Microsoft will scale its Azure Maia AI chip family to multi-gigawatt deployments starting in 2027, establishing the in-house hardware as the primary inference system for its products and services, the company announced Oct. 12. The shift aims to reduce reliance on GPUs from Nvidia and AMD by moving production AI workloads to its own silicon. According to The Verge, the strategy is intended to improve the cost and efficiency of running AI workloads in the cloud.
Microsoft is introducing the Maia 300, a third-generation accelerator designed for "agentic tasks," long-context reasoning, and latency-sensitive conversations, according to blogs.microsoft.com. The Maia 300 is a successor to the Maia 200, which was announced in January 2026 and is currently running production workloads for Microsoft Copilot, GitHub Copilot, and Microsoft Foundry, The Verge reports.
Microsoft expects the Maia 300 to deliver 40% more tokens per megawatt than leading GPU-based systems, according to blogs.microsoft.com. In a full 96-accelerator configuration, the system delivers up to seven times the rack-scale token throughput of the Maia 200, The Verge reports. Technical details released by azure.microsoft.com indicate the chip uses a multi-chiplet architecture built on TSMC’s N2P process node, separating compute, I/O, and memory controller functions. The system includes eight HBM4 stacks providing 18.0 TB/s of peak memory bandwidth within a 950-watt TDP.
Maia 300 racks integrate 96 accelerators within a 167-kilowatt power envelope, doubling the accelerator density of the Maia 200, azure.microsoft.com stated. These racks use a hybrid cooling system where liquid cooling removes more than 80% of the thermal load from accelerators and switch ASICs. The networking architecture employs a two-tier system that can scale to a "superpod" of 12,288 accelerators azure.microsoft.com. This setup uses Ethernet-based networking with a custom AI transport to allow accelerators to communicate across racks without involving a host or discrete scale-out NIC.
Microsoft said it will share further details on its strategy and technology adoption trends at the Microsoft Ignite 2026 event in San Francisco next month, according to blogs.microsoft.com.