Microsoft Launches Maia 200 AI Chip for Enhanced Inference Performance
Microsoft has unveiled the Maia 200, an in-house AI inference accelerator built on TSMC's 3nm process. The chip features over 140 billion transistors, 216GB of HBM3e memory, and achieves over 10 petaFLOPS of FP4 performance.
It is designed to enhance AI token generation efficiency, offering 30% better performance per dollar compared to the previous generation Maia 100. The Maia 200 supports multiple AI models, including OpenAI's GPT-5.2, and is deployed in Microsoft's US Central datacenter in Iowa, with plans for expansion to Arizona.
The architecture includes a two-tier scale-up network and a redesigned memory system to optimize data movement and inference efficiency. Microsoft aims to reduce reliance on NVIDIA by strengthening its AI chip and software ecosystem, positioning Maia 200 as a competitive alternative in the AI hardware market.
