Microsoft's MAI Self-Developed Model Implemented in Office Applications
Insiders reveal that Microsoft Office applications such as Excel and Outlook have begun using the self-developed MAI model to handle AI requests, currently completing tens of thousands of calls weekly, previously relying mainly on OpenAI and Anthropic models.
MAI's current share is still small, but it marks Microsoft's acceleration in deploying self-developed models in core products; MAI has been integrated into GitHub Copilot, and a self-developed speech transcription model will also be used in Teams in the coming months.
ABAB AI Insight
Microsoft AI head Mustafa Suleyman previously set a clear goal to reduce and ultimately eliminate spending on Anthropic models. The implementation of MAI in the office suite continues its strategy of building a self-sufficient AI stack, similar to the gradual internalization of core capabilities after the partnership with OpenAI expires.
On the capital path, self-development reduces inference costs and enhances bargaining power with partners, while strengthening ecosystem control in products like GitHub Copilot and Teams, thereby reducing external dependency risks.
Similar to Google's shift from reliance on external models to the deployment of its self-developed Gemini, large model cloud service providers are currently transitioning from cooperative distribution to a self-sufficient closed loop, with office and developer tools becoming key battlegrounds.
Essentially, this is about capital concentration: Microsoft achieves cost control and data closure through vertical integration of self-developed models, with the mechanism being the gradual replacement of external calls, strengthening product pricing power and providing a return path for long-term AI infrastructure investment.
ABAB News · Cognitive Law
Relying on external sources is like borrowing a knife; self-developed models are your own gun.
Weak bargaining at the end of partnerships; self-development reduces costs; whoever controls inference controls the lifeblood.
The entry point for office applications is small, but the ecological barriers are deep; one step replacement leads to a ten-year lead.