Meta and Microsoft Scale Back Internal Claude Use as AI Costs Rise
The growing cost of artificial intelligence is becoming an increasingly important factor in how major technology companies manage their internal AI tools.
According to a report by The Information, Meta and Microsoft are scaling back employees’ internal use of Claude, Anthropic’s AI model, while encouraging workers to rely more heavily on tools and systems developed by the companies themselves.
The shift does not mean either company has ended its commercial relationship with Anthropic. Claude remains available to customers through various AI services offered by both companies, highlighting their attempt to balance the benefits of external models with the growing use of internally developed AI products.
Microsoft Reassesses Its Claude Spending
Microsoft is taking a more cautious approach to internal Claude usage because of the model’s cost.
The company had expected its internal spending on Anthropic technologies to reach at least $1 billion during the current year. However, according to the report, Microsoft has reduced that spending forecast by more than one-third.
The company has also encouraged employees to use Microsoft’s own AI products whenever possible, particularly as Microsoft has developed a broad portfolio of competing AI tools.
Microsoft previously gave thousands of developers access to Claude Code in December of last year, and the Anthropic tool quickly gained popularity among engineering and development teams.
GitHub Copilot Takes a Larger Role
Microsoft subsequently began encouraging developers to use GitHub Copilot CLI instead of relying extensively on Claude Code.
The move is linked not only to cost considerations but also to Microsoft’s broader goal of increasing its engineers’ use of the company’s own products.
The Experiences + Devices division, which includes teams working on products such as Windows, Microsoft 365 and Surface, had been preparing to end most Claude Code licenses by the end of June.
By increasing internal adoption of its own AI tools, Microsoft can also gather direct feedback from its developers and use their experiences to improve its AI products.
Despite the reduction in internal usage, Claude remains part of Microsoft’s wider AI ecosystem. The model is available through Microsoft Foundry and is also used for certain tasks connected to Microsoft 365 and Copilot services.
Meta Shifts Employees Toward Its Own AI Tools
Meta is following a similar strategy.
The number of Meta employees using Claude Code has fallen to around 30,000, compared with approximately 60,000 earlier in the year.
Part of the decline is linked to workforce reductions at the company. At the same time, however, Meta is actively increasing its reliance on software tools built around its own AI models.
The company’s internal MetaCode tool has surpassed 30,000 users, while Muse Code, which Meta is testing as a competitor to Claude Code, has attracted more than 6,000 internal users.
AI Bills Become a Growing Corporate Challenge
The moves by Meta and Microsoft highlight a broader challenge facing large companies as AI adoption expands.
As more employees use advanced models and workloads increase, AI tools can shift from being technologies intended to improve efficiency and reduce costs into major corporate expenses.
An Axios report published in May highlighted a case in which one company’s Claude bill reportedly reached around $500 million in a single month after insufficient limits were placed on employees’ use of its licenses.
These developments suggest that competition in the AI market is increasingly about more than developing the most advanced model. Companies are also looking for ways to operate AI systems efficiently while keeping costs under control.
Companies Reconsider Their Dependence on External AI
As major technology companies continue developing their own AI models and software, their reliance on external systems appears likely to become more selective.
External models can still play an important role, particularly when they offer performance or capabilities that internal systems cannot match. However, companies may increasingly use their own tools for tasks that can be handled more affordably or that are closely connected to their internal workflows.
The strategies adopted by Meta and Microsoft indicate that enterprise AI may not ultimately revolve around choosing a single model. Instead, companies could distribute workloads across multiple systems based on cost, performance and the specific requirements of each team, while giving greater priority to AI tools they develop and control themselves.