The Army's AI Token Pool Depleted Months After Announcing 'Unlimited' Access

The Army’s Token Shortage Reveals Budget-Reality Mismatch

The U.S. Army’s Combat Capabilities Development Command exhausted its entire annual AI token allocation by mid-June 2026, just one month after the Department of Defense announced that nearly half of its 3.5 million employees were actively using generative AI at work. According to Wired AI, an internal Army email informed DEVCOM staff that despite a May 2026 announcement of unlimited token access, the Army CIO’s token pool had been depleted and usage caps were being reinstated. The Army had purchased 100 million tokens annually as part of an enterprise subscription to Ask Sage, a multimodal generative AI platform.

The speed of depletion—roughly 45 days to exhaust a year’s budget—underscores a stark disconnect between the Pentagon’s public enthusiasm for AI adoption and its inability to fund sustained enterprise-scale usage. An anonymous Army employee told Wired AI that “the whole Army burned through the whole year of tokens for just one service,” suggesting the depletion occurred faster than internal budgeting models anticipated.

Ask Sage’s Role in DOD AI Operations

Ask Sage functions as the Defense Department’s primary enterprise LLM workspace for unclassified and controlled-information tasks. According to Wired AI, the platform integrates models from Alphabet, Meta, and OpenAI—specifically Gemini, Llama, and ChatGPT—and is used by both Army personnel and the DOD’s Chief Digital and AI Office (CDAO) for acquisitions and personnel management. The platform has been accredited to handle Controlled Unclassified Information, expanding its utility across routine administrative functions.

Army employees received an initial monthly allotment of at least 200,000 tokens per person, with automatic top-ups available if individuals exceeded their quotas. This tiered approach was designed to encourage adoption without hard caps, but the absence of organizational-level spending controls appears to have contributed to rapid depletion once usage scaled across the 3.5-million-person DOD workforce.

Uncertainty Over Future Funding

The Army’s internal email notes that while the organization “has chosen to renew token usage at its current levels,” the renewal timeline remains uncertain—specifically whether the AI CIO pool will be funded beyond October 1, 2026. This ambiguity creates operational risk: teams relying on Ask Sage for personnel reclassification, job-duty alignment, and other administrative processes face a potential cliff if no budget is allocated for the next fiscal cycle.

The Army and Department of Defense did not respond to Wired AI’s requests for comment on the token exhaustion, the depletion timeline, or renewal plans.

Why This Matters

The Army’s token crisis exposes a critical gap in federal AI procurement: enthusiasm for adoption can outpace budget modeling. Teams now relying on generative AI for routine administrative work have a three-month window to either optimize token efficiency, migrate to on-premises models, or prepare contingency workflows if funding lapses. For vendors like OpenAI, Meta, and Google, the episode demonstrates that enterprise government contracts require guardrails against runaway consumption—and that “unlimited” announcements need accompanying budget authority or face rapid credibility damage. The Pentagon’s broader AI-adoption initiative also faces an implicit question: if a single internal tool exhausted a year’s allocation in 1.5 months, do existing budget lines account for the true cost of enterprise AI rollout across the 3.5-million-person workforce?