Microsoft has unveiled the Maia 300, its third-generation custom AI chip, with production deployment scheduled to begin in 2027. The processor is designed to support AI inference tasks for services such as Copilot.
The chip utilizes TSMC's N2P process and a chiplet architecture. Microsoft claims the Maia 300 offers a 40% increase in token processing efficiency per megawatt compared to leading GPU systems.
In a 96-accelerator rack configuration, the Maia 300 delivers up to 7 times the token throughput of the previous Maia 200 model. The design focuses on long-context inference and low-latency real-time tasks.
Microsoft aims to scale the Azure Maia series to gigawatt-level deployments by 2027. This strategy is intended to reduce reliance on external suppliers like Nvidia and AMD while optimizing costs and energy efficiency.
The company will continue to use a mix of hardware platforms for various training and inference needs. The Maia series is being integrated into the broader AI stack to ensure tighter synergy between custom models and infrastructure.
