Microsoft has reportedly begun canceling most of its direct licenses for Claude Code, according to The Verge, as the company shifts employees toward using GitHub Copilot CLI instead.
The move comes only six months after Microsoft expanded internal access to Anthropic’s AI coding tool across engineering, product, and design teams.
Claude Code reportedly saw rapid adoption inside Microsoft as employees increasingly used AI agents for software development and workflow automation tasks.
AI Adoption Creates New Cost Pressures
The reported rollback highlights a growing challenge facing enterprise AI deployments: rising operational costs tied to heavy usage of agentic systems.
While AI model pricing per token continues to decline, total enterprise spending is climbing as developers run increasingly complex workflows involving autonomous coding loops, tool calls, debugging chains, and multi-agent orchestration.
According to the report, Microsoft’s licensing pullback does not affect its broader relationship with Anthropic, which includes Azure infrastructure partnerships and access to Claude models through Microsoft Foundry.
Anthropic has also committed to purchasing billions of dollars in Azure compute capacity over the coming years.
Other Tech Companies Face Similar AI Spending Surges
Microsoft is not alone in confronting rapidly expanding AI infrastructure costs.
Uber CTO Praveen Neppalli Naga previously said the company exhausted its 2026 AI coding tools budget within the first four months of the year after aggressively encouraging internal adoption.
At Meta, employees reportedly created internal AI usage leaderboards nicknamed “Claudeonomics” to track AI tool adoption across teams.
Amazon has similarly encouraged employees to maximize AI token usage internally.
Cheaper Tokens Are Not Reducing Enterprise Costs
Analysts increasingly warn that lower token pricing may not translate into lower enterprise AI spending. Research firm Gartner recently projected that inference costs for frontier AI models could fall sharply by 2030, but noted that agentic systems require dramatically higher token consumption per task.
Meanwhile, Goldman Sachs forecasted that enterprise adoption of AI agents could drive a 24-fold increase in monthly token usage by the end of the decade.
As companies deploy more persistent AI agents, aggregate infrastructure costs could continue rising even as individual model queries become cheaper.
Agentic AI Economics Face Growing Scrutiny
The spending surge is beginning to raise broader questions about the economics of large-scale AI workforce automation. Nvidia executive Bryan Catanzaro recently said compute costs for his teams already exceed employee costs.
At the same time, Nvidia CEO Jensen Huang has predicted that future enterprises could operate with hundreds of AI agents working alongside each employee.
Those visions increasingly depend not only on model capability improvements, but also on whether enterprises can sustainably manage the infrastructure and inference costs required to run always-on autonomous systems at scale.
