Microsoft is quietly replacing OpenAI with its own AI in Excel and Outlook

Every week, tens of thousands of AI prompts inside Excel and Outlook are now handled by Microsoft’s in-house MAI model — not by OpenAI’s GPT or Anthropic’s Claude, according to Bloomberg.

That’s still a small fraction of Microsoft’s total AI usage across its products. But it signals a deliberate long-term shift. Over the past year, Microsoft has relied on its deep partnership with OpenAI to power most Copilot features at discounted rates. That arrangement is becoming less comfortable, and Mustafa Suleyman — the company’s AI chief — is building a Plan B.

“We pay Anthropic a lot of money, so our goal is to reduce and eventually eliminate that cost,” Suleyman said at Microsoft’s Build developer conference in June. At the same event, the company unveiled seven new AI models. One of them reportedly matches the coding ability of Anthropic’s previous flagship, Opus 4.6, at a significantly lower price.

MAI is already live in GitHub Copilot, Suleyman confirmed. Microsoft’s own transcription model will roll out to Teams and other products in the coming months.

The broader picture is financial. Copilot chews through enormous volumes of AI tokens across Office, Windows, and GitHub. Every prompt routed to an external provider carries a licensing cost. Every prompt handled by an in-house model brings that expense under Microsoft’s control. Building models that can shoulder production workloads at scale gives the company pricing power and, eventually, independence from the labs it currently pays.

For now, MAI handles only a sliver of the AI prompts flowing through Microsoft’s productivity suite. But the direction is unambiguous: Suleyman’s team is racing to make Microsoft’s own models good enough to render the rented ones unnecessary.