Microsoft unveils Maia 300 custom silicon for cloud AI inference workloads

Microsoft has announced its third-generation custom artificial intelligence chip, designated the Maia 300. As The Verge AI and CNBC both report, the processor is designed to serve as the primary AI inference system across Microsoft products and cloud services, reducing the company's reliance on Nvidia hardware. [1][2][3]

According to The Verge AI, the Maia 300 chip focuses on improved operational cost and efficiency for cloud workloads. The enterprise silicon is optimized to handle complex reasoning models and autonomous agentic software tasks. [1]

The Verge AI reports that Microsoft expects the Maia 300 to deliver 40 percent more tokens per unit of performance compared to prior setups. The new hardware architecture will begin rolling out to production AI infrastructure in 2027. [1]

As FTMQ SI reported earlier, Microsoft previously introduced local client AI hardware, including its Surface Laptop Ultra personal computers equipped with Nvidia RTX Spark processors. [8]