Members of the US Army's Combat Capabilities Development Command (DEVCOM) received an email warning that they were rapidly depleting their allotment of AI tokens and needed to limit their use of the service, according to a report from WIRED. The message reveals that a pool of computing capacity the Army thought was effectively boundless has run dry far faster than planned.
June 2026 · US Army DEVCOM
The Army Is Burning Through Its AI Tokens
A pool of AI computing capacity the Army thought was effectively boundless ran dry far faster than planned — forcing officials to reimpose caps just weeks after promising "unlimited" access.
~4 wks
From "unlimited" (May 2026) to pool exhausted (mid-June 2026)
1.75M
DoD employees using AI — roughly half of ~3.5M staff
$49M
Ask Sage contract cap over 5 years (IDIQ)
DAILY CONSUMPTION vs. THE TRIAL ALLOWANCE
During a 38-day operation, the DoD burned an estimated 20 billion tokens per day — dwarfing the 200,000-token monthly starter allotment that first onboarded users.
200K
Monthly trial allotment
20,000,000K
Per day, in a 38-day operation
Scales are not comparable — one is a monthly cap, the other a single day's operational burn.
THE "TOKENMAXXING" PATTERN
Unlimited launch
Access opened wide, usage unchecked
→
Runaway burn
Shared pool drained faster than planned
→
Caps reimposed
Cost discipline & metered allocations return
Echoing industry: Meta retired usage leaderboards and weighed limits; Uber reportedly spent an annual budget in four months.
WHERE IT HELPS
May assist with routine bureaucratic tasks; secure access to Gemini, Llama and ChatGPT cleared for Controlled Unclassified Information.
WHERE SKEPTICS PUSH BACK
Users call it "not especially useful" and at times unreliable — one model claimed to finish a task it hadn't. Indiscriminate rollout ≠ effective use.
THE OCTOBER 1 QUESTION
Token access was renewed at current levels, but renewal beyond October 1 is uncertain. Because the platform is billed per token, more headroom means more procurement — making this a budgeting and governance challenge that other institutions racing to adopt generative AI will likely face next.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…