At its Build developer conference in San Francisco, Microsoft announced a major expansion of its in-house artificial intelligence capabilities with the new MAI model suite, signaling a direct push against long-time partner OpenAI. The move, alongside a deepened collaboration with Nvidia, drew cautious investor reactions, with Microsoft shares dipping over 4% and Nvidia down nearly 0.7% on the day.
The MAI family includes MAI-Thinking-1, a medium-sized reasoning model that competes with leading systems on software engineering benchmarks, and MAI-Code-1-Flash, an efficient coding model already integrated into GitHub Copilot and Visual Studio Code. Notably, both were developed without distillation from third-party models, underscoring Microsoft’s goal to build original foundational AI. A suite of smaller “Aion” models for on-device Windows PC deployment was also introduced, alongside updates to speech, voice, and image generation tools.
The Nvidia partnership expands this vision to a full-stack platform for agentic AI across Windows, Azure, and on-premises setups. New hardware includes RTX Spark PCs (1 petaflop AI performance, 128GB unified memory, launching this fall) and DGX Station for Windows, a deskside supercomputer capable of running trillion-parameter models by Q4. Nvidia’s open models—including Nemotron 3 Ultra, Nemotron 3.5 ASR, and Cosmos 3—are being added to Microsoft Foundry, where they will sit alongside offerings from Anthropic (now running natively on GB300 Blackwell Ultra systems) and OpenAI.
The partnership also covers data infrastructure, with Microsoft Fabric Warehouse accelerated by Nvidia hardware delivering up to 7x faster SQL queries, and the Vera Rubin platform validated for Azure data centers, promising up to 10x inference throughput per megawatt. Microsoft’s Fairwater AI factory in Wisconsin is already live, ahead of schedule, running hundreds of thousands of Nvidia Grace Blackwell systems.
These moves come after a restructuring of Microsoft’s OpenAI tie-up that removed exclusivity on cloud usage, enabling the company to pursue a dual-track strategy: continuing to back OpenAI while building competing models in-house. Analysts see it as a natural evolution in an AI race where major cloud providers seek independence from external model suppliers.
