Major companies including Meta, Amazon, Uber and Walmart have reversed course and begun limiting employees' use of AI tools, after costs far exceeded expectations — an abrupt shift from a stance that, just months earlier, urged staff to use AI as much as possible.
June 2026 · AI Cost Reckoning
From "Tokenmaxxing" to Damage Control
Meta, Uber, Walmart and Amazon spent early 2026 urging staff to use AI as much as possible. Months later, "exponential cost increases" forced them to slam on the brakes — caps, removed leaderboards, and blown budgets.
4 mo
Time Uber took to burn its entire 2026 AI budget
~18 mo
Span over which the message flipped from "useful" to "contain costs"
−90%
Possible cost cut from switching to cheaper models on simple tasks
Uber Internal AI Adoption
A surge that outran the budget
Adoption more than doubled in a single month — and the full-year budget was gone by spring.
What a Task Costs
Why bills exploded: tokens per task
~hundreds
Summarize a
tens of thousands
Complex code &
Long-running AI agents and code generation can burn 100× the tokens of a simple query — pushing one engineer to tens of thousands of dollars a month.
The Crackdown
Uber
Burned 2026 budget in 4 months · $1,500/person monthly cap on coding tools
Meta
Notified usage limits · prioritized internal tools · removed leaderboard
Walmart
Introduced token caps on internal AI agents
Amazon
Removed leaderboard · introduced internal caps
THE EFFICIENCY CASE
Treating high token usage as a sign of competence was the mistake. Reserve cutting-edge models for complex work and "tokenminning" can cut costs by up to 90%. Output metrics like "agentic work units" should replace raw token counts.
THE UNEASE
The pendulum swung
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…