Microsoft and Meta scale back internal use of Anthropic's Claude models

Two Tech Giants Cut Claude Spending, Push In-House AI Tools

Microsoft and Meta are both scaling back their internal use of Anthropic's Claude AI models, redirecting staff and budgets toward proprietary alternatives. Microsoft has cut its projected annual internal Claude spending by more than one-third from an anticipated $1 billion, while Meta has halved its internal Claude Code user base from roughly 60,000 to 30,000.

The moves signal a strategic pivot at two of the world's largest AI investors. Both companies continue to partner with or invest in Anthropic externally, but inside their own engineering and product teams, the message is clear: build and use what you own.

Microsoft mandates a full transition by mid-2026

Microsoft began rolling back internal Claude usage in December 2025, according to reporting by Cryptobriefing. The company has set a June 30, 2026 deadline for a complete migration from Claude Code to GitHub Copilot CLI. The spending reduction - from a projected $1 billion annually to something significantly lower - reflects a broader push to consolidate around Microsoft-owned developer tools.

GitHub Copilot, built on OpenAI's models, sits at the center of that strategy. Microsoft's $13 billion OpenAI partnership gives it deep access to the underlying technology, and executives are now enforcing internal adoption of the tools they sell to enterprise customers.

Meta slashes Claude Code users and cites data concerns

Meta's reduction from 60,000 to 30,000 internal Claude Code users comes with an additional financial data point. The Decoder reported that Meta spent over $105 million on Claude Code in a single 28-day period. That burn rate, combined with concerns about training data access, accelerated the shift to internal tools like MetaCode and Muse Code.

Meta's motivation goes beyond cost. The company wants to restrict Anthropic's access to its internal usage patterns and training data, a consideration that matters more as AI competition intensifies. By routing more developer activity through its own platforms, Meta keeps sensitive engineering signals in-house.

Anthropic's infrastructure commitments add context

Anthropic disclosed $31.4 billion in non-cancelable infrastructure commitments with Microsoft in its IPO filing, a figure reported by Cryptobriefing. That long-term contractual relationship continues regardless of Microsoft's internal usage decisions. The two companies remain commercially intertwined even as Microsoft tells its own employees to use something else.

The dynamic is not unusual in enterprise technology. Companies routinely sell infrastructure to partners while standardizing internally on a different stack. What makes this notable is the scale of the spend and the speed of the pullback.

Why this matters for executives and strategy leaders

Internal procurement patterns often preview where an organization is placing its long-term bets. When Microsoft and Meta - both with deep visibility into AI model performance and costs - choose to reduce reliance on a third-party model, it raises questions for any enterprise making similar build-versus-buy decisions. The cost differential alone, with Meta burning $105 million in under a month, suggests that internal tooling can offer material savings at scale. For strategy leaders evaluating AI tooling budgets, the signal is worth watching: even the companies funding the AI boom are choosing to own more of the stack.