Microsoft Abandons Claude Code: AI Too Expensive

Microsoft has made the radical decision to stop using Claude Code among its engineers, not due to technical disappointments, but because of costs deemed exorbitant. Introduced six months ago, Claude Code proved so effective that its intensive use led to skyrocketing expenses.

Last December, Microsoft opened access to Claude Code to thousands of its employees, including engineers and project managers, as part of a large-scale test of AI coding tools. However, this initiative raised questions, as Microsoft was paying a direct competitor to equip its own staff, despite having its own AI models.

The Experiences & Devices division, responsible for products such as Windows, Microsoft 365, Outlook, Teams, and Surface, will have to abandon Claude Code by June 30, 2026. By that date, teams will transition to GitHub Copilot CLI, the internal tool that Microsoft aims to standardize to unify its tools. Rajesh Jha, vice president of the relevant unit, explained that this transition aims to unify tools and allow Microsoft to shape a product tailored to its own repositories, workflows, and security requirements.

Prohibitive Costs of Claude Code

The main issue stems from the billing model of the agentic AI, which is based on token consumption, proving to be costly. Unlike traditional subscriptions, AI tools like Claude Code charge per use, with each token representing a unit of processing by the model.

A code agent like Claude Code performs complex tasks, analyzing entire projects and generating code, which can consume a massive amount of tokens. Developers using the tool intensively can incur monthly costs ranging from $500 to $2,000 each.

An Unexpected Impact on the Industry

This situation is not isolated. In April, Uber announced that it had exhausted its AI coding budget planned for 2026 in just four months, with the adoption of Claude Code rising from 32% to 84% among its 5,000 engineers. Approximately 70% of the code produced at Uber is now generated by AI. No one had anticipated such rapid adoption or the volumes of tokens that would result. Gartner predicts that 25% of AI budgets for 2026 will be pushed to 2027 due to insufficient return on investment.