The newly unveiled Maia 300 accelerator is manufactured by TSMC on a 2nm process and features eight HBM4 memory stacks with 288 GB of capacity and 18 TB/s bandwidth per chip. Microsoft claims the Maia 300 delivers 40% more tokens per MW than leading GPU based systems and offers up to 7x the rack scale throughput of the Maia 200. The previous generation, Maia 200, is already active in the US, Europe, and Australia, where it provides up to 1.7x more tokens per watt for Teams and Copilot.
Tuesday, October 20, 2026
Story
Microsoft to Deploy 1 Million Maia 300 AI Chips Next Year for Primary Inference
AI5 days ago
Microsoft will make its in house Maia AI chips the primary system for running inference across all its services and products, including Copilot. The company intends to scale its Maia family to multi GW deployments starting in 2027, with reports from Reuters indicating plans to deploy more than 1 million Maia 300 accelerators next year. This transition to proprietary silicon aims to reduce dependence on Nvidia and AMD while optimizing data center margins by moving the majority of internal compute consumption to homegrown hardware.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.