Microsoft unveiled its third-generation Maia 300 AI accelerator on October 12 and said production deployments will begin in 2027. The company plans to scale its Azure Maia systems to multi-gigawatt deployments for Microsoft products and services, including AI workloads behind Copilot and Azure.
Maia 300 is designed for long-context and agentic workloads
Microsoft designed Maia 300 for inference-heavy tasks, including long-context reasoning, latency-sensitive conversational experiences and sustained, multi-step agentic work.
Microsoft says Maia 300 should deliver 40% more tokens per megawatt than leading GPU-based systems. Reuters reported that Microsoft plans to deploy potentially more than one million of the chips next year for its internal AI services. The company’s own announcement described the broader expansion as multi-gigawatt deployments.
One Maia 300 rack combines 96 accelerators and 27.6 TB of HBM
A full rack configuration uses 96 accelerators within a 167-kilowatt power envelope. Microsoft says the rack provides 27.6 terabytes of high-bandwidth memory and 1.7 petabytes per second of aggregate HBM bandwidth.
The accelerator uses a multi-chiplet design. Microsoft says its compute chiplets are built on Taiwan Semiconductor Manufacturing Co.’s N2P process, while the architecture separates compute, memory-controller and input/output functions.
Azure Maia 200 already serves Microsoft production workloads
Maia 300 follows Maia 200, which Microsoft says is already serving production workloads in the United States, Europe and Australia. Those workloads include Microsoft Copilot, Microsoft’s own MAI models, OpenAI models and leading open-source model families.
Microsoft says GitHub Copilot and Microsoft Foundry are scheduled to follow by the end of 2026. The company also says its experience operating Maia 200 informed the design of Maia 300.
Azure customers will encounter Maia 300 through hosted services
Maia 300 is a data-center accelerator rather than a consumer product or local developer download. Users and developers are therefore likely to encounter it through the performance, capacity and economics of Microsoft-hosted AI services after production deployments begin.
Microsoft says Maia 300 will deliver up to seven times the rack-scale token throughput of Maia 200 in a full 96-accelerator configuration. The company’s stated target is to make Maia its primary inference system for Microsoft products and services as the program expands.
References
Nellis, S., & Cherney, M. A. (2026, October 12). Microsoft plans huge deployment of its own AI chips, with an eye on saving costs. Reuters. https://www.investing.com/news/stock-market-news/microsoft-plans-huge-deployment-of-its-own-ai-chips-with-an-eye-on-saving-costs-4944480
Guthrie, S. (2026, October 12). Microsoft to expand Azure Maia system to multi-gigawatt deployments. Microsoft. https://blogs.microsoft.com/blog/2026/10/12/microsoft-to-expand-azure-maia-system-to-multi-gigawatt-deployments/