Meta and Microsoft Slash Anthropic AI Usage Ahead of IPO

Meta Platforms and Microsoft are significantly scaling back their internal use of Anthropic's Claude AI tools, opting instead for their own proprietary alternatives. At Meta, the number of employees using Claude Code has dropped by nearly 50%, from about 60,000 to roughly 30,000. Microsoft, meanwhile, has cut its projected annual internal spending on Anthropic technology by more than a third, down from an earlier estimate of at least $1 billion.

The shift highlights a broader trend among tech giants to reduce reliance on external AI solutions and develop their own in-house tools. Microsoft began reducing its dependence on Claude Code as early as May, aiming to cut employee token consumption and promote the use of its proprietary AI models. Meta's reduction in Claude usage is partly due to workforce reductions and the rapid maturation of its own AI development tools, including MetaCode and Muse Code.

This pullback comes at a critical time for Anthropic, which is preparing for an initial public offering. The company disclosed in its IPO prospectus that two customers account for approximately 25% of its revenue, widely believed to be Meta and Microsoft. While internal usage is declining, external demand for Claude through Microsoft's enterprise platforms continues to grow steadily.

The situation reflects a broader theme in the AI infrastructure sector, where large companies are increasingly determined to keep their AI tooling in-house. For Anthropic, the challenge will be whether enterprise deployments through external platforms can offset the revenue impact of shrinking internal usage at its largest customers.