Uber imposes $1,500 monthly AI cap after budget burn

Uber exhausted its annual AI budget within four months and has set a $1,500 monthly cap on AI spending per employee. The cap covers tools such as Claude Code and Cursor and operates through a token-based system; increases require explicit company approval.

The company’s chief technology officer disclosed heavy internal use drained the year’s allocation in a single quarter. Management encouraged experimentation and ran an internal leaderboard to rank consumption; the company now records individual spend on an internal dashboard.

Charges for large language models and other generative tools typically scale with tokens or model calls. Some employees used multiple coding and productivity models and cost exposure grew as usage spread across teams. The cap is applied per employee and varies by the set of tools accessed; managers can request exceptions for specific projects or heavy users.

A trend called tokenmaxxing measures AI use by counting tokens consumed. Several companies have introduced leaderboards and incentives tied to token counts. Proponents treat token volume as a sign of adoption; critics say higher token totals do not necessarily reflect useful outcomes.

Martin Reynolds, field chief technology officer at Harness, described reliance on consumption metrics as ‘treating the system rather than the cause’ and warned that usage metrics can reward generating more prompts, tokens and model interactions regardless of business value.

Surveys of executives report growing frustration about unclear financial returns from AI projects, and some firms maintain heavy AI spending in part from concern about falling behind competitors. Reynolds recommended improving cost visibility and linking AI usage to specific business outcomes such as customer experience or revenue impact.

Reynolds warned that without measuring value rather than spend, organizations could alternate between uncapped expenses and strict rationing, leaving the financial effect of AI use unclear.