Microsoft has unveiled its latest AI accelerator, the Maia 200, a next-generation chip designed to dramatically improve performance and efficiency for AI model inference, the company announced on January 26, 2026. The new silicon is a successor to Microsoft’s earlier Maia hardware and is targeted primarily at running AI workloads, the process of executing models in real-world applications rather than training them.
Built on Taiwan Semiconductor Manufacturing Co. (TSMC) 3-nanometer process technology, the Maia 200 incorporates more than 100 billion transistors and a redesigned memory and networking architecture that delivers over 10 petaFLOPS of performance at 4-bit precision and more than 5 petaFLOPS at 8-bit precision. These performance metrics make it one of the most powerful first-party AI inference chips deployed by a major cloud provider to date.
The accelerator is explicitly engineered to address large-scale AI inference workloads, meaning it’s optimized for running contextual prompts and serving responses from models rather than the compute-intensive training phase. It’s already being rolled out in Microsoft Azure datacenters in the United States, with deployments in the US Central region near Des Moines, Iowa, and the US West 3 region near Phoenix, Arizona planned to follow.
Microsoft says Maia 200 provides efficiency and cost-effectiveness benefits, with around 30 percent better performance per dollar compared with the company’s previous hardware generations, helping lower the economics of AI service delivery. The chip’s architecture includes high-bandwidth memory, on-chip SRAM and advanced data transport systems designed to keep AI models fed with data at scale.
In addition to raw performance, Microsoft is previewing a software development kit (SDK) for Maia 200, including tools for optimizing models and workloads for the new silicon, with support for popular frameworks such as PyTorch and integration with compiler and kernel libraries.
The Maia 200 initiative forms part of Microsoft’s broader strategy to develop custom AI hardware that integrates with Azure infrastructure, offering alternatives to third-party processors and strengthening the company’s ability to deliver AI-powered services such as Microsoft 365 Copilot and other enterprise offerings.
Microsoft acquires Osmos to strengthen data engineering and AI infrastructure
