Microsoft makes Maia its primary AI inference chip, with Maia 300 in production in 2027

Microsoft said it will mainly use its third-generation Maia AI accelerator, unveiled Monday, to handle incoming AI requests, reducing reliance on Nvidia GPUs.
Microsoft is targeting multi-gigawatt deployments across the Maia chip family, and production of the Maia 300 starts in 2027.
Microsoft expects 96 Maia 300 chips per data center rack, double the Maia 200 count, with up to seven times more AI tokens per rack and 40% more tokens per megawatt than GPUs.
Evercore ISI estimates Maia will represent about 10% of Microsoft's AI chip spending by Q1 2027.
Nvidia remains part of Microsoft's plans, with Microsoft partnering with Nvidia and AMD for model training.
Unlike Google TPUs or Amazon's Inferentia and Trainium, Microsoft's accelerators cannot be rented by the hour to other companies.

Quotes

“We need new architectures that will give us essentially better performance, better use of the power that we have”— Alistair Speirs

“Nobody has tested them outside of Microsoft yet”— Patrick Moorhead