Meta cuts Claude users by 50% as Microsoft slashes AI budgets from $100K to $10K per employee ahead of Anthropic IPO

Meta and Microsoft are pulling back on their employees’ use of Anthropic’s Claude, turning two of the AI startup’s biggest corporate customers into an early test of how much of the AI boom survives once companies start watching the bill.

At Meta, the number of employees using Claude Code has fallen to roughly 30,000 from around 60,000 earlier this year, according to The Information. Microsoft has cut its projected internal spending on Anthropic by more than a third after executives told employees to curb Claude use and spend more time with Microsoft’s own AI tools.

“Meta Platforms and Microsoft, two of Anthropic’s biggest corporate customers, are working to cut their employees’ use of Claude, a challenge for the AI firm’s efforts to sustain its explosive growth,” The Information reported.

The changes come at a sensitive moment for Anthropic. The company is preparing for an IPO that could value it at more than $2 trillion, putting its revenue growth, customer concentration and enormous computing costs under greater scrutiny. Anthropic’s draft IPO prospectus showed that two unnamed customers accounted for 24% of its 2025 revenue, Reuters reported.

The pullback is especially striking given how much freedom employees once had to spend.

Microsoft’s cloud and AI group, which employs more than 60,000 people, previously allowed individual employees to consume as much as $100,000 per month in AI models, according to The Information. That limit has now been reduced to around $10,000 per month in most cases.

Earlier this year, Microsoft executives projected the company was on pace to spend at least $1 billion annually on its own use of Anthropic technology. That projection has since fallen by more than a third.

“Microsoft executives earlier this year projected that the company was on track to spend at least $1 billion on its own internal use of Anthropic’s technology,” The Information added, citing two people with knowledge of the spending.

From tokenmaxxing to spending caps

Meta offers an even clearer example of what can happen when companies encourage employees to experiment with AI before imposing financial controls.

Meta employees embraced Claude Code so aggressively that some reportedly began “tokenmaxxing,” competing to demonstrate their AI usage on an employee-created leaderboard called “Claudeonomics.” Meta spent more than $105 million on Claude Code during one recent 28-day period, according to The Information.

Then the economics started changing.

Claude Code users inside Meta dropped from around 60,000 earlier this year to roughly 30,000. Layoffs accounted for part of that decline, but Meta has been pushing employees toward its own AI systems.

MetaCode, an internal coding tool, recently had more than 30,000 users. Muse Code, a Claude Code competitor built around Meta models and being tested with outside customers, had more than 6,000 employee users.

There is another twist. Meta used Claude during development of its Muse consumer AI agent, then switched the finished product to Meta’s own models before launching it last month, according to The Information.

That illustrates a risk Anthropic faces with some of its largest customers: the companies buying Claude are increasingly capable of building competing systems themselves.

The bigger question for Anthropic

The numbers do not suggest companies are abandoning Claude.

Microsoft’s total payments to Anthropic have remained roughly flat near their earlier highs, The Information reported, as growing customer use of Claude through Microsoft products has largely offset falling internal spending. Microsoft continues to make Claude available to employees, too.

Still, the shift exposes a question hanging over the enterprise AI market.

How much of the extraordinary spending seen during the first wave of corporate AI adoption represents durable demand, and how much came from experimentation budgets that companies will eventually tighten?

The issue reaches beyond Anthropic. Companies are discovering that usage-based AI costs can be difficult to predict, especially as employees and AI agents consume huge numbers of tokens. That is forcing finance and engineering teams to pay closer attention to which models employees use and what those models cost.

Anthropic has been one of the biggest beneficiaries of the enterprise AI spending surge. Its annualized revenue reportedly passed $65 billion in July after climbing more than sevenfold since the start of the year.

Now comes a different test.

Selling companies on AI is one thing. Keeping those companies spending at experimental-era levels once the CFO starts checking the meter may prove much harder.

Claude