Microsoft Slows Down on AI: Claude Code Considered Too Expensive

Microsoft has recently begun to cancel most of its direct licenses for Claude Code, an artificial intelligence tool developed by Anthropic. This decision, reported by The Verge, is driven by the realization that certain uses of AI agents are now more expensive than human labor. This measure comes as the Experiences + Devices division seeks to reduce its operational expenses before the end of the fiscal year on June 30.

The Experiences + Devices division, which includes teams responsible for Windows, Microsoft 365, Outlook, Microsoft Teams, and Surface, has decided to gradually decrease the use of Claude Code. The cancellation of these licenses would allow Microsoft to quickly cut certain expenses ahead of the new fiscal year opening in July.

Microsoft initially opened access to Claude Code in late 2025, inviting several thousand of its developers to use it daily. However, the intensive use of this tool revealed that, in some scenarios, the costs associated with AI agents exceeded those of human employees performing the same tasks. Consequently, Microsoft is now encouraging its engineers to use the command-line interface of GitHub Copilot, although the source does not explicitly state that this shift is motivated by economic reasons.

This strategic realignment does not undermine the major agreements between Microsoft and Anthropic, including the Foundry agreement, which outlines a potential investment of $5 billion. However, it illustrates the financial challenges companies face in the widespread adoption of AI.

Microsoft is not alone in this situation. Uber, for example, exhausted its annual budget for AI tools in just four months in 2026. These examples show that the cost of adopting AI remains a significant barrier, despite the expected decrease in unit token prices as projected by Gartner by 2030.

The Claude Code episode highlights the fact that companies must now consider not only the technological advancements of AI but also the accounting implications of its large-scale deployment.