Microsoft will abandon Claude: Is it too expensive or have they figured out how to exploit it?

Microsoft Can No Longer Afford Claude Code

Who would have thought that Microsoft—the tech giant that once invested over 10 billion dollars in OpenAI—recently halted internal use of Claude "because it's too expensive and unaffordable."

Here's what happened: Recently, news leaked from inside Microsoft that, starting June 30, thousands of engineers responsible for Windows, Microsoft 365, Teams, Outlook, and Surface-related work will no longer be allowed to use Claude Code. Microsoft is guiding them to switch to its own GitHub Copilot CLI.

Microsoft has not publicly disclosed the exact amount spent on Claude Code, but insiders reveal that the decision to stop using Claude Code was indeed due to costs being too high—so high that even Microsoft felt "painful."

Uber recently made a similar choice.

According to reports, Uber spends roughly $500 to $2,000 per month per engineer on AI tools.

What does this mean? For a team of 100 engineers, just this AI tool alone costs several million dollars a year. Uber's AI budget for 2026 was "burned through" by April.

Behind this lies a change many companies haven't fully realized yet but are already starting to worry about: the pricing model for AI is shifting from the old "fixed subscription" to a new "pay-as-you-go" model.

In the past, many AI tools charged a fixed monthly fee, making costs relatively predictable. But now, more and more programming-focused AI assistants are moving toward token-based billing—more complex questions, more frequent calls, deeper tasks, all lead to higher costs. For tech teams handling large amounts of coding daily, this expense is rapidly becoming a significant financial burden.

Against this backdrop, even tech giants like Microsoft and Uber have to reconsider: Are the high costs of third-party AI tools truly worth it? Should they continue paying rising bills, switch to more economical open-source solutions, or develop their own tools as replacements?

Microsoft's clear choice: replace Claude Code with its own GitHub Copilot CLI. Although it may offer a slightly inferior experience, it is cost-controlled and allows for more efficient internal resource flow.

This decision sends a clear signal—the rising "expensive" AI pricing is forcing companies to reevaluate their tech procurement strategies.

After all, the costs saved will ultimately reflect directly on profits.

However, The Verge also pointed out that canceling Claude Code licenses does not affect Microsoft's Foundry agreement with Anthropic, which includes investing up to $5 billion, providing Claude model access to Foundry clients, and Anthropic's commitment to spend $30 billion on Azure computing capacity.

Is letting employees use Claude Code just an experiment?

Microsoft suddenly revoked internal engineers' licenses to use Claude Code, just six months after allowing employees to use the tool. Many interpret this not as a hasty ban but as a carefully planned experiment.

According to an internal Microsoft memo, Rajesh Jha, EVP of the Experiences and Devices division, explained: "When we started offering both Copilot CLI and Claude Code, our goal was rapid learning—benchmarking these tools in real engineering workflows to see which best supported our teams. Claude Code played an important role in this learning process... At the same time, Copilot CLI brought us something particularly important: a product we could directly collaborate with GitHub to build, tailored to Microsoft's codebase, workflows, security expectations, and engineering needs."

In other words, Microsoft intentionally let a competitor's product into its engineering teams, exposing the shortcomings of its own Copilot CLI. Then, after six months of collecting feedback and fixing gaps, it shut down the competitor's tool and migrated all engineers back to its own product.

On LinkedIn, some users summarized this strategy as: "Let the competitor be the 'training partner,' learn from it, and then shut it down."

One LinkedIn user commented: "If Microsoft wanted to keep using Claude, cost wouldn't be an obstacle. Microsoft's previous Tokenmaxxing strategy seemed aimed from the start at learning."

Others noted, "Testing a competitor's product to pressure your own requires strong discipline. Applying what you've learned takes more effort."

In practice, Microsoft did just that. Over six months, Copilot CLI was iteratively improved based on engineers' feedback.

Thus, the abandonment isn't necessarily because it's "too expensive" and being passively given up, but rather a strategic move—leveraging the competitor's product to identify weaknesses, then ending the internal experiment after gaining insights.

However, opinions differ. Some developers point out that Microsoft's ability to do this depends on owning underlying cloud infrastructure, its own code hosting platform GitHub, and a large enough engineer base as "test subjects." Most companies lack these conditions—they simply can't afford to "stop using" and must pay, while Microsoft can "learn and pause."

The three major dilemmas behind stopping Claude Code

But the cost pressures and the "experiment" speculation are likely just the tip of the iceberg. Microsoft's decision to halt Claude Code involves much more than a financial ledger—it touches on a deeper, unsettling reality: in the era of large models, Microsoft is losing its defining power in the industry chain.

In March 2026, enterprise expense management platform Ramp released an AI Index, showing that among companies first purchasing AI services, Anthropic had about a 70% win rate over OpenAI in direct competition. This trend was the opposite of what Ramp observed in 2025, when OpenAI's popularity surged past all other model companies. Anthropic's annual revenue skyrocketed to $19 billion, nearly matching OpenAI's $25 billion.

By April, Anthropic's enterprise AI adoption rate reached 34.4%, surpassing OpenAI's 32.3% for the first time, becoming the new leading AI supplier in the enterprise market. The core driver of this reversal was Claude Code—launched just half a year earlier, it achieved an annualized revenue of $1 billion, accounting for 4% of all GitHub code commits at the time.

In this market round, Microsoft played almost no role.

As Microsoft was forced to rely on external models from OpenAI and Anthropic, by 2026, AI startups had an annual revenue of about $80 billion, with OpenAI and Anthropic splitting 89%.

This reveals a harsh truth: the commercial value of foundational models is flowing back to the model developers, while Microsoft—merely a channel—provides computing power and some investment but hasn't captured the lion's share of profits from the core model value.

In April 2026, Amazon announced a strategic partnership with OpenAI, committing up to $50 billion, with AWS acting as the exclusive third-party cloud distributor for OpenAI's enterprise platform, Frontier.

According to Business Insider, internal Microsoft assessments show that GitHub Copilot's market share in AI programming tools has fallen to about 25%.

These data points suggest that AI competition is shifting from "chatbots" to "engineering systems."

In this new wave, Claude Code is becoming a new infrastructure entry point. The problem is—Microsoft should have been the biggest beneficiary of this AI programming revolution, given that GitHub controls the world's largest developer ecosystem.

But now, with Claude Code occupying developers' minds, Anthropic gaining enterprise growth, and OpenAI gradually breaking free from Microsoft's exclusive grip, even GitHub Copilot is starting to be marginalized.

Microsoft suddenly realizes: owning GitHub doesn't necessarily mean owning the next-generation AI programming ecosystem.

One wrong step, and everything could go awry

The problems Microsoft faces today are no longer just about a product lagging behind.

On the surface, this is merely an "internal suspension of Claude Code," but deeper down, it reveals a chain of events spiraling out of control.

Initially, Microsoft failed to develop a truly competitive general-purpose large model comparable to GPT-4 or Claude. Without a strong foundational model, it could only rely long-term on OpenAI for core AI capabilities. But now, OpenAI is gradually shedding its exclusive ties with Microsoft, shifting from a "deeply bound" relationship to a "collaborative but non-exclusive" one.

Meanwhile, more dangerous developments are happening inside Microsoft.

An increasing number of engineers are now routinely using Claude Code instead of their own Copilot. While this appears to be a simple product choice, it actually impacts the entire development ecosystem: code workflows, debugging habits, engineering context, agent usage—all are migrating along with the tools. For a platform company, the greatest threat isn't competitors making money but its own developers working within competitors' ecosystems.

The problem then propagates further.

As developers shift heavily toward Claude Code, the real winners are Anthropic. Enterprise clients are also migrating, and Claude's influence in AI programming is rapidly expanding. Although Microsoft still earns from Azure cloud services, the most profitable and control-intensive part of AI—core models and platforms—is being taken over by model companies and agent platforms.

In this context, Microsoft is experiencing a subtle internal shift: product competition is intensifying, but it cannot simply let employees fully adopt external tools. The final move is to temporarily suspend internal use of Claude Code, not to catch up with Copilot, but to prevent losing control over core development tools.

This indicates that the problem has evolved from "product competition" to "organizational defense."

According to The Verge, Microsoft even considered acquiring Cursor to bridge the gap in AI programming experience, but later abandoned the plan due to regulatory concerns.

In a sense, this exposes Microsoft's current most awkward position: it owns one of the world's strongest developer platforms and a vast enterprise customer base, yet the most critical entry point in AI programming—the daily-used agent tools—is slipping into others' hands.

Once developer habits, workflows, and the engineering ecosystem are re-established elsewhere, regaining them later will be far more difficult than adding features or changing product strategies.