Microsoft to rely more on homegrown AI chips, reducing dependence on Nvidia

Microsoft plans to rely mainly on its own Maia chips to handle incoming AI requests, starting with the third-generation Maia accelerator unveiled on Monday, reducing its reliance on Nvidia GPUs. Nadella said Maia 200 delivers 30% better performance per dollar than the latest hardware in Microsoft's fleet, and Maia 300 production starts in 2027 with 96 chips per rack, double the Maia 200 count. Microsoft says Maia 300 can run up to seven times more AI tokens per rack and 40% more tokens per megawatt than GPUs. Evercore ISI estimates Maia will account for about 10% of Microsoft's AI chip spending by the first quarter of 2027, while Nvidia remains a major partner. Patrick Moorhead notes Microsoft's accelerators have not been tested outside the company and, unlike Google's TPUs or Amazon's Trainium, cannot be rented by the hour.

Summary written by Upside AI from the original article. Read the source for full details.

Read the original on CNBC ↗

All news →