Microsoft to expand Maia chip for the AI era
Microsoft announced plans to expand its Azure Maia AI inference chip family, scaling deployments to multi-gigawatt levels by 2027 to support core products like Copilot and emerging AI workloads.
Microsoft will scale its Azure Maia family to multi-gigawatt deployments starting in 2027, positioning Maia as the primary inference system for Microsoft products and services. The expansion marks a new phase in Microsoft’s AI infrastructure strategy, with Maia 200 already serving production workloads such as Microsoft Copilot and supporting the MAI team’s development of next-generation models. GitHub Copilot and Microsoft Foundry are expected to integrate Maia 200 by the end of the year, running OpenAI models, Microsoft’s MAI models, and leading open-source families.
Maia 300, designed for emerging AI workloads like long-context reasoning and latency-sensitive tasks, is projected to deliver 40% more tokens per megawatt than leading GPU-based systems. In a full 96-accelerator configuration, Maia 300 aims to provide up to 7x the rack-scale token throughput of Maia 200, reflecting significant efficiency gains in hardware and software co-design.
The Azure Maia system is engineered around Microsoft’s real-world workloads, with silicon and software teams collaborating to anticipate future requirements. Maia 200, already deployed across the United States, Europe, and Australia, delivers up to 1.7x more tokens per watt for AI capabilities in Microsoft Teams, Copilot, and Cowork experiences compared to leading GPU infrastructure.
Maia 300 introduces a chiplet architecture, industry-leading HBM bandwidth per watt, and a software-defined dataflow design to reduce data movement and improve utilization. Production deployments for Maia 300 are scheduled to begin in 2027, with the system designed to support multi-step agentic tasks, cloud-scale model serving, and latency-sensitive interactions across Microsoft’s AI services.