PANews October 12 news, according to CNBC, Microsoft officially released its third-generation self-developed AI accelerator chip Maia 300 on October 12 and announced that it will use the Maia series as its primary architecture for running AI models, gradually reducing its reliance on Nvidia GPUs.
Maia 300 is planned to enter production in 2027, with each data center rack able to carry 96 chips, a 7x increase in AI token processing volume per rack, and a 40% improvement in token processing efficiency per megawatt compared with GPUs. Microsoft aims to deploy multi-gigawatt-scale Maia chip clusters, and Microsoft 365 Copilot will be the first to benefit from the performance improvements.
Evercore ISI analysts expect that by the first quarter of 2027, Maia chips will account for about 10% of Microsoft's AI chip spending. Notably, Maia accelerators are currently for Microsoft's internal use only and are not available for external rental, while Nvidia and AMD will still participate in Microsoft's model training business.