Microsoft burned through its annual AI budget in just a few months. The Redmond giant has started massively restricting employee access to Anthropic’s Claude Code tool, redirecting developers to its own solution — GitHub Copilot CLI. The decision impacts thousands of engineers, project managers, and designers who had been relying on the tool in their daily work.
TL;DR: Microsoft exhausted its entire annual artificial intelligence budget in just a few months, primarily due to Claude Code token costs. The company is restricting access to the Anthropic tool and has set a June 30 deadline for complete license termination. Employees are being redirected to GitHub Copilot CLI.
Why Is Microsoft Restricting Access to Claude Code?
Microsoft previously expanded Claude Code access to thousands of employees — developers, project managers, and designers — to encourage experimentation with AI-powered coding tools. However, the token costs billed by Anthropic turned out to be so high that the company burned through its entire annual AI budget in just a few months. As a result, the decision was made to drastically cut access. According to Cybernews, Microsoft has set June 30 as the date for complete termination of the Claude Code license.
Anthropic’s token-based billing model proved too expensive for the Redmond giant. The company also needed to quickly rein in spending before the situation spiraled out of control. Redirecting employees to GitHub Copilot CLI is a strategic move — the tool is owned by Microsoft, giving the company full control over costs. Claude Code, described in detail in Claude Code Overview — Claude Code Docs, offers advanced coding features, but the price of those capabilities proved too high for mass deployment.
How Much Does AI Coding Actually Cost?
AI-assisted coding consumes significantly more computational resources than standard chatbot queries. Claude Code operates in an agentic mode — the tool reads files, analyzes the context of an entire project, executes terminal commands, modifies code, and runs tests. Each of these operations generates token traffic billed through the Anthropic API. As a result, a single development session can cost many times more than a regular conversation with a language model. As TheStreet reports, Microsoft canceled internal Claude Code licenses after costs spiraled out of control.
For example, tools like Claude Code /goal: auto-pilot for coding with evaluation after each round or claude code /ultraplan execute dozens of iterations in a single task. Each iteration requires sending the full context, which drastically increases token consumption. Uber has had similar experiences, with AI costs also growing at a rate that was difficult to control. This illustrates the scale of the financial challenge facing companies implementing AI.
How Many Employees Are Affected by the Changes?
The changes affected thousands of Microsoft employees. The company had previously expanded Claude Code access to a broad range of roles — not just developers, but also project managers and designers. The goal was to explore the tool’s potential across different positions. According to ETTelecom, the experiment proved too expensive to continue at its current scale.
Here is a summary of key facts about the cuts:
| Aspect | Details |
|---|---|
| Tool affected by cuts | Claude Code (Anthropic) |
| Replacement | GitHub Copilot CLI |
| License termination date | June 30 |
| Affected employees | Thousands (developers, PMs, designers) |
| Primary cause | Annual AI budget exceeded |
Employees are losing access in stages. Some teams can no longer use Claude Code, while others have reduced query limits. Reactions within the company are mixed — some engineers preferred Claude Code over GitHub Copilot, creating internal tensions. As USA Herald reports, the abrupt reversal has sparked internal chaos.
What Will Replace Claude Code Inside Microsoft?
Microsoft is redirecting employees to GitHub Copilot CLI — its own AI coding tool. This is a natural business decision. GitHub is owned by Microsoft, so costs are fully controlled internally. There is no external vendor that could raise prices or change billing terms. Additionally, integration with the Microsoft ecosystem is tighter than with external solutions.
Open-source tools offer an alternative for companies seeking savings. For example, claw-code: claude code rewritten from scratch (Open-source) offers similar functionality without commercial license costs. It is also worth checking out claude-warmup: managing the 5-hour Claude Code window, which helps optimize resource consumption. For Microsoft, however, the choice fell on a solution it fully controls.
Which Companies Besides Microsoft Are Struggling with AI Costs?
Microsoft is not the only company grappling with rising AI coding costs. Uber has also experienced rapid growth in AI spending, confirming that the problem is systemic. According to TheStreet, AI costs at Uber were growing at a rate that was difficult to control, forcing the company to review its spending strategy. Anthropic’s token-based billing model means that each agentic iteration generates costs proportional to the amount of code processed.
So the cost of Claude Code grows with every additional operation — file analysis, test execution, code modification. As a result, budgets intended for an entire year melt away in months. This shows that traditional budget planning methods are insufficient in the era of agentic AI tools.
Companies deploying AI at scale need to revise their approach to costs. Agentic tools consume significantly more resources than simple chatbots. Here are the key factors driving costs:
- Agentic mode requires repeatedly sending the full project context
- Each coding iteration generates separate token traffic billed through the API
- Lack of usage limits leads to uncontrolled spending growth
- Broad access across different roles (not just developers) multiplies costs
- Tools like Claude Code /goal: auto-pilot for coding with evaluation after each round execute dozens of iterations in a single task
- Costs grow proportionally to the amount of source code processed
- Experiments across different job roles increase token consumption
- Lack of internal monitoring tools makes budget control difficult
What Does This Situation Teach Us About AI Budgeting?
The Microsoft case serves as a warning for any company planning mass deployment of AI coding tools. Burning through an annual budget in a few months indicates that token-based cost models are difficult to predict at enterprise scale. The experiment, however, proved too expensive.
Moreover, the problem is not limited to one company. Uber is experiencing similar challenges with AI cost control. Companies must therefore implement cost control mechanisms before widely distributing AI tools.
Here are key takeaways for AI budgeting:
- Spending limits should be established before deploying agentic tools
- Pilot tests should include per-team cost measurement
- Open-source alternatives, like claw-code: claude code rewritten from scratch (Open-source), can significantly reduce costs
- Session window management tools, such as claude-warmup: managing the 5-hour Claude Code window, help optimize consumption
- AI budgets must account for agentic mode, not just chat queries
- Token consumption monitoring should be implemented from day one
- Experiments should be limited to small test groups
- Budgets must include a margin for unforeseen costs
How Are Microsoft Employees Reacting to the Changes?
Reactions among Microsoft employees to the restriction of Claude Code access are mixed. Some engineers preferred Claude Code over GitHub Copilot CLI, creating tensions within teams. According to USA Herald, the sudden reversal in AI policy sparked internal chaos, as thousands of employees lost access to a tool they had grown accustomed to.
Despite this, Microsoft is consistently redirecting engineers to GitHub Copilot CLI. The June 30 date marks the complete termination of the Claude Code license, as confirmed by Cybernews. Meanwhile, employees who integrated Claude Code into their daily workflow must now adapt to the new tool. This requires time and training, placing additional strain on teams.
Employees had been using various Claude Code features covered on this blog:
- claude code /ultraplan for advanced task planning
- Claude Code /buddy: 18 species of terminal pets as a terminal companion
- Claude Code Companion: How to Control Claude Code from Telegram for remote management
- Changelog — Claude Code Documentation for tracking updates
What Does the Future Hold for AI Coding Tools?
The Microsoft case illustrates the tension between the quality of AI tools and their costs. Claude Code offers advanced agentic capabilities, but the token-based billing model makes mass deployment financially risky. According to Social News XYZ, Microsoft began restricting Claude Code access and redirecting engineers to GitHub Copilot CLI due to rising costs. This suggests that companies will increasingly choose tools over which they have full financial control.
Furthermore, the growth of open-source tools could reshape the market landscape. Projects like claw-code: claude code rewritten from scratch (Open-source) demonstrate that alternatives exist without commercial license costs. The future of AI coding will therefore be shaped by the balance between functionality and cost. As Indica News reports, Microsoft scaling back Claude Code access as AI costs surge is a signal for the entire industry.
Key trends that will shape the AI coding tools market:
- Shift toward proprietary tools (like GitHub Copilot CLI) over external vendors
- Growth of open-source projects as alternatives to commercial solutions
- Implementation of cost control mechanisms and token limits
- Optimization of context sent to models to reduce costs
- Access segmentation — AI tools only for selected roles, not all employees
Frequently Asked Questions
When does Microsoft’s Claude Code license expire?
Microsoft has set June 30 as the date for complete termination of the Claude Code license, according to Cybernews. After this date, employees will use GitHub Copilot CLI exclusively.
Why is Claude Code more expensive than standard AI chatbots?
Claude Code operates in agentic mode, executing dozens of iterations in a single task, each requiring the full project context to be sent. As TheStreet reports, a single development session costs many times more than a regular conversation with a language model.
How many Microsoft employees lost access to Claude Code?
Claude Code access had been expanded to thousands of employees, including developers, project managers, and designers, as reported by ETTelecom. All of them have been redirected to GitHub Copilot CLI.
Are other companies also restricting access to AI tools?
Yes, Uber has also experienced rapid growth in AI costs, confirming that the problem is systemic. According to TheStreet, AI costs at Uber were growing at a rate that was difficult to control, forcing the company to review its spending strategy.
Summary
The Microsoft Claude Code case is a lesson for every company deploying AI at scale. Five key takeaways:
- An annual AI budget can be exhausted in a few months with mass use of agentic tools
- Token-based billing models are difficult to predict and control
- Proprietary tools (GitHub Copilot CLI) provide full cost control
- Open-source alternatives can significantly reduce AI spending
- AI budget planning must account for the specifics of agentic mode
Interested in the topic? Read the Claude Code Overview — Claude Code Docs to better understand how this tool works and why its costs can escalate quickly. Follow the gikiewicz.eu blog to stay up to date with AI coding tools.