Microsoft pushes in-house AI as Anthropic costs come under scrutiny

08:19

By: Dakir Madiha

Microsoft pushes in-house AI as Anthropic costs come under scrutiny

Microsoft's artificial intelligence chief Mustafa Suleyman has publicly argued that Anthropic's Claude models are too expensive for large-scale enterprise deployment, underscoring Microsoft's growing effort to reduce reliance on external AI providers and strengthen its own model ecosystem.

The comments come as Microsoft prepares to discontinue Claude Code licenses across its Experiences and Devices division by June 30, the end of the company's fiscal year. The division oversees major products including Windows, Microsoft 365, Outlook, Teams and Surface. Thousands of engineers, product managers and designers are being redirected toward GitHub Copilot CLI as part of a broader effort to consolidate development tools.

Microsoft has officially described the move as a toolchain unification initiative. However, reports indicate that escalating token-based AI costs played a significant role in the decision. Ending enterprise licenses before the start of a new budget cycle offers an immediate way to reduce operating expenses while Microsoft evaluates long-term AI spending.

The economics of AI coding assistants have become a growing concern across the technology sector. Enterprise usage of Claude Code has generated monthly costs ranging from $500 to $2,000 per engineer depending on activity levels. One major technology company reported exhausting its annual AI budget within four months after adoption of Claude Code expanded rapidly among thousands of software developers.

Despite reducing its own internal use of Claude Code, Microsoft continues to sell Anthropic-powered products to customers. Earlier this year, the company introduced Copilot Cowork, a service built on Anthropic's models and agent-based architecture. The offering became a centerpiece of Microsoft's latest Microsoft 365 Copilot expansion and is sold through a premium enterprise subscription package.

At the same time, Microsoft is accelerating development of its own AI models. During its recent Build developer conference, Suleyman unveiled seven new MAI models developed internally, including MAI-Thinking-1, which he said achieves performance comparable to leading frontier reasoning models on programming benchmarks. The company has also begun deploying MAI-Code-1-Flash, a coding-focused model integrated directly into Visual Studio Code and GitHub Copilot workflows.

The shift reflects a broader transition across the software industry. After years of experimentation with external AI models, major technology companies are increasingly scrutinizing operational costs and seeking greater control over their AI infrastructure. While Anthropic's models remain available through Microsoft's platforms alongside offerings from OpenAI and Microsoft's own systems, the era of unrestricted enterprise AI licensing appears to be giving way to stricter cost management and vertically integrated model strategies.