Microsoft Unveils Maia 200 Custom AI Inference Chip on TSMC 3nm

Summary

Microsoft announced Maia 200, a custom AI inference accelerator built on TSMC's 3nm process, designed to reduce its reliance on Nvidia for AI workloads in Azure data centers. The chip represents the most performant first-party silicon from any hyperscaler, with 140+ billion transistors, 10+ petaFLOPS in FP4 precision and 5+ petaFLOPS in FP8, 216GB HBM3e memory at 7 TB/s bandwidth, 272MB on-chip SRAM, and a 750W power envelope. Microsoft claims Maia 200 delivers 30% better performance per dollar