Microsoft Replaces OpenAI and Anthropic Models With In-House MAI

Microsoft Corporation has begun replacing third-party AI models from OpenAI and Anthropic with its proprietary Microsoft AI (MAI) model family across key applications, including Excel, Word, Outlook, and Teams. The company is currently routing tens of thousands of weekly user prompts through these in-house models to lower high operational costs associated with AI tokens and decrease financial reliance on external providers.

This strategic transition follows the debut of seven new MAI models at the June Build developer conference, including MAI-Thinking 1 for efficient reasoning and a coding assistant designed to match Anthropic's Opus 4.6 at a lower cost. Microsoft is also integrating a proprietary transcription model into Teams and expanding MAI capabilities into GitHub Copilot. The shift is supported by a 2025 renegotiation that ended Microsoft's exclusivity with OpenAI, allowing the company to develop competing technology.

Mustafa Suleyman, CEO of Microsoft AI, characterized Anthropic as extremely expensive and stated that the company's goal is to reduce and eventually eliminate payments to the provider. Despite the move, Microsoft maintains a hybrid strategy, continuing its partnership with OpenAI while selecting between in-house and third-party models based on performance and cost. The pivot occurs as OpenAI reportedly seeks to reduce its own dependence on Microsoft Azure by developing custom AI chips and competing with LinkedIn. Following reports of the transition, Microsoft shares declined 1.52% to $377.52.