Microsoft starts replacing OpenAI inside its own products

Facts July 24, 2026 · Reading July 24, 2026

Microsoft put two in-house generative AI models into Microsoft Foundry and said it has begun moving workloads including Bing Image Creator, PowerPoint image features, OneDrive image editing, Dynamics 365 Contact Center and Azure Voice Live from OpenAI models to these MAI models. It said the shift cut GPU costs by up to 84% against GPT-Image-2.

This is likely to matter less as one more model launch than as a purchasing decision inside Microsoft itself. When Microsoft moves named, user-facing services off OpenAI and onto its own models, it appears to show that model supply is becoming substitutable inside a large platform as soon as quality is good enough and operating cost falls far enough. That reading likely sharpens the pressure on OpenAI in the one place where its position once looked safest: inside Microsoft's estate.

The move also appears to change what "AI at Microsoft" means for customers. If the company can swap the model layer beneath products such as PowerPoint, OneDrive and Dynamics while keeping the application in place, the customer relationship is likely to sit even more with distribution, workflow and cloud control than with allegiance to a single model provider. Codeberg separately changed its terms to ban use of its services and hosted code as training data for generative AI and to allow excluding AI-generated mass projects, citing load from AI crawlers and the energy and hardware demands of large language models. That is likely to show the same market from the other side: as more value is pushed into applications and platforms, the fight over who may extract data and computing resources grows more explicit.