Microsoft unveils Maia 300 AI chip for 2027 rollout

Microsoft unveiled Azure Maia 300 on October 12, expanding its push into custom AI chips with plans for production deployments in 2027. The company said the Maia family would become its primary inference system for Microsoft products and services as it scales deployments to multiple gigawatts.

Inference is the work of running an AI model to produce responses. Microsoft said Maia 200 is already serving production workloads, including Microsoft Copilot. Maia 300 is designed for longer-context reasoning, conversational applications and AI agents that carry out multiple steps.

The new system packs 96 accelerators into a rack with a 167-kilowatt power envelope, according to Microsoft. Its technical announcement describes compute chiplets built on TSMC's N2P process and eight HBM4 memory stacks delivering peak bandwidth of 18 TB/s per accelerator.

Microsoft says the full rack configuration can deliver up to seven times the token throughput of Maia 200 and more than 2.5 times its performance per watt. These are company-reported performance claims.

Reuters places the launch within a broader effort by cloud companies to loosen Nvidia's hold on AI computing. Maia gives Microsoft another way to supply its own services.

Disclosure: Newsah's editorial process uses OpenAI's Codex models.