Overview
- Microsoft announced Maia 300 on Monday, October 12, 2026, and said it will begin scaling the Maia family to multi‑gigawatt deployments starting in 2027 to make Maia the primary system for AI inference across its services.
- Maia 300 uses a chiplet architecture with compute chiplets on TSMC’s 2‑nanometer N2P node and other functions on 3nm parts to reduce data movement and increase on-chip compute efficiency.
- The company gave rack specs that house 96 Maia 300 accelerators in about a 167 kW power envelope with very large HBM capacity and bandwidth, and it says the design will deliver about 40% more tokens per megawatt than leading GPU systems.
- Microsoft will continue to rely on Nvidia and AMD for model training and some workloads, but plans to shift most inference traffic to its in‑house Maia accelerators to lower energy and operating costs.
- Independent reporting says Microsoft is negotiating large-volume manufacturing with TSMC for hundreds of thousands to over a million units for 2027 and beyond, though those production figures and the company’s efficiency claims are not yet independently verified.