Microsoft Unveils Maia 300 Inference Chip With Rack-Scale Specs

Microsoft announced on 12 October 2026 its third-generation Azure Maia chip, the Maia 300, designed to run large AI models in its datacenters. The multi-chiplet accelerator pairs compute dies on TSMC's N2P process with eight HBM4 memory stacks delivering 288 GB and 18.0 TB/s of bandwidth. Each rack holds 96 accelerators within a 167-kilowatt envelope, and up to 128 racks can be linked into a 12,288-accelerator superpod. Microsoft claims 40% more tokens per megawatt than leading GPU-based systems, with the comparison baseline not named, and up to 7x the rack-scale throughput of its Maia 200. Maia 300 production deployments are set to begin in 2027, with a multi-gigawatt rollout planned as the primary inference system for Microsoft products.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.



Discussion (0)
Log in to join the discussion and vote.
Log in