Microsoft has introduced Maia 200, its first custom accelerator built specifically for large-scale AI inference. Designed to improve tokens per dollar and per watt, Maia 200 uses narrow-precision compute, software-managed SRAM, and Ethernet-based scale-up to support next-generation AI services. #AIInfrastructure #DataCenter #EnterpriseIT