Representative image
It added that Microsoft previously projected spending at least $1 billion per year on Anthropic tools for internal operations, but has scaled back that budget estimate by more than a third. Within its 60,000-plus cloud and AI unit, individual monthly spending ceilings on AI models that previously reached $100,000 per person have been cut to roughly $10,000 in most instances, frustrating some engineers who feel their technical workflow is now constrained. The shift, reported by The Information, highlights a ‘complicated problem’ confronting Anthropic: its most lucrative corporate buyers also double as its fiercest market rivals. Citing earlier internal communications from Parikh, the report says that the company introduced division-wide token budgets and shifted standard defaults toward lower-cost OpenAI models.
Inside Microsoft, top leaders, including Azure Cloud head Scott Guthrie and developer tools leader Jay Parikh, have instructed teams to practice greater discipline with coding tools, monitor employee consumption and prioritise non-Anthropic alternatives, the report said. Microsoft and Meta are reportedly curbing their internal staff’s reliance on Anthropic’s Claude models, attempting to control mounting computing expenses while steering engineers toward their own in-house artificial intelligence (AI) tools. Even as both tech giants continue to offer Claude to external clients through enterprise products and cloud infrastructure, their executive leadership is taking direct measures to limit how much their own developers lean on Anthropic’s systems. Internal managers now receive daily tracking metrics detailing exact employee expenditures and specific models accessed via GitHub Copilot. Despite the internal cutbacks, Microsoft’s commercial partnership with Anthropic remains intact. Microsoft continues to support Anthropic models across Azure and integrates Claude into Copilot productivity applications, including Word, Excel and PowerPoint.
The Information reports that internal usage of Claude Code at the social media company tumbled from roughly 60,000 users earlier this year to around 30,000 recently. While part of that drop was tied to spring layoffs that eliminated roughly 10 percent of Meta’s then 78,000-person staff, the primary catalyst has been the rapid rollout of internal alternatives. An internal-only coding assistant, MetaCode has expanded its user base to more than 30,000 employees. Meta previously utilised around 7,000 weekly active users on the platform in August to gather feedback and refine training runs for its underlying architectures. Its external-facing rival to Claude Code, known as Muse Code, which began testing with third-party clients in August, recently counted over 6,000 employee users. The AI startup’s annualized revenue climbed past $65 billion by the close of July, a rise from approximately $9 billion at the end of 2025, according to reporting from The Information and Reuters.
Meta is also said to be directing personnel toward its proprietary Muse Spark model lineup. Meta has similarly taken a step back from Anthropic’s developer assistant, Claude Code. The internal pullbacks by Meta and Microsoft provide a contrast to Anthropic’s otherwise runaway commercial momentum. You use AI every day. Now get your AI Quotient. Take the AIQ test.