Meta and Microsoft Cut Internal Use of Anthropic’s Claude

Meta Platforms and Microsoft are reducing their employees’ internal reliance on Anthropic’s Claude as both technology companies push workers toward their own AI tools, according to a report by The Information. Microsoft has cut its projected internal Claude spending by more than one-third, while the number of Meta employees using Anthropic’s Claude Code has fallen to about 30,000 from roughly 60,000 earlier this year.

The development is notable because both companies remain major commercial partners of Anthropic. Microsoft continues to make Claude available through its enterprise products and cloud ecosystem, while Meta has been developing its own AI coding products and recently launched Meta Enterprise Platform around its proprietary AI stack. The shift therefore does not represent a complete break with Anthropic; it shows the companies are becoming more selective about where they pay for an outside model and where they use or develop their own systems.

Key takeaways

• Meta’s internal Claude Code users have fallen to about 30,000, down from roughly 60,000 earlier this year, according to The Information.
• Some of that decline reflects Meta’s spring layoffs, but the larger factor reported by The Information is the company’s push toward its own coding tools.
• Meta’s internal data showed more than $105 million spent on Claude Code during one recent 28-day period.
• MetaCode, an internal coding tool, had more than 30,000 users, while external-facing Muse Code had more than 6,000 employee users.
• Microsoft had been on pace for at least $1 billion a year in internal Anthropic spending before reducing that projection by more than one-third.
• Microsoft is encouraging employees to use GitHub Copilot and prioritize OpenAI models for internal coding workloads.
• Microsoft has not stopped using Claude: its customers can still access Anthropic models through Microsoft products and Azure.
• The development highlights a growing tension in enterprise AI: companies increasingly want access to the best model, but they also want to reduce dependency on expensive external AI providers.
• Anthropic is facing these changes while rapidly expanding revenue, customers and infrastructure commitments.

Meta and Microsoft are pulling back from Claude—but not abandoning Anthropic

The headline can make the development sound like Meta and Microsoft are walking away from Anthropic.

That would be misleading.

Both companies continue to have substantial relationships with the AI company.

Microsoft’s own documentation confirms that Anthropic models remain available in Microsoft Copilot, giving users the ability to select Claude for tasks such as complex analysis, document understanding and structured content generation.

The change is primarily about internal consumption.

For Microsoft, executives have been pushing employees to use Microsoft’s own AI infrastructure and products more efficiently rather than independently consuming large quantities of Anthropic models.

For Meta, employees are increasingly being directed toward AI coding products built by Meta itself.

That distinction matters because the economics of AI are changing.

During the early phase of generative AI adoption, companies often had a straightforward strategy: find the strongest model available and give employees access to it.

At scale, that can become extremely expensive.

A large technology company may have tens of thousands of engineers running coding agents continuously. If each agent consumes large numbers of tokens, model usage can turn into a major operating expense.

The companies are now discovering that AI capability and AI economics are two different problems.

Meta’s Claude Code usage has fallen by half

According to The Information, approximately 30,000 Meta employees were recently using Anthropic’s Claude Code, compared with around 60,000 earlier in 2026.

The decline has two reported causes.

The first was Meta’s spring workforce reduction, which cut about 10% of a workforce that had numbered roughly 78,000 people at the time.

The second—and reportedly more important—factor was Meta’s increasing push toward its own AI coding tools.

Meta has been developing products under its Muse family of AI systems. One of those products, Muse Code, is positioned as a competitor to Claude Code and uses Meta’s own models.

The Information reported that Muse Code had more than 6,000 Meta employee users, while another internal tool, MetaCode, had more than 30,000 users.

That gives the shift a clear strategic dimension.

Meta is not simply trying to reduce its AI bill.

It is also using its own employees as an internal testing and adoption base for products that it ultimately wants to sell to other businesses.

That strategy fits with Meta’s broader enterprise push.

In September, Meta announced Meta Enterprise Platform, saying it would bring its AI technology—including Muse, Meta Business Agent, Muse API and Muse Code—to businesses and developers.

Internal adoption therefore becomes commercially useful.

If Meta employees can perform their own coding work with Meta’s tools, the company gains practical feedback while reducing the amount it spends on an outside competitor.

Meta was spending heavily on Claude Code

The economics help explain why Meta is making the transition.

The Information reported that Meta spent more than $105 million on Claude Code in a recent 28-day period.

That is not a full-year spending figure and should not be extrapolated directly into an annual Claude bill.

But it illustrates the scale that AI coding tools can reach inside a large technology company.

Meta employees had reportedly begun using Claude heavily enough to create an internal culture around maximizing model usage. The Information described employees engaging in “tokenmaxxing” and competing on an employee-created leaderboard called “Claudeonomics.”

The resulting usage helped trigger tighter controls.

In June, Meta leadership reportedly moved toward budgets, token limits and centralized monitoring for external AI tools.

The underlying problem is familiar to companies adopting AI agents.

Traditional software generally has a relatively predictable licensing cost per employee.

AI agents can behave differently.

An engineer can repeatedly ask a coding model to inspect a repository, write code, run tests, diagnose failures and make additional changes. A complicated task can therefore generate a large number of model calls.

When multiplied across thousands of employees, apparently small per-user costs can become very large.

Meta wants to turn its internal AI tools into a business

The shift is particularly important because Meta is no longer developing AI tools only for its own employees.

Its September announcement of Meta Enterprise Platform explicitly positioned the company’s AI stack as a commercial offering for businesses and developers.

Muse Code is part of that strategy.

This creates a feedback loop.

Meta can give employees its own tools.

Employees use those tools on real engineering problems.

The company can collect feedback and improve the products.

Those products can then be offered to outside customers.

The more successful the internal transition becomes, the stronger Meta’s argument becomes that enterprises can use Meta’s AI systems instead of relying entirely on third-party coding assistants.

That is a very different strategy from simply purchasing the best external model.

Microsoft is taking a similar approach

Microsoft’s internal Claude strategy is more complicated because the company has a particularly deep relationship with Anthropic.

The Information reported that Microsoft employees were on track to spend at least $1 billion annually on Anthropic technology earlier this year. That estimate has since been reduced by more than one-third after executives encouraged employees to use less Claude and rely more heavily on Microsoft’s own AI tools.

Microsoft executives including Scott Guthrie, who leads Azure, and Jay Parikh, who leads the company’s developer tools division, reportedly urged employees to be more cautious about AI spending.

Microsoft subsequently made OpenAI’s models the default for internal AI coding, according to a Microsoft spokesperson quoted by The Information. Employees can still use Claude, however.

This is especially significant because Microsoft owns and operates a large part of the software infrastructure used by developers.

Rather than allowing engineers to independently purchase or consume external AI tools, Microsoft has an incentive to route that activity through products such as GitHub Copilot.

That gives the company greater visibility into usage, spending and model selection.

Why GitHub Copilot matters

GitHub Copilot has become a key part of Microsoft’s strategy for managing AI coding internally.

The product can provide access to different models, including Anthropic’s Claude. Microsoft’s documentation confirms that Anthropic models are also available within its broader Copilot environment.

That creates an important distinction between using Claude and buying Claude directly.

Microsoft can allow employees to access Anthropic’s capabilities through a controlled Microsoft product while still managing how the models are consumed.

The Information reported that Microsoft’s “auto” mode in GitHub Copilot routes queries to OpenAI models by default in most cases, reducing Microsoft’s need to pay separately for OpenAI model usage because of its commercial relationship with OpenAI.

The result is a more centralized AI strategy.

Employees can still have access to multiple models, but Microsoft can influence which models are used and how much AI compute each business division consumes.

Microsoft is not cutting Anthropic out of its customer products

This is perhaps the most important qualification in the story.

Microsoft’s internal cost-cutting should not be interpreted as a decision to stop using Claude across its commercial products.

The Information reported that Microsoft continues to spend substantially on Anthropic models that power customer-facing Copilot features.

Microsoft’s own documentation also confirms that organizations can enable Anthropic models within Microsoft online services and Copilot.

That creates an unusual relationship.

Microsoft can simultaneously:

• Reduce its own employees’ direct consumption of Anthropic models.
• Continue offering Claude to enterprise customers.
• Sell Anthropic models through Azure.
• Use OpenAI models heavily across its products.
• Develop its own AI capabilities.

In other words, Microsoft is treating AI models increasingly as interchangeable components rather than committing its entire software ecosystem to one provider.

The AI industry is becoming a multi-model market

This is the broader lesson from the Meta and Microsoft moves.

The enterprise AI market is moving toward a multi-model architecture.

Large companies increasingly want the ability to choose between OpenAI, Anthropic, Google and their own models depending on the task.

The reason is straightforward: no single model is guaranteed to be best at every workload, and model prices can change quickly.

A company might prefer one model for coding, another for reasoning, another for document processing and an internal model for highly sensitive workloads.

Microsoft’s own product architecture illustrates this direction.

Its Copilot products can expose multiple models to users, while Microsoft controls the surrounding infrastructure, security and administration.

That gives the software company an important strategic advantage even when it is not the company that created the underlying model.

Anthropic is becoming both partner and competitor

The situation is particularly complicated for Anthropic because companies like Microsoft and Meta increasingly compete with it.

Microsoft has a commercial relationship with Anthropic and provides access to Claude through its ecosystem.

But Microsoft is also developing Copilot products that increasingly perform tasks that overlap with enterprise software.

Meta is in an even more direct position with its Muse products.

Anthropic is therefore selling technology to companies that are simultaneously building competing AI products.

That can create an uncomfortable economic dynamic.

The better Claude becomes, the more valuable it is to customers.

But the more deeply those customers integrate Claude, the more money they may spend on an outside provider that could eventually compete with their own AI products.

Meta’s move toward Muse Code and MetaCode demonstrates how that tension can play out.

Anthropic’s growth makes the shift more significant

The pullback is happening despite Anthropic’s rapid growth.

The Information reported that Anthropic’s annualized revenue had passed approximately $65 billion in July, more than seven times its level at the start of the year, based on reporting around the company’s IPO preparations.

Reuters has separately reported that Anthropic has made enormous long-term infrastructure commitments as it prepares to scale its AI operations. A confidential IPO filing showed at least $518 billion in planned infrastructure spending over a decade, including major commitments involving Google, Amazon, Microsoft and other technology providers.

That creates a paradox.

Anthropic is growing extremely quickly and attracting huge amounts of enterprise demand.

At the same time, some of its largest customers are actively trying to control how much they consume.

For Anthropic, this means revenue growth cannot simply depend on maximizing usage by a handful of giant customers.

The company needs a broader customer base and must continue proving that its models deliver enough value to justify their costs.

Cost is becoming a competitive weapon

AI companies initially competed heavily on model quality.

Now cost is becoming equally important.

If two models can perform a coding task with similar quality but one costs significantly less, a large enterprise has a strong reason to shift workloads.

That is particularly true when a company is processing billions of tokens.

The Information reported that other businesses have also looked for cheaper alternatives as AI costs rise.

Open-source models add another pressure.

Companies with strong engineering teams can increasingly customize or deploy open models for selected workloads instead of paying a frontier AI provider for every request.

That does not eliminate demand for frontier models.

It changes where companies use them.

A business may reserve its most expensive models for difficult reasoning tasks while using cheaper models for routine summarization, classification or coding.

The result is a market where model routing becomes an important part of AI infrastructure.

What the shift means for AI developers

The changes at Meta and Microsoft also reveal something about the future of AI coding.

Coding assistants are becoming deeply embedded in software engineering workflows.

But the winning product may not necessarily be the model with the highest benchmark score.

The winner may be the platform that combines:

• A strong model.
• Low inference cost.
• Access to the company’s codebase.
• Security and permissions.
• Developer tooling.
• Monitoring and usage controls.
• Integrated testing and deployment.
• A simple billing model.

This is why Microsoft can tolerate Claude inside Copilot while simultaneously pushing employees away from standalone Claude Code.

The value increasingly lies in the entire workflow rather than just the model.

Meta is pursuing a similar strategy with Muse Code and its broader enterprise platform.

India relevance: enterprise AI buyers may face the same decision

The development is relevant to Indian technology companies because many large Indian IT services firms and startups are also experimenting with multiple AI models.

The lesson is not that companies should stop using Anthropic.

It is that enterprises are likely to become more deliberate about which model is used for which job.

For an Indian software company serving global customers, an AI coding assistant may be extremely valuable. But if thousands of developers use an expensive model continuously, the cost can quickly become significant.

Companies may therefore combine proprietary models, commercial APIs and open-source systems.

The same approach could eventually become common across software development, customer support, analytics and internal knowledge management.

However, there is no evidence from the reported Meta and Microsoft changes that Indian companies are following the same policy. The India connection is an industry implication rather than a reported development.

What happens next

Meta is likely to continue expanding Muse Code and its broader enterprise AI stack as it tries to make its proprietary models useful outside the company. Its Meta Enterprise Platform announcement makes clear that enterprise AI is now a major strategic focus.

Microsoft, meanwhile, is unlikely to abandon Claude. Its current strategy appears more nuanced: control internal AI consumption while continuing to give customers access to Anthropic models through Copilot and Azure.

For Anthropic, the key question is whether rapidly expanding external demand can offset efforts by its biggest technology customers to reduce their own internal consumption.

The answer will depend on three variables: model quality, price and the number of enterprise workloads that Claude can win against increasingly capable alternatives.

FAQs

Are Meta and Microsoft stopping the use of Claude?

No. The reported changes concern primarily internal employee usage. Microsoft continues to make Anthropic models available through Copilot and its enterprise ecosystem, while Meta continues to have a relationship with Anthropic.

Why is Meta reducing Claude Code usage?

The Information reports that Meta is increasingly pushing employees toward its own AI coding products, including MetaCode and Muse Code. Employee layoffs also contributed to the decline in Claude Code users.

How much was Meta spending on Claude Code?

The Information reported that Meta spent more than $105 million during a recent 28-day period on Claude Code. This is a reported period-specific figure, not an annual spending estimate.

How much was Microsoft spending on Anthropic AI?

Microsoft employees were reportedly on pace to spend at least $1 billion annually on Anthropic technology earlier in 2026. The Information reported that Microsoft subsequently reduced that projected internal spending by more than one-third.

Looking Ahead

The Meta and Microsoft moves show that the next stage of enterprise AI adoption will not simply be about finding the smartest model. Large technology companies are increasingly treating AI models as components that can be switched, routed and optimized according to cost, performance, security and strategic value. That puts pressure on Anthropic, OpenAI, Google and other model providers to compete not only on intelligence but also on economics.

For Anthropic, the situation is particularly important because some of its biggest customers are also developing competing AI products. Meta and Microsoft can continue buying Claude where it gives them an advantage while simultaneously building alternatives that reduce their dependence on it. If that pattern spreads across the enterprise market, Anthropic’s extraordinary growth will increasingly have to come from a broad base of customers rather than unlimited consumption by a few technology giants.