Microsoft Unveils Maia 300 AI Chips to Cut Dependence on Nvidia

Microsoft has unveiled its third-generation Maia 300 custom artificial intelligence processor to handle AI workloads across Azure cloud and Copilot services, curbing reliance on Nvidia. The company plans multi-gigawatt deployments, including over one million chips next year, as its calendar-year capital expenditures are projected to exceed $175 billion. The chiplet architecture uses TSMC's 2-nanometer and 3-nanometer processes, packing 96 chips per rack to deliver 40% more compute performance per megawatt than leading GPU systems. Full production begins in 2027, with Evercore ISI projecting Maia will account for roughly 10% of Microsoft's AI chip spending by Q1 2027.