Microsoft is increasingly using its in-house AI models within applications like Excel and Outlook to reduce dependence on external vendors and cut costs, after previously relying on models from OpenAI and Anthropic. According to a source familiar with internal developments reported by Bloomberg, tens of thousands of AI requests are now processed weekly by Microsoft’s own MAI models, though this accounts for only a small portion of total AI usage, as products like Copilot process many millions of AI prompts each week. Microsoft declined to comment. The strategy has a financial rationale: while Microsoft benefits from favorable terms with OpenAI, it wants to prevent future price increases from external providers. At the Build developer conference in June, Microsoft unveiled seven new MAI models, including MAI-Thinking 1, its first reasoning model, which delivers performance comparable to Anthropic’s Claude Opus 4.6 for programming tasks at lower costs. Mustafa Suleyman, who leads Microsoft’s AI model development, said “Anthropic is extremely expensive, and many organizations are urgently seeking alternatives,” with the goal to completely eliminate costs associated with using Anthropic models. Microsoft is now deploying MAI models within GitHub Copilot and plans to roll out an internally developed transcription model to Teams and other products in the coming months.