Claude Pullback: Meta and Microsoft Cut Internal Use

Two of Anthropic's largest enterprise customers are scaling back. Meta and Microsoft are both reducing how much their own staff use Claude, Anthropic's AI model family. The cuts are large, and they say something about how the relationship between AI labs and Big Tech is changing.

The figures come from a report by The Information. The motive appears to be more than cost control.

Microsoft redirects staff to in-house options

Microsoft's Claude bill had been on course to pass $1 billion a year. That projection is no longer the plan. Executives Scott Guthrie and Jay Parikh told employees to use internal tools instead, including GitHub Copilot and models from OpenAI, Microsoft's long-standing AI partner.

The effect was immediate. Claude spending dropped by more than a third. In the cloud division, the monthly per-employee budget reportedly fell from $100,000 to roughly $10,000.

Meta halves its Claude Code user base

Meta's cuts look similar. The number of employees using Claude Code, Anthropic's agentic coding tool, fell from about 60,000 to around 30,000.

Two things drove that drop. Part of it came from layoffs. The rest came from Meta steering engineers toward its own tools, such as Muse Code and MetaCode.

The scale of the spending explains why Meta is paying attention. In one 28-day period, the company spent more than $105 million on Claude Code alone. Meta had already introduced measures to cut AI costs in June, so this move continues a trend that was already underway.

Not just about the money

Saving money is one likely reason. It is probably not the only one.

Meta wants to sell products that compete with Anthropic's. Paying a rival large sums to power internal work does not fit that plan. According to the report, Meta also wanted to limit Anthropic's access to its training data. That concern is somewhat ironic, given Meta's own track record on using other people's data.

There is also the issue of scale. Anthropic has grown very quickly, and incumbents now see it as a real problem. Microsoft has its own reasons to be cautious. Products like Claude Cowork and ChatGPT Work are increasingly becoming alternatives to Office, which is one of Microsoft's core businesses.

So the same company that looked like a supplier a short time ago now looks a lot like a competitor.

Our Take

This suggests the era of easy partnerships between frontier labs and Big Tech is ending. Anthropic used to sell models that other companies built on. It now ships products, including coding agents and workplace tools, that compete directly with what its biggest customers sell. When that happens, customers start asking whether they are paying to fund a rival.

The two companies' responses are telling. Neither is dropping AI-assisted work. Both are moving spending to tools they own or partners they trust more. For Microsoft that means GitHub Copilot and OpenAI. For Meta it means Muse Code and MetaCode. The demand for AI is still there. What has changed is who gets paid for it.

For teams buying AI tools, there is a practical lesson. Vendor choice is not only a question of benchmarks and price. It also depends on where a provider is heading strategically and who it might compete with next. That matters for training data and for long-term lock-in.

Several things are worth watching. One is whether other large customers that compete with Anthropic follow Meta and Microsoft. Another is whether Anthropic's growth makes up for the lost revenue, or whether losing two major accounts shows up in its numbers. A third is whether the in-house tools at Meta and Microsoft are good enough that engineers stay with them, or whether usage drifts back toward Claude over time. The answers should show whether this is a short-term budget correction or a lasting change in how Big Tech deals with AI labs.