Meta and Microsoft Reduce Internal Reliance on Anthropic Claude Models

Meta and Microsoft are scaling back their internal reliance on Anthropic’s Claude models. They are shifting focus toward proprietary AI development and stricter cost controls. This reduction applies to internal operations, while external enterprise demand for Anthropic through commercial platforms remains steady.

Internal Spending and Usage Reductions

Microsoft has curtailed its internal investment in Anthropic’s technology. It lowered its projected annual spending by more than one-third from an initial $1 billion estimate, according to a report from Investing.com. The company also tightened operational budgets by reducing monthly AI usage caps for employees within its Cloud and AI division from $100,000 to approximately $10,000.

Meta has reduced its internal headcount utilizing third-party AI coding assistants. Usage of the Claude Code assistant among Meta employees has fallen from roughly 60,000 earlier this year to approximately 30,000.

Transition to Proprietary Tooling

The shift is driven by a strategy to prioritize internal AI infrastructure. Meta is migrating its workforce toward its own coding solutions. Its proprietary tool, MetaCode, has surpassed 30,000 internal users. The company’s Muse Code tool, which is currently in external testing, has reached more than 6,000 internal users.

"The reduction in Claude usage applies exclusively to internal operations; customer spending on Anthropic models via Microsoft's enterprise platforms continues to see steady growth," according to the Investing.com report.

Why This Matters for AI Infrastructure

For major technology firms, the cost of running large language models internally has become a significant line item as AI deployment scales. By developing proprietary alternatives, companies like Meta can exert greater control over inference costs and data privacy. Firms are maintaining external commercial availability while limiting internal reliance. This allows them to avoid vendor lock-in for their own engineering workflows while still capitalizing on the broader market demand for third-party AI services.