Microsoft has officially released its third-generation self-developed AI chip, Maia 300, with plans to deploy it in production environments by 2027 to support AI inference tasks for products and services such as Copilot. Microsoft claims that compared to current leading GPU systems, the Maia 300 can process 40% more Tokens per megawatt of power; in a full rack configuration consisting of 96 accelerators, its rack-level Token throughput can reach up to 7 times that of the previous-generation Maia 200.
