Microsoft to Discontinue Using Claude: Is It Too Expensive or Have They Figured It Out?

1  Even Microsoft Can't Afford Claude Code Anymore 

Who would have thought that Microsoft, the tech giant that has invested over $10 billion in OpenAI, has recently stopped using Claude internally because it's "too expensive."

Here's what happened: Recently, news spread within Microsoft that starting from June 30th, thousands of engineers working on Windows, Microsoft 365, Teams, Outlook, and Surface will no longer be allowed to use Claude Code. Microsoft has directed them to switch to its own GitHub Copilot CLI.

Microsoft hasn't disclosed the exact amount it spent on Claude Code, but insiders revealed that the decision to stop using it was indeed due to high costs, so high that even Microsoft felt the pain.

Uber made the same choice as Microsoft not long ago.

According to reports, Claude Code costs each Uber engineer about $500 - $2000 per month for AI tool usage.

What does this mean? For a technical team of 100 people, this single AI tool would cost millions of dollars a year. Uber burned through its AI budget for 2026 in April.

Behind this lies a change that many companies haven't noticed yet but are starting to worry about: the pricing model of AI is shifting from the previous "package deal" to the current "pay - as - you - go."

In the past, many AI tools adopted a fixed - monthly - fee model, and the costs were relatively predictable. However, now, more and more AI assistants for programming scenarios are switching to a token - based billing method. The more complex the questions, the more frequent the calls, and the deeper the tasks, the higher the costs. For technical teams that need to handle a large amount of coding work daily, this expense is rapidly becoming a significant financial burden.

In this context, even tech companies of Microsoft and Uber's scale have to recalculate: Are the high costs of third - party AI tools really worth it? Should they continue to pay the rising bills, or switch to more cost - effective open - source solutions or use their own tools instead?

Microsoft's choice is clear: replace Claude Code with its own GitHub Copilot CLI. Although the functionality and user experience may be slightly inferior, the costs are controllable, and the internal resource flow is more efficient.

This choice sends a clear signal - the AI pricing that even Microsoft finds "expensive" is forcing enterprises to re - evaluate their technology procurement strategies.

After all, the saved costs will directly reflect in the profits.

However, The Verge also pointed out that canceling the Claude Code authorization won't affect Microsoft's Foundry agreement with Anthropic. This agreement includes investing up to $5 billion in Anthropic, providing Foundry customers with access to the Claude model, and Anthropic's commitment to spend $30 billion on Azure computing power.

2  Was Letting Employees Use Claude Code Just an Experiment? 

Microsoft suddenly revoked the internal engineers' permission to use Claude Code six months after allowing employees to use it. So, the outside world believes that this is not a hasty ban but rather a well - planned experiment.

According to an internal Microsoft memo, Rajesh Jha, the executive vice - president of the Experiences and Devices Division, explained: "When we started offering both Copilot CLI and Claude Code, our goal was to learn quickly, benchmark these tools in real - world engineering workflows, and understand which tools best support our teams. Claude Code played an important role in this learning process... Meanwhile, Copilot CLI also brought us something particularly important: a product that we can directly collaborate with GitHub to build according to Microsoft's codebase, workflows, security expectations, and engineering requirements."

In other words, Microsoft actively introduced a competitor's product into its engineering team to expose the shortcomings of its own Copilot CLI. Then, it spent six months collecting feedback, fixing the gaps, and finally shut down the competitor's tool and migrated all engineers back to its own product.

On the LinkedIn platform, some users summarized this strategy as: let the competitor's product be the "sparring partner" first, and then end it after learning.

A LinkedIn user commented: "If Microsoft wanted to continue using Claude, cost would definitely not be a hindrance. Microsoft's previous Tokenmaxxing strategy seemed to be aimed at learning from the start."

Another user said, "Using a competitor's product to conduct stress tests on your own product requires strong self - discipline. And putting the learned knowledge into practice requires even more effort."

As a result, Microsoft did just that. Copilot CLI has gone through multiple iterations in six months based on the engineers' feedback from comparative use.

So, this abandonment is interpreted not as a passive move due to "unaffordability" but as an active end to this internal experiment after leveraging the competitor's product to make up for its own shortcomings.

However, the outside world has different views. Some developers pointed out that Microsoft can do this because it has the underlying cloud infrastructure, its own code hosting platform GitHub, and a large enough engineer group as an "experimental sample." Most companies don't have these conditions - they can only simply "not afford it," while Microsoft can choose to "stop after learning."

3  Behind the Decision to Stop Using Claude Code, Microsoft Faces Three Major Dilemmas 

However, cost pressure and the outside world's speculation of an "experimental test" may just be the tip of the iceberg. Microsoft's decision to stop using Claude Code is far more than just a financial calculation - it touches on a fact that makes this software giant even more uneasy: in the industrial chain of the large - model era, Microsoft is losing its defining power.

In March 2026, the corporate spending management platform Ramp released an AI Index. Among the enterprises that purchased AI services for the first time, Anthropic won about 70% of the direct confrontations with OpenAI. This is completely opposite to the trend observed by Ramp in 2025, when OpenAI's popularity outpaced that of any other model company. Anthropic's annualized revenue soared to $19 billion, approaching OpenAI's $25 billion.

By April, Anthropic's corporate AI adoption rate had reached 34.4%, surpassing OpenAI's 32.3% for the first time and becoming the new top AI supplier in the corporate market. The core engine driving this reversal is Claude Code - this programming tool achieved an annualized revenue of $1 billion in just six months after its release, accounting for 4% of all GitHub code submissions at that time.

In this round of the market, Microsoft hardly has a place.

While Microsoft is forced to rely on the external models of OpenAI and Anthropic, the annualized revenue of AI startups in 2026 has reached $80 billion, and OpenAI and Anthropic together account for 89% of it.

This reveals a cruel fact: the commercial value of the basic model is flowing back to the model developers, and Microsoft is just a distributor. When the distributor tries to block the source, it only means one thing: it can no longer buy a real ticket to enter the game.

In summary, it can be said that Microsoft has been marginalized in three battlefields: models, developers, and ecological control.

Dilemma 1: Lack of Cutting - Edge Basic Models and Heavy External Dependence 

To this day, Microsoft's biggest concern is that it still doesn't have a truly cutting - edge general large model of its own.

Since 2019, Microsoft has invested over $13 billion in OpenAI and acquired about 27% of its equity, but its self - developed large - language model has never been able to match GPT - 4 or Claude.

In April 2026, Microsoft's AI research laboratory released three MAI - series models - MAI - Transcribe - 1, MAI - Voice - 1, and MAI - Image - 2 - which only cover voice transcription, voice generation, and image creation, and no general large - language model was launched.

Although Microsoft has one of the world's strongest AI business entrances, it lacks the "basic - model control power" that truly determines the upper limit of AI capabilities.

Without a self - developed general model, Microsoft cannot achieve a technological closed - loop in core scenarios such as general conversation and programming reasoning, and its core AI capabilities rely on OpenAI. In April 2026, Microsoft and OpenAI jointly announced the end of their seven - year exclusive binding. Azure is no longer the only cloud outlet for OpenAI, and the IP authorization has changed from exclusive to non - exclusive.

One of Microsoft's biggest moats in the past was "exclusive access to OpenAI." But now, this exclusivity is loosening.

Once Microsoft loses its exclusive binding with OpenAI, it has to face a cruel reality: it doesn't have an underlying model that can replace GPT - 4 or Claude. This is why Microsoft presents a very fragmented state in the AI field today: it is one of the companies with the deepest AI commercialization globally, but at the same time, it highly depends on external models for core capabilities. This "strong platform, weak model" structure is essentially a technological hollowing - out.

Dilemma 2: Its Own Product Power Is Inferior to Competitors 

What makes Microsoft even more sad is that not only does it not have a presentable general large model, but even Copilot, which once seized the early window of intelligent programming, has been substantially crushed by Claude Code.

In the past two years, GitHub Copilot has always been synonymous with AI programming. However, the AI programming market in 2026 has undergone fundamental changes. What truly changed the industry about Claude Code is that it transformed the "code completion tool" into a "long - context engineering agent."

The traditional Copilot is more like "helping you write a few lines of code," while Claude Code can directly participate in the entire software engineering process.

Within Microsoft, the most popular programming tool is not its own Copilot but Claude Code.

According to Tom Warren, a reporter from The Verge, Microsoft engineers have "obviously favored" Claude Code over their own tools in the past few months. This preference is not because employees "don't like to use their own products" but because there is a substantial gap in product power.

According to test data, Claude Code scored 80.8% on the SWE - bench, while GitHub Copilot based on GPT - 4o only scored 72.5%, a difference of 8.3 percentage points.

Claude Code supports a context window of one million tokens and can process about 3000 files in a single session, while the upper limit of Copilot CLI is only 128K tokens. In scenarios of refactoring or debugging across dozens of files, Claude Code's completion rate reaches 89%, while Copilot's is only 60%.

Engineers' daily use of Claude Code means that the development workflow, debugging data, and operating habits are all being deposited in Anthropic's ecosystem. According to The Verge, before opening Claude Code internally, 91% of Microsoft's engineering teams were using GitHub Copilot, but the usage of Claude Code in the past six months has "seriously eroded" this proportion.

Rajesh Jha, the head of Microsoft's Experiences and Devices Division, admitted in the memo that Claude Code was "an important part of the learning process," but still required a mandatory switch. The root of this contradiction lies in the fundamental anxiety at the strategic level - when engineers rely on external products for key links in the development toolchain, Microsoft's control over its own technology stack is gradually weakening.

Employees' long - term use of external tools means cultivating the users' habits of competitors and bringing development skills and process knowledge directly to competing companies in the future.

A Microsoft insider told The Information that the satisfaction rate of Claude Code among Microsoft engineers is as high as 91%. When a company's core developers are less satisfied with their own tools than external competitors, its technological cohesion faces a major challenge. This is not about "fear of competitors making money" but about worrying that the development culture will be penetrated by external tools, leading to the locking of core talents and development processes in competitors' products.

Dilemma 3: Weakening Ecological Control Power 

What makes Microsoft even more uncomfortable is that not only are internal engineers switching to Claude, but a similar trend is also emerging in the entire corporate market.

Microsoft has invested in two major partners, OpenAI and Anthropic, but both are gradually getting rid of their dependence on Microsoft.

According to the Ramp AI Index data, in April 2026, Anthropic's corporate paid adoption rate reached 34.4%, surpassing OpenAI's 32.3% for the first time. In the past 12 months, Anthropic's corporate adoption rate has soared from only 9% to 34.4%, nearly a four - fold increase, while OpenAI's corporate adoption rate has only increased by 0.3% during the same period.

When enterprises purchased AI services for the first time in 2026, in about 70% of the direct confrontations, the final contract was signed for Claude rather than ChatGPT.

The core engine driving this reversal is exactly Claude Code.

According to market estimates, about 4% of global public GitHub submissions are completed with the participation of Claude Code, and Anthropic expects this proportion to exceed 20% by the end of 2026. Claude Code occupies 54% of the AI programming tool market, and 8 out of the world's top ten Fortune 500 companies are its paying customers. In terms of annualized revenue, Claude Code exceeded $1 billion in November 2025 and reached $2.5 billion at the beginning of 2026. In contrast, OpenAI's Codex has an annualized revenue of about $1 billion.

According to Goldman Sachs statistics, the total annualized revenue of AI startups in 2026 is about $80 billion,