Microsoft Pushes Its Own Models in Excel and Outlook

According to Bloomberg, Microsoft is progressively replacing OpenAI and Anthropic models with its own MAI (Microsoft AI) family in some services, notably Excel and Outlook. Tens of thousands of weekly queries previously handled by third-party models are now supported by in-house technologies. This marks a step in Microsoft's AI strategy: long seen as OpenAI's main supporter and financier, the company now wants to capitalize on acquired expertise to reduce its technological and financial dependence on partners. Since 2019, Microsoft has invested over $13 billion in OpenAI, including a major $10 billion injection announced in early 2023, gaining privileged access to advanced models integrated into Azure, Bing, and the Copilot line. During testimony in May, a Microsoft executive indicated that the cumulative cost of the partnership, including investments and infrastructure, exceeded $100 billion. The collaboration allowed Microsoft to develop expertise in training, optimizing, and deploying large language models. Last month, Microsoft presented models in the MAI family, including MAI-Thinking 1, its first reasoning model with 35 billion active parameters, designed to balance performance and inference costs. Other MAI models target image generation, transcription, speech recognition, and coding assistance. Microsoft already uses these models for some queries in Excel and Outlook.