Microsoft on Monday announced Maia 300, the third generation of its custom Azure AI chip, saying the company plans large-scale deployments in 2027 and intends Maia to become the primary AI inference system for Microsoft products and services, GeekWire (opens in new tab) and The Verge (opens in new tab) reported.

In a blog post cited by both outlets, Azure chief Scott Guthrie said: “We expect Maia 300 to deliver 40 percent more tokens per megawatt compared to leading GPU-based systems.” The Verge (opens in new tab) also quoted him saying a full 96-accelerator configuration “delivers up to 7x the rack-scale token throughput of Maia 200,” and that Microsoft will “scale our Azure Maia family to multi-gigawatt deployments starting in 2027.”

CEO Satya Nadella called the move “our biggest bet yet on our own silicon” in a post on X, GeekWire (opens in new tab) reported. Maia 300 is built on TSMC’s N2P process with a chiplet design that separates compute, memory controller and I/O functions, according to The Verge (opens in new tab)’s account of Microsoft’s hardware team.

Maia 200 already runs production workloads including Microsoft 365 Copilot, GitHub Copilot and Microsoft Foundry, GeekWire (opens in new tab) and The Verge (opens in new tab) reported. Microsoft AI CEO Mustafa Suleyman told The Verge (opens in new tab): “We have to work across the full stack, all the way down to the silicon,” and that running Microsoft’s models on Maia is “a critical focus for our team.”

Microsoft still partners with Nvidia and AMD. Nadella said on the July earnings call that Microsoft will be among the first cloud providers to deploy systems based on Nvidia’s Vera Rubin and AMD’s Helios, GeekWire (opens in new tab) reported. Evercore ISI analysts estimate Maia will account for about 10% of Microsoft’s spending on AI chips by early 2027, GeekWire (opens in new tab) reported, citing CNBC.