Microsoft has unveiled the Maia 300, a 2nm chiplet-based AI accelerator designed to serve as the primary platform for the company's AI services and products. The hardware is expected to begin deployment in 2027.
The new chip is built using the TSMC N2P process, utilizing a chiplet architecture that separates computing, memory management, and I/O functions. This design allows for optimized allocation of silicon area for specific tasks.
According to Scott Guthrie, executive vice president of Microsoft's cloud division, the Maia 300 is projected to deliver 40% more tokens per megawatt compared to leading GPU-based systems. A 96-accelerator rack configuration offers up to seven times the throughput of the previous Maia 200 model.
The Maia 300 is tailored for complex AI workloads, including long-context processing, logical reasoning, and latency-sensitive conversational systems. It will support the company's internal AI models and agents.
Microsoft plans to scale the Azure Maia family to gigawatt-level systems starting in 2027. This initiative aims to reduce reliance on third-party GPU providers like Nvidia and AMD by transitioning to proprietary hardware.
