Microsoft plans a much bigger role for its own AI chips

Microsoft has introduced Maia 300, the third generation of its in-house AI chip, and says it plans to deploy the Maia family at multi-gigawatt scale in its data centers starting in 2027. These chips are built mainly for inference—the constant work of generating answers, code, and other outputs after an AI model has been trained—rather than for building the model in the first place. That matters because serving services such as Copilot at enormous scale makes electricity, cooling, and chip availability as important as raw computing speed. Microsoft says Maia 300 should let it get more AI output from a fixed amount of data-center power, while giving it a larger alternative to buying the GPUs that still underpin much of the AI industry.