Microsoft Introduces Maia 200 AI Chip: A Potential Threat to Nvidia’s Market Dominance
Microsoft has recently unveiled its Maia 200 AI accelerator chip, the most powerful custom silicon to date. The chip is designed to reduce reliance on Nvidia’s expensive GPU offerings and power the next generation of AI workloads across Azure cloud services.
The Maia 200 is built on Taiwan Semiconductor Manufacturing Company’s advanced 3-nanometer process. It packs a whopping 144 billion transistors capable of delivering over 10 petaFLOPS of FP4 performance and approximately 5 petaFLOPS of FP8 performance. The chip features 216GB of HBM3e high-bandwidth memory with 7TB/s of memory bandwidth, specifically optimized for AI inference workloads including large language models and chain-of-thought reasoning.
According to Scott Guthrie, Microsoft’s Executive Vice President of Cloud and AI, the Maia 200 delivers “30% better performance per dollar” than current-generation alternatives. This makes it the company’s most cost-efficient AI processor to date. At 750 watts, it also consumes considerably less power than Nvidia’s Blackwell GPUs, which can exceed 1,200 watts.
The chip is already deployed at Microsoft’s data centers in Iowa and Arizona. It’s running critical workloads including OpenAI’s GPT-5.2 models, Microsoft 365 Copilot, and the company’s AI Superintelligence team projects led by Mustafa Suleyman. Microsoft has announced plans for wider customer availability in the future and has released a software development kit for developers, academics, and AI labs.
While Microsoft claims the Maia 200 outpaces both Google’s latest tensor processing unit and Amazon’s newest Trainium chip in several categories, experts note that the company is unlikely to significantly reduce its Nvidia GPU orders given the chip’s specialized focus on inference rather than training.
Source: CNBC