Meta and Microsoft strive to reduce employees’ reliance on Claude: Meta’s internal usage has halved, and Microsoft has cut its budget by one-third.

Meta and Microsoft are working to reduce employees’ use of Claude, putting pressure on Anthropic’s revenue outlook.

On October 5th, according to The Information, Meta and Microsoft are drastically cutting internal use of Anthropic’s Claude AI in favor of promoting their own in-house tools among staff.

This shift directly impacts the core revenue structure of Anthropic. In its recent IPO prospectus, the company disclosed that these two major customers together contribute around 25% of its revenue.

Microsoft previously projected its annual internal spending on Claude to be at least $1 billion, but this budget has already been reduced by more than one-third. Meanwhile, the number of Meta employees using Claude Code internally has plummeted from about 60,000 earlier this year to around 30,000—a drop of nearly 50%.

This adjustment is of great significance for Anthropic. While both tech giants have emphasized that external enterprise customers’ use of Claude remains on the rise, the shrinking internal usage signals that major technology companies are accelerating the move to bring AI capabilities in-house.

The core logic behind Microsoft’s move is to control operating costs.

According to reports citing informed sources, Microsoft began reducing its dependence on Claude Code as early as May this year, aiming to limit employees’ token consumption and promote the adoption of its own internal models.

Microsoft had previously estimated internal use of Claude AI would reach $1 billion in spending, but after instructing employees to cut back on Claude to save costs and spend more time on Microsoft’s proprietary AI tools, this budget has been reduced by over one-third.

It is noteworthy that this adjustment is limited to internal operations; Anthropic model services offered to external clients via the Microsoft enterprise platform are continuing to grow steadily.

The drop in Meta’s internal usage of Claude is due to both personnel changes and strategic intentions.

This spring, Meta laid off around 10% of its 78,000 employees, which helped lower the baseline of Claude usage. However, reports indicate that the deeper driver is the rapid maturation of Meta’s own AI ecosystem.

Meta has launched two internal programming assistant tools: MetaCode, which is for internal use only and now has over 30,000 users; and Muse Code, which began external client testing in August, with internal users now exceeding 6,000.

Muse Code is seen as a direct competitor to Claude Code, built on Meta’s proprietary models and is part of the Muse Spark series product line.

Despite a sharp decline in the number of Claude Code users, Meta’s spending on Claude Code in just the past 28 days has still exceeded $105 million.

In its recent IPO prospectus, Anthropic disclosed that two customers together contributed about 25% of its revenue, without naming specific clients. The market widely believes that Meta and Microsoft are likely these two customers or at least one of them.

Both key customers are simultaneously reducing internal usage, and the rapid rise of in-house alternative tools has left Anthropic facing the structural risk of overreliance on key clients during this critical IPO window.

Although external enterprise demand is still growing, whether it can compensate for the revenue lost from shrinking internal usage remains an uncertainty closely watched by investors.