Microsoft deploys its first homegrown AI chips

Microsoft deployed its first crop of homegrown AI chips, named the Maia 200, in one of its data centers this week, with plans to roll out more of the hardware in the coming months. The company designed the Maia 200 to serve as what it calls an AI inference powerhouse. This means the chip is optimized specifically for the compute-intensive work of running AI models in production, rather than training them. According to Microsoft, the Maia 200 outperforms Amazon’s Trainium chips and Google’s Tensor Processing Units (TPUs) in processing speed. The move comes as cloud providers increasingly turn to their own silicon designs, driven by the difficulty and expense of obtaining chips from external suppliers amid a supply crunch.

Despite launching its own hardware, Microsoft will continue to purchase chips from Nvidia and AMD to meet its infrastructure demands. Microsoft CEO Satya Nadella confirmed that the company plans to maintain its partnerships with these chip suppliers even as it deploys its own silicon. Nadella emphasized that building internal hardware does not mean the company will rely solely on its own designs. “Because we can vertically integrate doesn’t mean we just only vertically integrate,” Nadella said, defining vertical integration as building systems from top to bottom without using wares from other vendors. He added that both Microsoft and its partners are continuing to innovate in parallel, noting that while many observers focus on who is currently ahead, long-term leadership requires sustained innovation.

The Maia 200 chip will support OpenAI’s models running on Microsoft’s Azure cloud platform. However, the first group to use the new hardware will be Microsoft’s internal Superintelligence team, which is led by Mustafa Suleyman. Suleyman celebrated the deployment, noting that his team will be the first to use the Maia 200 as they develop their frontier AI models. Securing access to AI hardware remains a challenge for both paying customers and internal teams alike. By developing its own models, Microsoft is working to perhaps one day reduce its reliance on model makers like OpenAI and Anthropic.

Why it matters

Microsoft is attempting to balance the efficiency of vertical integration with the reality of supply constraints, signaling that while it is building its own infrastructure, it remains dependent on the broader semiconductor ecosystem.