When Uber set its 2026 AI budget, the company presumably employed people who could do arithmetic. Yet by late May, employees had consumed the entire year's allocation of AI tools and services—leaving finance scrambling to impose caps that will govern the remaining eight months. The episode is less about one company's forecasting failure than about a structural reality now confronting every large enterprise: AI usage, once unlocked, grows faster than anyone anticipates.
The details, first reported by TechCrunch, reveal a familiar pattern. Uber gave employees access to a suite of AI coding assistants, productivity tools, and internal model APIs. Usage was encouraged. Dashboards were built. And then the dashboards turned red, months ahead of schedule.
The adoption curve nobody modeled
Corporate AI spending has a peculiar quality: it compounds. An engineer who uses a coding assistant for one task discovers it handles three others. A product manager who queries an internal LLM for competitive analysis starts using it for meeting prep, then for drafting specs, then for everything. Multiply this by tens of thousands of employees, and even generous budgets evaporate.
Uber is not alone. Microsoft's own internal data suggests that enterprise Copilot usage typically exceeds initial projections by 40-60% within six months of deployment. But Uber's case is notable for its velocity—a complete budget exhaustion in a single quarter—and for the company's willingness to impose hard caps rather than simply expand the envelope.
What the caps reveal
The new restrictions reportedly include per-employee monthly limits on API calls, tiered access based on role, and approval workflows for high-consumption use cases. This is, in effect, AI rationing—a concept that would have seemed absurd two years ago when enterprises were begging employees to adopt these tools.
The shift matters because it signals a transition from the "experimentation" phase of enterprise AI to something more closely resembling utility management. Electricity is not unlimited; neither, it turns out, is AI. Companies that treated these tools as essentially free productivity boosters are now discovering they have created new cost centers that behave like usage-based cloud bills: unpredictable, fast-growing, and politically difficult to constrain once employees have built workflows around them.
Our take
Uber's budget blowout is the kind of problem most enterprises would kill to have. It means adoption worked—spectacularly, in fact. The harder question is whether the productivity gains justify the spend, and on that point, the data remains thin. What's clear is that the AI vendor ecosystem—OpenAI, Anthropic, Google, and the tooling layer built atop them—now has proof that enterprise demand is not merely robust but insatiable. Pricing power, for those who hold it, just got a lot more interesting.
