Microsoft is increasing its investment in self-developed AI chips, planning a large-scale deployment of its Maia series chips to reduce reliance on NVIDIA GPUs and control AI infrastructure costs by enhancing inference efficiency. Microsoft announced that its new-generation Maia 300 chip, specifically designed for AI inference tasks, will be deployed in data centers starting next year. This chip can process 40% more tokens per megawatt compared to mainstream GPU systems, enabling it to handle more AI workloads under the same energy consumption. Unlike Google and Amazon, which open up their self-developed chip computing power to cloud customers, Microsoft currently has no intention of directly leasing Maia chip computing power to external parties. Instead, it prioritizes meeting its own vast AI service demands, transforming hardware efficiency advantages into business competitiveness. Microsoft's self-developed chip strategy was initiated early on, with the previous-generation Maia 200, installed since the beginning of this year, already deployed in data centers across the Americas, Europe, and Asia, supporting AI operations for multiple products including Copilot. In the field of cloud giants' self-developed AI chips, Microsoft's layout (layout) is relatively late. It has chosen to prioritize serving internal AI businesses, with the commercial value of Maia chips primarily reflected in improving AI service operational efficiency, reducing inference costs, and enhancing the profit margins of related businesses.