years- Graus-supply-of-ai-tokens/).

Aggressive Push for AI Adoption Backfires

The rapid depletion stems from the Army's aggressive push for AI adoption among its workforce. Employees received an allotment of at least 200,000 tokens per month and were automatically allocated more if they burned through their initial allocation 1

1

. Workers who signed up for Ask Sage but weren't regularly using it even received emails encouraging them to use more of their allocated tokens 1

1

. "Apparently the whole Army burned through the whole year of tokens for just one service," an Army employee told WIRED 1

1

.

For context, a single token on Ask Sage equates to about 3.7 characters 1

1

. The scale of military AI consumption becomes clearer when examining Operation Epic Fury in Iran, where the Defense Department burned through approximately 20 billion tokens per day during the 38-day operation, according to Breaking Defense 1

1

.

Unpredictable Consumption-Based Pricing Models Challenge Budgeting

The US Army's experience reflects broader industry struggles with unpredictable consumption-based pricing models for AI services 2

2

. After the Army chose to reinstate limits on AI token usage at current levels, uncertainty looms over whether the Army CIO pool will be renewed after October 1, 2026 1

1

. This case demonstrates how aggressive AI pushes might not always consider cost impacts, especially as many AI vendors shift from flat per-seat subscriptions to output-based models 2

2

.

The Pentagon continues to lean heavily into AI tools despite these budgeting challenges. The Intercept reported that the Pentagon has cut staff at the Civilian Protection Center of Excellence while developing AI tools to speed up assessments that human staff would normally make 1

1

. The Army isn't alone in rethinking unlimited AI access—Meta quietly removed its leaderboard tracking token usage after encouraging employees to "tokenmaxx," and Uber engineers burned through a year's worth of generative AI tokens in just four months 1

1

.

One Army employee questioned the effectiveness of the rollout, noting that generative AI tools have proven unreliable for their work, with one model falsely claiming it had completed a task 1

1

. "I think there are definitely several aspects of the bureaucracy of the US federal government that these tools might be helpful with. But an unthinking application and use is not going to result in an effective, efficient, and trustworthy rollout," they said 1

1

.