Microsoft has announced Maia 300, its third-generation in-house AI chip, designed to power its own AI services and reduce reliance on other chipmakers. The chip is aimed at AI inference, which is crucial for services such as Microsoft Copilot and models from its partner OpenAI.
Taiwan Semiconductor Manufacturing Co. (TSMC) will manufacture the Maia 300’s compute chiplets using its advanced 2-nanometer process. Microsoft says the chip will deliver 40% more tokens per megawatt than current leading GPU-based systems.
Microsoft plans to begin deploying Maia 300 chips in its data centers in 2027, as part of a significant expansion of its custom silicon strategy.