Microsoft Replaces OpenAI and Anthropic With Proprietary AI Models

Microsoft is shifting its artificial intelligence strategy by integrating its own internally built MAI models into key software products, including Excel and Outlook. The move marks a strategic effort by the tech giant to reduce its reliance on external AI providers like OpenAI and Anthropic and lower the substantial costs associated with AI computing.

Expanding Internal Model Usage

According to a person familiar with the internal developments, tens of thousands of AI prompts within Excel and Outlook are now being processed by Microsoft’s proprietary MAI models each week. While these internal models currently account for a small portion of the company's total AI usage, the transition represents a significant shift for applications that previously relied heavily on technology from OpenAI and Anthropic.
Microsoft’s AI model chief, Mustafa Suleyman, has been vocal about the company’s objective to curb spending on third-party services. In June, Suleyman stated that the company aims to reduce and eventually eliminate costs associated with Anthropic by prioritizing the use of MAI models.

Managing AI Infrastructure Costs

The financial incentive for this transition is tied to the massive volume of AI tokens consumed by products like the Copilot workplace assistant. Although Microsoft currently benefits from discounted rates through its partnership with OpenAI, the company is preparing for a future where it is not dependent on the pricing structures set by leading AI labs.
To support this transition, Microsoft unveiled seven new AI models at its annual Build conference in June. Among these is a model designed to match the coding capabilities of Anthropic’s popular Opus 4.6 model, but at a lower operational cost.

Future Integration Plans

The deployment of MAI models is not limited to spreadsheet and email software. These models are already being utilized within GitHub Copilot, the company’s AI-assisted development service. Furthermore, Microsoft plans to expand the use of its own technology to other areas of its ecosystem.
Suleyman has indicated that a Microsoft-built model for transcription is scheduled to be implemented within the Teams videoconferencing application and other products in the coming months. As these internal models continue to scale, they are expected to play an increasingly central role in Microsoft’s broader software suite.