Microsoft is paving the way for a significant expansion in its Azure Maia system, aiming for multi-gigawatt deployments by 2027. This move will make Maia the central inference system for its wide range of products and services as part of the company’s push into the era of Super Intelligence.
The Azure Maia program, which spans multiple generations, plays a crucial role in Microsoft’s AI infrastructure strategy. Currently, the Maia 200 is actively supporting Microsoft’s production workloads, such as Microsoft Copilot. It also aids the MAI team in developing future models, with GitHub Copilot and Microsoft Foundry expected to come on board by the end of the year. Maia 200 is utilized across various workloads, running models from OpenAI, Microsoft’s in-house MAI models, and prominent open-source model families. The extensive scale of Microsoft’s AI services facilitates a unique co-design of hardware systems and software applications, enhancing optimization for real-world enterprise applications and industry scenarios.
Looking ahead, the Maia 300 is being developed as a comprehensive intelligent system to handle the new AI workloads emerging in today’s technological landscape. These include complex long-context reasoning, latency-sensitive conversational interfaces, and enduring, multi-step agentic tasks. With improvements focused on efficiency, the Maia 300 is projected to achieve 40% more tokens per megawatt than leading GPU-based systems. In its full 96-accelerator setup, it is expected to provide up to seven times the rack-scale token throughput compared to Maia 200.