Microsoft announced on Monday that its next-generation Maia 300 chip, designed for AI inference tasks, will begin deployment in data centers next year, Bloomberg reported. Microsoft expects Maia 300 to process 40% more tokens per megawatt than mainstream GPU systems, aiming to boost AI inference efficiency and reduce operating costs. Unlike Google and Amazon, Microsoft currently has no plans to directly lease Maia chip computing power to external clients, prioritizing its own vast AI service demands.