News
Microsoft announced October 12, 2026 that Azure Maia 200 is serving production workloads including Copilot, while Maia 300 will begin deployment in 2027 with multi-gigawatt scale planned.
Microsoft announced October 12, 2026 that it will scale its Azure Maia family of custom AI accelerators to multi-gigawatt deployments starting in 2027, positioning Maia as the primary inference system for Microsoft products and services. The announcement, published by Scott Guthrie on the Official Microsoft Blog, confirmed that Azure Maia 200 is now serving production workloads and introduced the next-generation Maia 300 system.
Maia 200 in Production Across Three Continents
According to the October 12 announcement, Maia 200 is currently serving Microsoft Copilot production workloads and supporting the MAI team’s development of next-generation models. The system is deployed across the United States, Europe and Australia. Microsoft reported that Maia 200 delivers up to 1.7 times more tokens per watt for built-in AI capabilities in Microsoft Teams, Copilot and Cowork experiences compared with what the company described as leading GPU infrastructure. GitHub Copilot and Microsoft Foundry are planned to follow by the end of 2026.
The Maia 200 system runs OpenAI models, Microsoft’s own MAI models and leading open-source model families, according to the blog post. Microsoft first introduced Maia 200 on January 26, 2026, describing it as an AI accelerator built for inference with 216 GB of HBM3e memory and 7 TB/s memory bandwidth.
Maia 300 Architecture and Specifications
Maia 300, announced October 12, 2026, is designed for long-context reasoning, latency-sensitive conversational experiences and sustained multi-step agentic tasks. Microsoft stated it expects Maia 300 to deliver 40 percent more tokens per megawatt compared to leading GPU-based systems. In a full 96-accelerator configuration, the company reported Maia 300 delivers up to seven times the rack-scale token throughput of Maia 200.
The Maia 300 system features eight HBM4 stacks providing 18.0 TB/s of peak memory bandwidth within a 950-watt thermal design power (TDP), according to Microsoft’s specifications published October 12. The architecture uses a multi-chiplet design separating compute, memory-controller and I/O functions, with compute chiplets built on TSMC’s N2P process node. Maia 300 rack systems bring together 96 accelerators within a 167-kilowatt power envelope, delivering 27.6 TB of HBM capacity and 1.7 PB/s of aggregate HBM bandwidth across a high-speed scale-up domain, Microsoft reported.
The system is designed with what Microsoft described as a software-defined dataflow architecture and AI-assisted kernel optimization to maximize locality and reduce data movement. Support for advanced low-precision formats, including enhanced MXFP4, is intended to reduce memory and bandwidth requirements for both computation and key-value cache storage.
Multi-Gigawatt Scale and Production Timeline
Microsoft stated that Maia 300 will begin production deployments in 2027, when the company plans to start scaling the Azure Maia family to multi-gigawatt deployments. The announcement described Azure Maia as a foundational component of Microsoft’s approach to what the company termed “Frontier Transformation,” emphasizing model and silicon choice on an open, heterogeneous platform.
Scott Guthrie wrote in the October 12 blog post that insights from the scale of Microsoft’s AI services allow for co-design between hardware systems and software applications to optimize real-world use cases. He stated that experience running AI models and services on Maia 200 directly informed the design of Maia 300. The two generations are designed to support multi-step agentic work for GitHub, cloud-scale model serving for Microsoft Foundry, inference and data generation for the MAI team, and latency-sensitive interactions for Dragon Copilot to reduce clinicians’ administrative burden, according to the announcement.
Microsoft’s efficiency and throughput comparisons are vendor-reported claims based on the company’s internal testing and projections. The announcement referenced a technical paper on Maia 300 architectural details available on the Azure blog. Microsoft Ignite 2026, scheduled for next month in San Francisco, is expected to provide additional information on the company’s AI strategy and technology adoption trends.
Source: blogs.microsoft.com.
Reporting source
Original reporting or announcement. Other cited links provide additional evidence; a source link alone does not establish independent verification.
Be the first to comment