Microsoft is quietly shifting the AI powering its biggest productivity tools. Inside Excel and Outlook, some of the AI tasks previously handled by OpenAI and Anthropic models are now being processed by Microsoft’s own in-house models, called MAI. Bloomberg reported on this transition on July 7, 2026.
This is not a visible change for users. Microsoft confirmed it is a server-side update — no toggle, no setting to adjust. The AI that helps summarize emails, generate formulas, or draft responses is simply routing to different models under the hood.
At Microsoft’s Build 2026 conference in June, the company introduced seven MAI models covering reasoning, coding, image generation, speech, and transcription. The lineup includes MAI-Thinking-1 for complex reasoning tasks, MAI-Code-1-Flash for software development and GitHub Copilot integration, MAI-Image-2.5 Pro for image tasks, MAI-Voice-2 Flash for voice features, and MAI-Transcribe-1.5 for audio transcription.
The company’s stated reason is straightforward. One of its AI leaders said: “We pay a lot of money to Anthropic — so our goal is to reduce and ultimately eliminate that cost.” Microsoft processes enormous volumes of AI requests across its suite of products, and every request costs money when it routes through a third-party model. Internal models reduce that cost once development expenses are covered.
On performance, Microsoft claims one of its coding models matches the capability of Anthropic’s Claude Opus 4.6 at a lower cost. A model tuned specifically for consulting firm McKinsey reportedly beat OpenAI’s GPT-5.5 on cost efficiency by a factor of ten.
Importantly, Microsoft is not cutting its ties with OpenAI or Anthropic. The company describes its approach as building a multi-model platform where different models are selected based on the task. External models remain available. But the internal MAI models are becoming the default for the highest-volume operations in Microsoft 365 applications.
The strategy mirrors what other large tech companies are doing. Google has built Gemini for internal and external use. Meta has developed the Llama family of open models. Each found that at scale, depending entirely on third-party models becomes expensive enough to justify significant internal research investment.
For the broader AI industry, Microsoft’s shift raises a real question about the long-term revenue outlook for companies that depend on selling model access to large enterprise customers.