Satya Nadella announced Maia chips will become primary inference system at multi-gigawatt scale; Maia 300 introduced with claimed 40% efficiency gain over GPU-based systems
Following·Day 4·first covered 12 Oct 2026·2 sources
Microsoft to Make Maia Chips Primary Inference System at Multi-Gigawatt Scale
The gist
Microsoft shifting its own products to in-house silicon at multi-gigawatt scale represents a significant move away from reliance on third-party GPU suppliers. The Maia 300's claimed efficiency gains, if realized, would reduce inference costs across Copilot, Teams, and other Microsoft services.
The full picture
Microsoft CEO Satya Nadella announced that the company's Maia chip family will become the primary inference system for Microsoft's products and services, with multi-gigawatt deployments scaling from 2027. The third-generation Maia 300, launching in 2027, claims 40% more tokens per megawatt than leading GPU-based systems and 7x the token throughput of a Maia 200 rack. Maia 200 already serves production traffic for Microsoft Copilot and Teams. The Maia 300 uses a multi-chiplet design separating compute, memory, and I/O onto specialized silicon. Microsoft is reportedly negotiating with TSMC for over 300,000 Maia 300 units for 2027 delivery, though no purchase order has been signed.
How it developed
Sources
Rohan Paul Twitter
Related
The daily email
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free