Microsoft has unveiled Maia 200, its in-house AI chip aimed at lowering the cost and improving the efficiency of large-scale AI model deployment as demand for AI workloads continues to rise.
This is the latest chip from its Maia AI family, succeeding the Maia 100 AI chip, which Microsoft first unveiled in 2023. The earlier chip was designed for Azure’s AI workloads and marked Microsoft’s initial effort to build its own chip tailored to AI tasks rather than relying solely on third-party GPUs.
Microsoft said Maia 200 delivers more than three times the FP4 performance of Amazon’s latest Trainium chip and higher FP8 performance than Google’s newest TPU, while offering 30% better performance per dollar than current hardware in its fleet.
Maia 200 will support multiple models, including the latest GPT-5.2 models from OpenAI, and will power services such as Microsoft Foundry and Microsoft 365 Copilot. The chip is part of Microsoft’s broader push to control AI infrastructure costs, diversify beyond GPUs and vertically integrate silicon, software and data center design as AI workloads scale globally.
Built on TSMC’s 3-nanometer process, Maia 200 is optimized for low-precision inference, with native FP4 and FP8 tensor cores, a high-bandwidth memory system and a custom networking architecture designed to keep large models fully utilized.