Microsoft Launches Maia 300 AI Chip to Enhance In-House AI Capabilities

Microsoft has introduced the Maia 300, its third-generation custom AI chip, aimed at serving as the primary AI inference system for its products and services. The company claims that the Maia 300 will deliver 40% more tokens per megawatt compared to leading GPU-based systems, enhancing energy efficiency in data centers.

Scott Guthrie, executive vice president of Microsoft’s cloud group, highlighted that the chip is expected to significantly increase AI output while managing electricity costs, a crucial factor for data center operations. The Maia 300 is set for production rollout in 2027, with Microsoft negotiating with TSMC to manufacture over 300,000 units, potentially scaling to over a million units in the long term.

The Maia 300 is built using a chiplet architecture on TSMC’s N2P process node and is designed for inference workloads such as conversational AI. It aims to support Microsoft 365 Copilot, GitHub Copilot, and various models developed by OpenAI and Microsoft.

Microsoft's strategy to develop its own chips is part of a broader trend among major tech companies to reduce reliance on single suppliers like Nvidia, while also aiming for long-term cost savings. The success of the Maia 300 will depend on its deployment speed, efficiency in real-world applications, and potential adoption by external customers.