Microsoft Unveils Maia 300 AI Chip, Plans Million-Unit Deployment to Cut Nvidia Dependence

Microsoft on Monday unveiled Maia 300, the third generation of its custom AI chip, and said it plans to deploy potentially more than a million of the chips next year to power its internal AI services. The announcement marks the company’s most aggressive push yet to design its own silicon and reduce its dependence on Nvidia.

Microsoft said it plans “multi-gigawatt deployments” of the Maia 300 starting in 2027. The company has said it expects to spend more than $175 billion in capital expenditures this calendar year, much of it on the infrastructure behind its AI products.

“Our biggest bet yet on our own silicon”

Microsoft CEO Satya Nadella called the Maia 300 “our biggest bet yet on our own silicon” in a post on X. In a blog post, Azure chief Scott Guthrie said Maia will become “the primary AI inference system for Microsoft products and services.”

The chip takes a different technical path from Nvidia’s accelerators, which use one massive computing chip surrounded by high-bandwidth memory. The central computing processor in the Maia 300 is built from “chiplets” — specialized chips bonded together — with 2-nanometer technology from Taiwan Semiconductor Manufacturing Co. handling key computing functions and cheaper 3-nanometer chiplets moving data around the processor.

Each server rack holds 96 Maia 300 chips and draws about 167 kilowatts of power, compared with about 142 kilowatts for Nvidia’s most recent GB300 AI server racks with 72 graphics processing units. Microsoft said the new chips will get about 40% more useful computing done per megawatt than “leading GPU-based systems,” though it did not name the systems used for the comparison. A full rack will produce up to seven times as much AI output as the current generation, according to the company, and the chips are tuned for fast-growing workloads such as AI agents that work through multi-step jobs and models that handle long documents or conversations.

“It’s basically delivering more intelligence for every watt, every dollar, and every byte of memory by optimizing across the entire system,” Rani Borkar, president of Azure hardware systems and infrastructure at Microsoft, told Reuters in an interview.

A crowded field, and a long runway

The Maia 300 arrives amid a flurry of custom chips from cloud computing rivals and AI companies such as OpenAI, all aimed at loosening Nvidia’s hold on the AI computing market. Google has started selling its Tensor Processing Units directly to some customers, Amazon has won over AI companies including Anthropic with its Trainium chips, Meta has rolled out its own AI chips, and OpenAI unveiled its first custom chip, built with Broadcom, in June.

Microsoft remains deep in Nvidia’s orbit even as it builds an alternative. Nadella and Nvidia CEO Jensen Huang shared a stage last week to launch Windows PCs built on Nvidia chips, and Nvidia remains a key data center supplier. On the company’s July earnings call, Nadella said Microsoft will be among the first cloud providers to deploy next-generation systems based on Nvidia’s Vera Rubin and AMD’s Helios.

Microsoft’s custom-silicon program has also been playing catch-up: unlike Google’s and Amazon’s chips, Maia is not available for outside customers to rent, and as of this summer the current-generation Maia 200 was running in just two U.S. data centers, according to a report by The Information. Microsoft says the Maia 200, which powers Microsoft 365 Copilot and runs OpenAI models as well as Microsoft’s own, is now being deployed across the U.S., Europe and Australia, with GitHub Copilot and Microsoft Foundry set to move onto it by year’s end. Evercore ISI analysts estimate Maia will account for about 10% of Microsoft’s spending on AI chips by early 2027, CNBC reported. Microsoft said it will share more details next month at its Ignite conference in San Francisco.

Reporting based on coverage by Reuters and GeekWire.