Microsoft plans to expand its Azure Maia AI infrastructure to multi-gigawatt deployments starting in 2027, making the system the main platform for running AI models across its products and services.
The expansion includes Maia 300, a new generation of hardware designed to handle longer conversations, faster responses, and AI agents that complete tasks through several steps. Its focus is inference: the process of using an AI model to generate answers or perform work after training.
“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,” said Scott Guthrie, executive vice president, Cloud + AI, Microsoft.
Microsoft expects Maia 300 to produce 40% more tokens per megawatt than leading systems built around graphics processing units, or GPUs. Tokens are the small pieces of information AI models process and generate. The measure indicates how much AI output a system can deliver for the power it uses.
In a configuration with 96 AI accelerators, the chips that process AI workloads, Microsoft says Maia 300 can deliver up to seven times the output of Maia 200 at the rack level. Production deployments will begin in 2027.
Maia 200 already runs Microsoft Copilot workloads and supports the company’s MAI team as it develops new models. GitHub Copilot and Microsoft Foundry are scheduled to follow by the end of the year. The system runs OpenAI models, Microsoft’s MAI models, and leading open-source model families.
Microsoft said Maia 200 is being deployed across the United States, Europe, and Australia. For built-in AI features in Microsoft Teams, Copilot, and Cowork experiences, it delivers up to 1.7 times more tokens per watt than leading GPU infrastructure, according to the company.
Maia 300 builds on that experience with more memory capacity, faster data transfers, and networking that allows multiple accelerators to work together. Its design reduces how much data must move between components, helping the system spend more resources on processing AI requests.
Each Maia 300 rack combines 96 accelerators within a 167-kilowatt power limit. Microsoft says this allows each rack to support more simultaneous AI interactions and larger workloads.
Planned uses include coding assistance in GitHub, running different AI models through Microsoft Foundry, and supporting model development by the MAI team. Maia will also support Dragon Copilot, where faster responses are intended to help scale clinical workflows and reduce clinicians’ administrative work across healthcare settings.