Walmart tightens AI access as token costs trigger budget discipline

The transition reflects a broader shift in how large enterprises handle the economics of generative AI. While companies initially encouraged staff to experiment with tools for spreadsheet analysis and presentation design, the reality of per-token billing has forced a rethink. With a workforce of 2.1 million, even minor increases in query frequency translate into significant operational expenses. Walmart’s new guidance now directs staff to select specific AI models based on task complexity, hoping to prevent the use of expensive, high-reasoning models for trivial administrative chores.

This policy change signals the end of the "token maxxing" trend, where firms once celebrated employees for maximizing AI usage to hit productivity targets. That approach often led to the gamification of internal metrics, creating hidden costs as employees ran iterative, multi-agent loops to refine simple outputs. Now that these interactions are billed directly, companies like Walmart are discovering that AI productivity is only valuable if it remains cheaper than the human labor it replaces.

Walmart is not alone in this reckoning. As major providers like OpenAI and Anthropic pivot away from flat-rate subscriptions toward usage-based pricing, other corporations are feeling the squeeze. Uber recently disclosed that it exhausted its entire 2026 AI budget within the first four months of this year. By imposing strict token limits, Walmart is attempting to curb this "financial normality," forcing employees to consider the tangible cost of every automated request.