Microsoft Deploys Maia 200 AI Chips in Azure Data Centers

Overview

• Microsoft says Maia 200 units are live in its Iowa and Arizona data centers and are running internal services such as Microsoft 365 Copilot and OpenAI models.
• The Maia 200 is built on TSMC’s 3nm process, contains over 140 billion transistors, and ships with 216GB of HBM3e memory and native support for FP8 and FP4 low-precision formats used for AI inference.
• Microsoft reports performance gains versus competitors and roughly 30% better performance-per-dollar versus its prior fleet silicon, but those comparative claims have not been independently verified in the reporting.
• The Maia program recovered from earlier design and staffing delays to reach production, and Microsoft is already designing a Maia 300 successor while keeping a multi-foundry manufacturing strategy that includes TSMC and a reported Intel arrangement.
• This move follows a wider cloud-industry trend of custom AI accelerators to cut costs and secure supply, a shift that could give Azure more capacity and lower AI serving prices for enterprise customers over time.