Microsoft is starting to route some AI requests in select products away from OpenAI and Anthropic models and toward its own in-house AI models, according to reports including Bloomberg. The change is described as incremental, not a full switch. In products such as Excel and Outlook, Microsoft is testing configurations where certain tasks are handled by Microsoft’s internal “MAI” models when that better fits priorities like cost efficiency or data residency requirements. Despite the early replacements, OpenAI and Anthropic continue to handle much of the production AI traffic within Microsoft’s Copilot experience. The reported approach suggests Microsoft is selectively choosing which model processes particular parts of user features rather than changing the entire underlying AI stack at once. The move is framed as part of Microsoft’s broader effort to manage costs and maintain control over how AI workloads are served, while still relying on partner models for most current demand.