Microsoft's Maia 300 delivers 7x token throughput and 2.5x performance per watt over Maia 200
Deep dive into Azure Maia 300 architecture
Azure Maia 300 is Microsoft's third-generation AI accelerator, co-designed with software, networking, and datacenters for cloud-scale agentic inference. It packs 96 accelerators per rack and scales to a 12,288-accelerator superpod. Architectural advances include a multi-chiplet design, NUMA memory, and low-precision MXFP4 formats, yielding 2.5x HBM bandwidth and 33% more capacity than Maia 200. The result: up to 7x rack-scale token throughput and 2.5x performance per watt.
The opportunity ahead is not simply to build more compute. It is to make advanced AI capabilities more accessible, affordable, and useful.