Microsoft Rolls Out New Proprietary AI Chip, Aims for Gigawatt-Level Deployment in the Coming Year
5 day ago / Read about 0 minute
Author:小编   

On October 12, 2026, Microsoft took the wraps off its latest proprietary AI chip, the Maia 300. Leveraging TSMC's cutting-edge 2-nanometer manufacturing process, this chip is outfitted with HBM4 memory, marking a significant leap in technological sophistication. Microsoft has ambitious plans to roll out this chip on a massive, gigawatt-level scale starting in 2027, with a primary focus on AI inference tasks.

Microsoft anticipates that the Maia 300 will deliver a remarkable 40% increase in Tokens processed per megawatt of power when compared to mainstream GPU systems. Moreover, when configured with 96 accelerators, the rack-level Token throughput of the Maia 300 can soar up to 7 times that of its predecessor, the Maia 200, showcasing its superior performance capabilities.