Even Microsoft reconsidered? Why a popular AI tool was replaced
A tool can be popular with users and still face review over its pricing. Lessons from The Verge's reporting
Key takeaway
In May 2026, Microsoft reportedly told divisions responsible for products including Windows and Office to end internal use of Claude Code and move to GitHub Copilot CLI. The stated rationale was completion of the evaluation and tool standardization. Several reports also suggested that growing popularity and rising usage-based costs formed part of the background.
A good tool and a sustainable bill are separate management questions
June 30, 2026The reported internal-use deadline, coinciding with Microsoft's fiscal year-end
Only six months after access was opened in December 2025, a tool reportedly described as almost too popular became the subject of a transition.
What was reported: from popular tool to transition in six months
- December 2025
Reported development: Claude Code was made available to internal engineers for comparison alongside Copilot CLI.
- Through spring 2026
Reported development: Claude Code quickly grew popular among engineers, reportedly becoming almost too popular.
- May 2026
Reported development: Divisions including Windows and Microsoft 365 were notified of license termination, according to The Verge.
- June 30, 2026
Reported development: The reported deadline for Claude Code use, followed by transition to GitHub Copilot CLI.
Executive Rajesh Jha reportedly explained in an internal memo that the goal had been to compare both tools in real development work, and that Microsoft's ability to directly influence Copilot CLI's direction mattered. The company's stated explanation was the practical decision to finish an evaluation and standardize on its own tool.
One important detail is that Copilot CLI reportedly continued to offer a choice of underlying AI models, including Anthropic's Claude. The change therefore concerned the Claude Code tool and its license arrangement, not an end to using Claude models. Reports also noted that some engineers preferred Claude Code's user experience and regretted the transition.
The background identified by reporting: more than simply owning another tool?
The stated explanation remained completion of the evaluation and standardization. Alongside it, several reports discussed cost.
- Under the usage-based token arrangements discussed in the reports, more Claude Code usage meant higher costs. Other organizations reported heavy-user costs of USD 500–2,000 per month (article 15).
- The June 30 deadline coincided with Microsoft's fiscal year-end.
- Around the same time, Amazon and Meta were also the subject of reports about AI costs and usage metrics (article 12 and article 13).
The notable point is that Microsoft was successfully adopting AI: CEO Satya Nadella had publicly said that AI wrote around 30% of Microsoft's code. This is not a story of AI being useless. Even useful AI has metered costs that management must control and optimize, including at a company of Microsoft's scale.
Predictable pricing is a strength in its own right
Microsoft could switch to Copilot CLI, developed by its GitHub subsidiary. What options do organizations have when they cannot build or own their own AI tool?
One option is an upfront-purchase on-premises AI system such as Sovereign GaiXer. Local inference has no per-token fee, so a tenfold increase in use does not itself produce a tenfold token bill. Electricity, capacity, operations, and applicable maintenance still need budgeting. This changes the specific dilemma in which popularity alone creates a metered-cost problem.
Tool selection often focuses on capability and usability. This case also highlights predictability and control of cost. See article 5 on evaluating quality and article 9 on pricing changes.
It resembles switching from a popular metered car service to a company-owned vehicle. People may prefer the hired car, but its monthly meter charges are hard to forecast, and continuing to pay an outside provider is harder to justify when the company owns an alternative. This is the pattern suggested by the reporting.
Frequently asked questions
Did Microsoft reduce AI adoption itself?
No. The reports describe standardization on GitHub Copilot CLI, with AI use continuing. The question was which tool to pay for and under what pricing structure.
Could employees still use Claude through Copilot CLI?
According to reporting, yes. Copilot CLI allowed selection of underlying models, including Anthropic's Claude. The change concerned the Claude Code tool's internal licensing arrangement.
Was the reason cost or strategy?
The stated explanation was completion of testing and tool standardization; the company did not explicitly identify cost as the reason. Several reports separately analyzed rising usage costs and the fiscal year transition as background. This article distinguishes those accounts.
Does that mean we should standardize on our own tool?
Not necessarily. The lesson is to estimate company-wide costs before rollout and choose a pricing arrangement you can forecast, rather than to standardize for its own sake. Article 3 explains estimation.
We are piloting a metered tool. What should we watch?
Do not assume pilot spending scales only with headcount. In a wider rollout, experienced users can use the tool more intensively, so consumption per person may also rise. Popularity-driven overruns are possible in any organization.
Will an upfront-purchase system have lower capability than cloud AI?
It depends on the work. Cloud services have advantages where the latest very large models are required, while on-premises models can handle many internal drafting, summarization, and inquiry tasks. Validate the model against your own workload.
Summary
- Reports said Microsoft notified Windows- and Office-related divisions in May 2026 that internal Claude Code use would end by June 30, with a move to GitHub Copilot CLI.
- The stated reason was completed evaluation and standardization. Several reports also discussed increased metered costs from the tool's popularity.
- Even organizations succeeding with AI must manage its usage-based costs.
- Tool selection should include predictability and control of cost, alongside capability.
- Upfront-purchase local AI does not add per-token charges as usage grows; capacity and operating costs remain.
Related articles
A year's AI budget, gone in four months Uber's challenge of successful adoption
Adoption reaching 84% upset the budget assumptions. Lessons from figures shared by the people involved
How often prices change
Unit prices move. Keep reviewing your choice.
Accuracy and quality
Why the cheapest option is not always the best value
This article is based on public reporting, without independent interviews. It distinguishes the company's stated explanation from media analysis. Details reflect the reporting dates. Please contact us with factual corrections. The media, researchers, and companies discussed do not endorse Sovereign GaiXer.
Company, product, and service names mentioned are trademarks or registered trademarks of their respective owners.