Microsoft Drops Most Internal Claude Code Licenses. The Enterprise AI Cost Crisis Is Real.

Here's the counterweight to the PMF story. Microsoft began revoking internal Claude Code licenses for most employees on May 14 with a June 30 deadline. The Experiences and Devices division, the team behind Windows, Microsoft 365, Outlook, Teams, and Surface, is moving engineers to GitHub Copilot CLI instead.

The cost data is staggering. Uber burned through its entire $3.4B 2026 AI budget in four months after Claude Code usage jumped from 32% to 84% of its 5,000-engineer org. Individual engineers were spending $500-$2,000/month in tokens. AI software prices across the US climbed 20-37% this year. FinOps teams managing AI spend doubled from 31% to 63% of enterprises in one year.

This is the part of the coding agent gold rush nobody wants to talk about. Usage-based pricing means your costs are a function of how much your engineers love the tool. And engineers really love these tools. That's the whole problem. Microsoft isn't cutting Claude Code because it doesn't work. They're cutting it because it works too well and the bill got out of control.

I've been watching enterprise AI procurement closely, and there's a pattern forming. Phase 1: pilot with 50 engineers, costs look fine. Phase 2: roll out to 5,000 engineers, costs explode. Phase 3: panic, pull licenses, redirect to cheaper alternatives. Microsoft is in phase 3. Uber hit phase 3 in four months.

GitHub is moving all Copilot plans to usage-based billing via AI Credits starting June 1. So Microsoft's own alternative is heading toward the same pricing model that caused the problem. I don't think anyone has figured out the right pricing model yet. Flat-rate subsidizes heavy users. Usage-based creates budget chaos. Something in between, probably with hard caps and tiered pricing, is where this lands.

What builders should do: If you're deploying AI coding tools at scale, build a cost dashboard before you build the rollout plan. Track per-engineer, per-project, and per-task spend. Set token budgets with hard caps. The tool vendors won't do this for you, so build it yourself or check out CodeBurn, which tracks spending across 25+ AI coding tools.