October 12, 2026: Microsoft plans to scale its in-house Azure Maia AI accelerator programme to multi-gigawatt deployments from 2027, as the company expands infrastructure designed to handle growing demand for AI services. The company also confirmed that Maia 200 is already running production workloads and outlined plans for the next-generation Maia 300 system.
Scott Guthrie, Microsoft’s executive vice president for Cloud and AI, said Maia would become the primary inference system for Microsoft products and services as the programme expands. Maia 200 is currently supporting workloads including Microsoft Copilot and the company’s MAI model development team. GitHub Copilot and Microsoft Foundry are expected to follow by the end of 2026. <Cite refs={[“turn248780search0”]} />
Microsoft said Maia 200 is being deployed across the United States, Europe and Australia. The company reports that the system delivers up to 1.7 times more tokens per watt for built-in AI capabilities in Microsoft Teams, Copilot and Cowork experiences compared with leading GPU infrastructure.
The next generation, Azure Maia 300, is designed for demanding inference workloads, including long-context reasoning, conversational AI and multi-step tasks performed by AI agents. Microsoft expects it to deliver 40% more tokens per megawatt than leading GPU-based systems. In a full configuration of 96 accelerators, the company says Maia 300 can provide up to seven times the rack-scale token throughput of Maia 200. <Cite refs={[“turn248780search0”]} />
The Maia 300 system combines compute, memory and networking components to improve how AI workloads use available power and processing resources. It features eight HBM4 memory stacks delivering peak bandwidth of 18 terabytes per second, with a stated thermal design power of 950 watts.
At rack level, a 96-accelerator configuration provides 27.6 terabytes of HBM capacity and 1.7 petabytes per second of aggregate HBM bandwidth within a 167-kilowatt power envelope.
Microsoft said Maia 300 production deployments are scheduled to begin in 2027. The system is intended to support services spanning Microsoft Foundry, GitHub, the company’s MAI model programme and Dragon Copilot.
The expansion reflects Microsoft’s effort to develop specialised AI infrastructure alongside its broader Azure computing platform, with an emphasis on improving inference efficiency as AI services and agent-based applications scale.
source:Microsoft to expand Azure Maia system to multi-gigawatt deployments – The Official Microsoft Blog
