Meta and Microsoft Are Pulling Their Own Engineers Off Claude Just as Anthropic Gets Ready to Sell Stock

Meta and Microsoft spent much of this year handing their engineers Claude and watching them run with it. Now both companies are working to pull their own staff back, The Information reported Monday, and the timing could hardly be worse for Anthropic, which is preparing to go public at a valuation reported above $2 trillion.

The two are among Anthropic's largest corporate customers. They are also two of the companies best placed to build a replacement for Claude, and that is exactly what they are doing.

Microsoft tightens the tap

Earlier this year, Microsoft executives projected the company was on track to spend at least $1 billion on its own internal use of Anthropic technology, covering Claude Code, Claude models inside Copilot and Claude Mythos. That projection has since fallen by more than a third, after executives Scott Guthrie and Jay Parikh told employees to curb their Claude use and spend more time with Microsoft's own tools, like GitHub Copilot and OpenAI's models.

The budget change is blunt. Microsoft's cloud and AI group, which counts more than 60,000 people, used to let individual employees burn through as much as $100,000 a month in AI models. That ceiling is now around $10,000 a month in most cases.

Claude has not been banned. A Microsoft spokesperson said the company has been steering staff toward GitHub Copilot, but engineers can still choose other models, and Claude remains available internally. There is also a twist: Microsoft's total payments to Anthropic have stayed roughly flat near earlier highs, because customers using Claude through Microsoft products have offset the falling internal spend.

Meta's Claude habit, and its cure

At Meta, the enthusiasm got almost comical. Employees embraced Claude Code so hard that some were "tokenmaxxing" on an employee-built leaderboard called "Claudeonomics." In one recent 28-day stretch, Meta spent more than $105 million on Claude Code alone.

The company had been on track to spend billions this year on internal AI tools, with Claude Code the most popular, before it imposed token limits, spending controls and monitoring. Claude Code users at Meta have since dropped from about 60,000 earlier this year to about 30,000. Layoffs explain part of that. Most of it came from pushing staff onto in-house alternatives.

Those alternatives are growing fast. MetaCode, which is internal only, recently passed 30,000 users. Muse Code, a Claude Code competitor built on Meta's own models and being tested with outside customers, had more than 6,000 employee users. Meta even used Claude while building its Muse personal AI agent, then switched the finished product to its own models before launching it last month. Meta did not comment.

The worst possible moment

Anthropic is heading toward a listing that could land around the holidays, with a pre-Thanksgiving IPO target at up to $2 trillion. Its annualized revenue reportedly passed $65 billion in July. But its draft prospectus also showed that two unnamed customers accounted for 24% of its 2025 revenue, a concentration that makes any pullback by a big buyer hard to ignore.

The prospectus already carries other warnings. On the Pentagon's supply chain risk designation, it cautions that "The company may experience material revenue losses or business disruptions attributable to these events."

The bigger worry is structural. Anthropic sells the most coveted coding tool in the industry to the very companies with the money, talent and models to replicate it. For its biggest clients, the startup is turning from partner into competitor. Meta and Microsoft are now acting like it, just as Anthropic asks the public to put a price on its growth.