Microsoft Accelerates AI Self-Sufficiency with In-House Models and Chips

Microsoft is accelerating AI self-sufficiency, reducing reliance on third-party models while maintaining its partnership with OpenAI through 2032, including post-AGI models. The company rolled out an in-house mixture-of-experts model in August 2025, pre-trained and post-trained on about 15,000 NVIDIA H100 GPUs, for select Copilot text tasks. Microsoft is also shipping the Maia 200 chip to improve inference economics and compete with CUDA, and is co-hosting and integrating models from xAI, Meta, Mistral, Black Forest Labs, and Anthropic into its Azure and Office platforms.