Microsoft Unveils Maia 300 AI Chip to Reduce Nvidia Reliance
Microsoft is deploying its proprietary Maia 300 AI accelerators at a multi-gigawatt scale to power inference for Copilot and other AI services.
Microsoft announced the Maia 300, its third-generation in-house AI accelerator designed to serve as the primary inference system for the company's products and services. The chip, which begins production in 2027, utilizes a chiplet architecture built on TSMC's N2P process node to improve cost and efficiency. The company plans to scale the Maia family to multi-gigawatt deployments to support services such as GitHub Copilot and Microsoft 365 Copilot.
The Maia 300 is specifically optimized for agentic tasks and long-context reasoning. Microsoft expects the hardware to deliver 40 percent more tokens per megawatt and up to seven times the rack-scale token throughput of the previous Maia 200. This strategic shift aims to reduce the company's dependence on silicon from Nvidia and AMD for AI workloads, although Microsoft will continue to partner with both firms for model training.
Executive leadership emphasized the necessity of owning the hardware stack to match the scale of their models. While the Maia 200 already provided a 30 percent performance-per-dollar improvement over previous fleet hardware, the Maia 300 represents a further leap in infrastructure efficiency.
