Microsoft Debuts Maia 300 AI Processor for Azure Data Centers
Microsoft introduced the Azure Maia 300, a custom AI inference chip that it plans to deploy in its cloud data centers starting in 2027. The company says a rack with 96 Maia 300 processors can deliver seven times the token throughput of its Maia 200 system and process 40% more tokens per megawatt than leading GPU-based systems. The launch reflects Microsoft's push to rely less on Nvidia and AMD while expanding its AI infrastructure.
Microsoft announced the Azure Maia 300 at the OCP Global Summit 2026 in San Jose. The chip is intended mainly for inference, meaning it runs already-trained AI models to answer queries and power software. Deployment across Azure data centers is planned to begin in 2027.
The rack design holds 96 processors. Microsoft says it offers seven times the token throughput of a Maia 200 rack and 40% more tokens per megawatt than leading GPU systems, plus high memory bandwidth. These figures have not been independently verified. Earlier Maia chips already support Copilot, Microsoft models, OpenAI models, and open-source workloads.
Microsoft's move could affect cloud customers, AI developers, and businesses relying on Azure by potentially lowering inference costs and expanding access to AI services. It may also influence competition among chip suppliers, as Nvidia and AMD face a large customer building alternatives. Greater efficiency per megawatt could ease some data-center power pressures, though overall electricity demand may still rise as AI use grows.