Microsoft / The Verge / Reuters
Important
MicrosoftChipsAzureInfrastructureMicrosoft unveils Maia 300 — own silicon to be primary inference from 2027
October 12, 20262 min read
Microsoft announced Maia 300, its third in-house AI accelerator (TSMC N2P chiplets, 288 GB HBM4, 18 TB/s bandwidth), and said multi-gigawatt Maia deployments will make Maia the primary inference system for Microsoft products from 2027 — targeting ~40% more tokens per megawatt than leading GPU systems and up to 7x rack throughput vs Maia 200.
Why it matters
Nvidia still owns training GPUs, but Microsoft is locking in inference on its own silicon at scale. That is a direct cost and supply play for Copilot, OpenAI workloads on Azure, and agent traffic — and a signal every hyperscaler will push harder on custom chips.
Microsoft announced Azure Maia 300 on October 12 and said it will expand the Maia system to multi-gigawatt deployments starting in 2027, making Maia the primary inference system for Microsoft products and services.
Maia 300 uses a multi-chiplet design with compute on TSMC’s N2P process, eight HBM4 stacks (about 288 GB, 18 TB/s bandwidth), and ~950 W SoC power. Racks hold 96 accelerators in a ~167 kW envelope. Microsoft claims up to 7x rack-scale token throughput versus Maia 200 and about 40% more tokens per megawatt than leading GPU-based systems, aimed at long-context reasoning, latency-sensitive chat, and multi-step agent workloads.
Maia 200 is already in production powering Copilot and other Azure AI services. Nadella called Maia “our biggest bet yet on our own silicon.” The move reduces reliance on Nvidia for inference while Microsoft remains a top GPU buyer for training and peak capacity.