Microsoft is quietly replacing OpenAI and Anthropic’s AI models in Word and Excel with its own to reduce the soaring costs of running generative AI at scale. According to Bloomberg, a portion of AI-powered requests in these flagship Microsoft 365 products now run on Microsoft’s internal family of MAI models. The motivation is straightforward: generative AI remains prohibitively expensive for everyday use, especially across hundreds of millions of users.
This shift marks a notable change in how Microsoft presents its AI architecture. Until recently, the company emphasized that Microsoft 365 heavily relies on OpenAI and Anthropic models. Now it’s quietly redirecting some workloads to its own AI systems. That doesn’t mean Microsoft is cutting ties with third-party providers-the company continues to use OpenAI and Anthropic where their models outperform or fit specific tasks better.
Microsoft’s development of proprietary MAI models for AI in Word and Excel
Microsoft has been accelerating development of its MAI (Microsoft AI) models. At its Build conference, the company unveiled seven new models, including tools for code generation and text-to-image creation. Bloomberg’s report sheds some light on how Microsoft distributes AI processing across MAI, OpenAI, and Anthropic, though the company declined to publicly clarify the exact split.
Why shifting AI workloads to Microsoft’s own models reduces cloud costs
The economics are clear: requests to large language models remain one of the priciest parts of AI infrastructure. Incorporating AI deeply into cloud-based office products magnifies these expenses. Shifting more processing to proprietary MAI models helps Microsoft reduce dependency on external providers and control the cost of features like Copilot in Word and Excel.
Balancing partnerships with OpenAI and Anthropic alongside proprietary AI
There’s also a strategic angle in Microsoft’s complicated relationship with OpenAI. Microsoft has invested roughly $13 billion into OpenAI and has long served as its primary commercial partner, while simultaneously building out its own AI ecosystem within Azure and Microsoft 365. So the partnership persists, but the highest-volume AI tasks in Word and Excel are increasingly handled in-house.
Industry-wide efforts to reduce AI cloud service costs
This trend isn’t unique to Microsoft. Amazon, Uber, Meta, and Accenture have all been reported exploring ways to slash AI service costs after a surge of heavy investment. The AI industry is moving beyond the question of ”whose model is smarter” toward ”who can deliver affordable AI at cloud scale,” especially for enterprise and office applications.
Note: Meta is designated an extremist organization in Russia and banned from operating in the country.