Microsoft has cancelled most internal Claude Code licences for its Experiences and Devices division — the team behind Windows, Microsoft 365, Outlook, Teams, and Surface — after a token-billing explosion consumed the division’s entire annual AI budget well ahead of schedule.

An internal memo from Executive Vice President Rajesh Jha directed engineers to stop using Anthropic’s AI coding assistant and migrate to GitHub Copilot CLI instead. The cutover deadline is June 30, 2026.

What Happened

Microsoft rolled out Claude Code to approximately 5,000 engineers in December 2025. Adoption was fast and sticky: usage rates climbed to between 84 and 95 percent of the engineering cohort by April 2026. Engineers loved the tool. That was the problem.

Because the pilot launched under flat seat licensing, the full cost of token consumption was invisible during the early months. When Microsoft shifted to usage-based billing, the true numbers became immediately and painfully visible. Per-engineer API costs were running between $500 and $2,000 per month — well beyond what flat-rate budget modelling had assumed.

The result: an annual AI tools budget exhausted in a matter of months.

Microsoft is not alone. Uber’s Chief Technology Officer publicly confirmed that the company burned through its planned 2026 AI coding budget in four months, with the full annual allocation spent before the year was half over.

The Budget Modelling Gap

The core issue is structural. Enterprises are accustomed to software procurement models where the price of a seat is the price of a seat. When a thousand engineers hold a licence, the cost is predictable.

AI coding assistants break that model entirely. Usage-based billing means a highly engaged engineering team — exactly what you want when deploying a productivity tool — can generate token costs an order of magnitude higher than seat-count projections suggest.

Microsoft’s situation illustrates this clearly. A 90 percent adoption rate sounds like a success story. In token-billing terms, it translated to a budget overrun that forced a retreat.

What Microsoft Is Doing Instead

Engineers are being transitioned to GitHub Copilot CLI, Microsoft’s own command-line AI coding tool. The move consolidates AI coding spend under a product Microsoft controls and owns commercially, reducing exposure to third-party token pricing.

There is an obvious strategic dimension here: Microsoft holds a multi-billion dollar stake in OpenAI and is deeply invested in the Copilot product line. Directing engineers toward GitHub Copilot CLI serves both cost discipline and competitive alignment.

What This Means for Business

The Microsoft story carries lessons well beyond big tech.

Flat-rate thinking does not survive contact with usage-based AI. Any business evaluating AI tools priced on consumption — whether coding assistants, voice agents, or LLM APIs — needs to model usage scenarios at the high end, not the average. A tool that 90 percent of your team uses every day will cost far more than a tool only half your team uses occasionally.

Popularity is not the same as ROI. A high adoption rate is a promising signal, but it is not a substitute for measuring actual output improvements. Microsoft’s engineers loved Claude Code. Whether the productivity gains justified $500 to $2,000 per engineer per month is a separate, harder question — one the company ultimately decided did not justify continued spend.

Lock-in to your own stack has real value. Microsoft’s exit strategy — GitHub Copilot CLI — was available because the company had built its own AI coding tool. Businesses that rely entirely on third-party AI vendors have fewer options when costs become unmanageable.

Governance needs to come before adoption, not after. The pattern here — deploy widely, discover costs later — is increasingly common as AI tool adoption accelerates. Enterprises that establish usage monitoring, per-team budget alerts, and cost-per-output metrics before they roll out AI tools will avoid the scenario Microsoft found itself in.

The Bigger Picture

The Microsoft and Uber situations mark a turning point in enterprise AI adoption. The first wave was about proving what AI tools could do. The second wave — the one we are now entering — is about proving the economics.

That is a healthier phase for the industry, even if it feels like a setback. Tools that deliver real productivity gains at defensible cost will survive this scrutiny. Tools that are impressive to use but impossible to budget for will face exactly this kind of retreat.

For businesses currently evaluating AI tools — whether coding assistants, AI agents, voice systems, or custom applications — the Microsoft story is a useful reference point. Get the usage economics clear before you deploy widely, not after the budget has already run out.

---

Thinking about deploying AI across your business? Enterprise DNA’s Omni Advisory service helps business leaders build AI roadmaps that account for real-world cost modelling, not just headline capability. Book a discovery call to talk through your situation.