Meta and Microsoft curb Claude use as Anthropic prepares for a blockbuster IPO

Meta and Microsoft, two of Anthropic’s largest corporate customers, are working to cut employees’ use of Claude — an early test of how much of the AI boom survives once companies watch the bill. Citing The Information, Tech Startups reported that Meta employees using Claude Code have fallen to roughly 30,000 from about 60,000 earlier this year.

Microsoft’s cloud and AI group previously allowed individuals to consume as much as $100,000 a month in AI models; that cap has been cut to around $10,000 in most cases. An earlier internal projection that Microsoft was on pace to spend at least $1 billion a year on Anthropic technology for its own use has fallen by more than a third. Total Microsoft payments to Anthropic have remained roughly flat, however, as customer use of Claude through Microsoft products has offset lower internal spend.

Inside Meta, aggressive Claude Code adoption included “tokenmaxxing” on an employee leaderboard dubbed “Claudeonomics,” with more than $105 million spent on Claude Code in one recent 28-day period. Meta is now pushing staff toward internal tools such as MetaCode and Muse Code. The Information also reported that Meta used Claude while building its Muse consumer agent, then switched the finished product to Meta’s own models before launch — underscoring the risk that major customers can build competing stacks.

The pullback lands as Anthropic prepares for an IPO that could value the company above $2 trillion. Reuters has reported that Anthropic’s draft prospectus showed two unnamed customers accounted for 24% of 2025 revenue. Annualized revenue reportedly passed $65 billion in July after climbing more than sevenfold this year. The larger question for Anthropic — and for enterprise AI vendors broadly — is how durable experimental-era token budgets remain once CFOs tighten the meter.

Source: Tech Startups (citing The Information)