2 min read
Add as a preferred source on Google

Microsoft Unveils Maia 300 as Azure Maia Family Targets Primary Inference Role

The third-generation accelerator is scheduled for production deployments in 2027 as Microsoft targets multi-gigawatt Azure Maia deployments.

Gloved hands fit an accelerator module into a computing enclosure / TokenPost.ai
Gloved hands fit an accelerator module into a computing enclosure / TokenPost.ai

Microsoft unveiled Maia 300 on Oct. 12 as its third-generation in-house AI accelerator and said the Azure Maia family will become the primary inference system for Microsoft products and services.

Microsoft plans to begin production deployments in 2027 and scale Azure Maia to multi-gigawatt deployments. The company designed Maia 300 for long-context reasoning, latency-sensitive conversational applications and sustained, multi-step agentic tasks.

A complete rack houses 96 Maia 300 accelerators and can produce as much as seven times Maia 200’s rack-level token throughput. The rack draws 167 kilowatts, offers 27.6 terabytes of HBM and reaches aggregate memory bandwidth of 1.7 petabytes per second.

Every accelerator pairs eight HBM4 stacks with 288 gigabytes of memory and 18 terabytes per second of bandwidth. The compute chiplets use Taiwan Semiconductor Manufacturing Co.’s N2P process node.

Microsoft said Maia 300 increases compute performance by 36%, HBM bandwidth by 2.5 times and scale-up bandwidth by 14% compared with Maia 200. The company also said the new system provides more than 2.5 times Maia 200’s performance per watt.

“We will scale our Azure Maia family to multi-gigawatt deployments starting in 2027, making Maia the primary inference system for Microsoft products and services,” Scott Guthrie, Microsoft executive vice president for Cloud + AI, said.

Microsoft projects a 40% improvement in tokens per megawatt compared with leading GPU systems.

Microsoft’s Maia program combines chip design, software, networking, racks and data centers. Maia 200 is already serving some Microsoft production workloads, including Microsoft Copilot. GitHub Copilot and Microsoft Foundry are scheduled to follow by the end of 2026.

Maia 200 is being deployed in the United States, Europe and Australia. Microsoft said it delivers up to 1.7 times more tokens per watt for certain built-in Teams, Copilot and Cowork experiences than leading GPU infrastructure.

Production deployments of Maia 300 are scheduled to begin in 2027.

Simon Yoon

Reporter

Simon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.

Loading…