Microsoft unveils Maia 300, its third-generation custom AI chip for Azure
Microsoft introduced Maia 300, the third generation of its Azure Maia custom accelerator, built to run large language model inference. Microsoft says Maia 300 delivers up to 7 times the rack-scale token throughput of Maia 200 in its full 96-accelerator configuration, and more than 2.5 times the performance per watt. Each accelerator has eight HBM4 stacks with 288 GB of capacity and 18 TB/s of bandwidth, which Microsoft says is 2.5 times the memory bandwidth and 33% more capacity than Maia 200. A rack holds 96 accelerators across 24 trays and draws 167 kW, and a two-tier network links up to 12,288 accelerators into one superpod. Microsoft also says it will expand the Azure Maia system to multi-gigawatt deployments.