Microsoft announced the launch of its third-generation in-house AI chip Maia300, planned to enter production workloads in 2027, with the goal of becoming the primary AI inference system for Microsoft's products and services. The new chip focuses on improving the performance and energy efficiency of cloud AI inference, and is optimized for long-context reasoning, low-latency conversations, and multi-step agent tasks.
Microsoft Cloud and AI Group Executive Vice President Scott Guthrie said that compared with leading GPU systems, Maia300 is expected to increase token delivery per megawatt by 40%; in a full 96-accelerator configuration, its rack-level token throughput will reach 7 times that of the previous-generation Maia200. Microsoft plans to expand the deployment scale of the Azure Maia series to the hundreds-of-gigawatts level starting from 2027, gradually increasing the share of in-house chips in its internal AI infrastructure.
Maia300 is the successor to Maia200, which was announced in January this year. Maia200 is already used in production-grade AI workloads such as Microsoft Copilot, GitHub Copilot, Microsoft Foundry, and Microsoft's in-house AI models. Guthrie said that Maia200 is currently among the best-performing infrastructure in Microsoft's chip cluster as measured by actual AI service metrics, and Maia300 will further adapt to the needs of next-generation AI applications.
As inference models and agent tasks continue to drive growing compute demand, Microsoft is using in-house chips to improve inference efficiency, reduce infrastructure costs, and strengthen its control over the AI compute supply chain.