Uber Caps Employee AI Spending at $1,500/Month After Budget Collapses in Four Months
Uber has imposed a hard $1,500/month per-employee cap on AI tool spending, confirmed by Bloomberg, the Los Angeles Times, and TechCrunch on June 3. The policy applies company-wide and covers Claude Code, Codex, and related AI coding systems that engineers have been using to automate development work.
The backstory: Uber burned through its full 2026 AI budget before April ended — less than four months into the fiscal year. Claude Code costs individual engineers $500 to $2,000 per month at typical usage intensity. The company’s COO had already flagged the spend publicly, questioning ROI. The $1,500 cap is the formal policy response.
The Arithmetic
At $1,500 per seat, the math clarifies quickly. Uber employs roughly 7,000 engineers. Full deployment at the cap: $10.5M/month, $126M annually — for AI coding tools alone, before any other AI spend. That’s before accounting for infrastructure, API costs for non-engineering use cases, or the broader AI stack the company runs across its platform.
The cap sits at the midpoint of Claude Code’s observed per-user cost range, which means it constrains heavy users without fully blocking moderate adoption. Engineers spending $500-$1,000/month are unaffected. Engineers at $2,000 need to cut usage by 25%.
What Simon Willison Said
Simon Willison, writing on simonwillison.net, called the $1,500 number “a useful signal for AI tool pricing.” His argument: the cap implicitly sets a market rate. Tool vendors now have a benchmark for what enterprise buyers will sustain per seat. Products priced above $1,500/month all-in face structural procurement resistance at major employers.
That framing matters for the competitive landscape. OpenAI’s Codex Pro plan runs $100/month — well inside any enterprise ceiling. Anthropic’s Claude Code at the frontier, depending on actual usage, regularly exceeds $1,500 for power users. The gap between flat-rate plans and consumption-based usage is where enterprises are now setting limits.
The Pattern Is Spreading
Uber is the first major tech company to formalize a per-employee AI cap with a publicly known number. It will not be the last. Several large organizations have already cut AI tool access, frozen subscriptions, or opened internal audits of AI spend. Cloudflare cut 1,100 jobs and cited agentic AI on the same day it beat earnings. Microsoft is canceling blocks of Claude Code licenses.
The first wave of enterprise AI adoption was permissive: give engineers access, measure later. The second wave has a number attached to it.
For Anthropic and OpenAI, the question now is whether enterprise accounts can be structured to fit inside those numbers at scale, or whether the high-end usage that drives their revenue figures is concentrated in a small enough subset of users that broad enterprise caps don’t materially affect it. The answer will show up in ARR figures before the end of the year.