To reduce heavy reliance on external AI suppliers and improve its profit structure, Microsoft (MSFT) has quietly initiated a major strategic shift in AI. People familiar with the matter revealed that Microsoft is beginning to replace models from OpenAI and Anthropic with its self-developed “MAI” models in flagship office software such as Excel and Outlook. Currently, tens of thousands of AI prompts are being processed by the in-house models each week.

This marks the first time the scale of Microsoft’s in-house MAI model deployment in office software has been disclosed. While MAI models still account for a relatively small portion of Microsoft’s overall AI usage, the move signals critical progress in the race to build low-cost, high-performance AI models and represents a first step toward shedding expensive external model licensing fees.

Following the news, Microsoft shares rose 2% in Tuesday trading, indicating investor approval of this cost-reducing infrastructure adjustment. According to a Bloomberg report, this transition marks the first large-scale departure from external AI suppliers within Microsoft’s core business software.

Cost Pressures Drive In-House Development

Mustafa Suleyman, head of Microsoft’s AI models, stated publicly last month that the company is reducing its reliance on Anthropic by expanding the use of MAI models, with the ultimate goal of eliminating that spending entirely. He said bluntly, “We pay Anthropic a significant amount of money annually, so our goal is to reduce, and eventually eliminate, that cost entirely.”

Currently, Microsoft’s Copilot-branded AI assistant products consume massive volumes of tokens daily. While Microsoft’s long-standing partnership with OpenAI and its $13 billion investment have allowed it to secure large amounts of computing power and model services at preferential rates, the cost advantages of this collaboration will not last indefinitely. Suleyman’s team is working to ensure Microsoft will not be forced to accept the high pricing standards set by leading AI labs in the future.

At the Build developer conference in June, Microsoft unveiled seven new AI models in one go. One of these models reportedly achieves code generation capabilities comparable to Anthropic’s previous-generation programming model Opus 4.6 at a lower cost. This means Microsoft is not only seeking to cut costs in consumer-facing products but also aiming to compete directly with external suppliers on technical capability.

In-House Models Permeate Product Lines

Beyond Excel and Outlook, the deployment map for Microsoft’s in-house models is expanding rapidly.

The MAI models have reportedly also been integrated into GitHub Copilot, the AI-assisted programming service. Furthermore, Suleyman revealed that in the coming months, Microsoft’s in-house speech transcription models will be progressively applied to Teams video conferencing and other products to deliver lower-cost AI services.

This series of adjustments shows Microsoft is systematically shifting a large volume of routine AI computing demands from expensive external APIs to its own end-to-end proprietary infrastructure. For office software handling massive volumes of user queries daily, even minuscule savings per individual call can aggregate into a significant impact on profit margins.

Market Reaction and Future Outlook

The market responded positively to Microsoft’s strategic adjustment. Tuesday’s 2% share price increase not only recovered some of the losses from the previous trading session triggered by Xbox layoff news but also reflected shareholder approval of the company’s pursuit of technological independence and improved unit economics.

In the long run, Microsoft is establishing a clear “off-ramp” through its in-house models, gradually reducing its structural dependence on third-party AI suppliers. While the volume of requests processed by in-house models still represents a low percentage of the overall Copilot ecosystem, this shift establishes a clear path toward AI self-sufficiency. As Microsoft continues to expand the application scenarios for its in-house models, the intelligent experiences driving hundreds of millions of Office users worldwide—whether analyzing data in Excel or organizing emails in Outlook—will increasingly be powered by Microsoft’s own technology.