Meta and Microsoft pull back from Claude as Anthropic transforms from partner into competitor

Meta and Microsoft are Anthropic's two largest enterprise customers, and they are now significantly cutting back internal Claude usage. This is not a localized adjustment in one department, but a simultaneous pivot by the two largest buyers.

According to The Information, Microsoft's annual spending on Claude was originally projected to exceed $1 billion. That figure itself illustrates the scale of penetration Claude had achieved within Microsoft, and explains why the cut is so conspicuous.

A distinction is needed: the fact that both companies are cutting Claude usage is confirmed; while Microsoft's annual spending 'originally projected to exceed $1 billion' comes from The Information's report, a source claim without first-party confirmation.

Microsoft's move comes with clear personnel directives and budget figures. Executives Scott Guthrie and Jay Parikh instructed employees to switch to internal tools such as GitHub Copilot and OpenAI's models.

The spending contraction followed: Microsoft's spending on Claude was cut by more than a third, and the cloud division's monthly budget per employee dropped from $100,000 to about $10,000.

Put together, these two figures point not to a fine-tuning but to a shift in procurement logic—redirecting budget that once flowed to external models back toward its own and its partners' tools.

The change on Meta's side is equally dramatic. Its Claude Code user count fell from about 60,000 to 30,000, a halving.

The reasons for the user decline are not singular: partly due to layoffs, partly due to Meta promoting its own tools such as Muse Code and MetaCode. These two factors are difficult to separate based on available information, so it cannot be simply attributed to either one.

Cost pressure is real. In 28 days, Meta spent over $105 million on Claude Code alone. Meta had already introduced AI cost-cutting measures in June, indicating this round of contraction was not a spur-of-the-moment decision.

Attributing both companies' moves solely to saving money misses a more critical layer. Cost savings is one possible driver of the Claude usage cuts, but not the only one.

The more fundamental change lies in the nature of the relationship. Anthropic is growing extremely fast and is shifting from partner to competitor of incumbent vendors. When a supplier starts selling products that compete head-on with its customers, procurement decisions are no longer just about price.

Meta is also reportedly seeking to restrict Anthropic's access to its training data. This claim has not been confirmed by first-party sources, but it aligns with the logic that 'a customer is also a potential competitor': data is both an asset and a competitive bargaining chip.

Microsoft's position is more delicate than it appears. It is both a customer of Anthropic and a target being encroached upon by Anthropic's product line.

Tools such as Claude Cowork and ChatGPT Work are increasingly becoming alternatives to Office, and even Microsoft has reason to worry. This judgment comes from source reports and points to direct product-level competition, not merely cost considerations.

In other words, Microsoft's cut in Claude spending is driven both by the push of internal tool replacement and by defensive considerations about the competitive landscape.

From Anthropic's perspective, this round of contraction exposes a structural problem: reliance on a few large customers is becoming a strategic risk.

When the two largest customers simultaneously cut usage, revenue-side volatility is amplified. And the motivation for these two customers' cuts partly stems from Anthropic's own expansion—the more successful it becomes, the more it looks like a rival.

This judgment is an analytical inference, not a confirmed fact. The available evidence comes from a single report and lacks first-party sources, so judgments about long-term impact should remain cautious.