Microsoft has introduced its new AI chip, the Azure Maia 300. This move aims to decrease the company's dependence on chips from competitors like Nvidia and AMD. The Maia 300 was revealed at the OCP Global Summit in San Jose. It will augment the existing Maia 200 and Maia 100 chips in large-scale deployments starting next year. Microsoft intends for its Maia chips to be the main AI inference system for its products and services. The Maia 300 reportedly offers 40% more tokens per megawatt than leading GPU systems. A rack of 96 Maia 300 chips can achieve seven times the token throughput of the Maia 200. Currently, the Maia 200 powers Microsoft's internal AI workloads and OpenAI models. These include Microsoft 365 Copilot and other proprietary and open-source models. The Maia 200 also provides significantly better token efficiency for AI features in Microsoft Teams and Copilot.