Microsoft Introduces Maia 200 AI Inference Chip Built on 3nm Process

Microsoft has launched its latest AI chip, the Maia 200, announced on its official blog. The AI inference accelerator is built on TSMC’s 3-nanometer process and includes native FP8 and FP4 tensor cores, 216GB of HBM3e memory, and 272MB of on-chip SRAM. Each chip delivers over 10 petaflops in 4-bit precision and 5 petaflops in 8-bit precision, offering a 30% improvement in performance per dollar compared to Microsoft’s previous hardware.

The Maia 200 is being deployed in Microsoft’s U.S. Central data center near Des Moines, Iowa, with expansion planned for the U.S. West 3 region in Arizona. The chip integrates natively with Azure and supports a new software development kit (SDK) that includes PyTorch integration, a Triton compiler, and a low-level programming language for fine-grained control.

The Maia 200 will support multiple workloads, including GPT-5.2 models from OpenAI and Microsoft Copilot. It will also be used by Microsoft’s Superintelligence team for synthetic data generation and reinforcement learning. Developers, startups, and academics can now sign up for early access to the Maia SDK preview.