Microsoft unveils Maia 300 and plans to make its own chips the main way it runs AI from 2027
Microsoft says the third-generation inference chip will produce 40 percent more tokens per megawatt than leading GPU systems, and that deployments will be measured in gigawatts.
Microsoft announced the Maia 300 on Monday. It is the third generation of the company's own AI chip, and Microsoft says it will become the main system for running AI across its products starting in 2027, with deployments measured in gigawatts. The Next Web says that could mean more than a million Maia 300 chips in 2027. The current generation, Maia 200, already runs production workloads including Microsoft Copilot. It serves OpenAI models, Microsoft's own MAI models and open-source models, and Microsoft says GitHub Copilot and Microsoft Foundry will follow by the end of the year. The Maia 300 is built for inference, the work of running trained models, with a focus on long-context reasoning, conversations that need fast replies and multi-step agent tasks. Microsoft expects it to deliver 40 percent more tokens per megawatt than leading GPU-based systems, but did not say which systems it compared against. It says a full rack produces up to seven times the output of a Maia 200 rack. Each rack holds 96 chips and draws about 167 kilowatts. The chips are built from smaller pieces, called chiplets, using TSMC's N2P process, with eight stacks of HBM4 memory each. The Verge says the move makes Microsoft less dependent on Nvidia GPUs.