Microsoft Unveils New Maia AI Chips. What to Know.

A prototype of Microsoft’s Maia 300 chip, which will fill its newest artificial-intelligence system. (Courtesy Microsoft)

Key Points

• Microsoft unveiled its latest artificial-intelligence chip, the Maia 300, to deploy custom hardware across its data centers.
• The Maia 300 system will be deployed in Microsoft’s cloud-computing facilities beginning in 2027.
• Microsoft expects Maia 300 to compute 40% more tokens per megawatt of power than leading GPU-based systems.

Microsoft on Monday unveiled its latest artificial-intelligence chip, dubbed Maia 300, as part of an effort to deploy custom hardware across its data centers.

The Maia 300 system, which is an entire server rack filled with 96 Maia processors, will be deployed in Microsoft’s cloud-computing facilities beginning in 2027, joining predecessor Maia 200. The company plans to run multiple gigawatts worth of AI workloads on Maia hardware starting next year.

Microsoft stock has risen 11% in 2026, trailing the S&P 500 by a few percentage points. It was a Barron’s stock pick in March.

The expansion of Maia is the latest sign that large AI companies want to expand the use of custom, in-house hardware and lessen their reliance on pricey third-party chips.

The Barron’s Daily

A morning briefing on what you need to know in the day ahead, including exclusive commentary from Barron’s and MarketWatch writers.

Email addressSubscribe

I agree to the Terms of Use, Privacy Notice and Cookie Notice.

I would like to receive updates and special offers from Dow Jones and affiliates.

Graphics processing units, or GPUs, from Nvidia and Advanced Micro Devices have long been the go-to choice for training AI models. With attention turning to inference—using models to solve problems, complete tasks, and power autonomous agents—other types of chips have become an increasingly vital piece of the AI ecosystem.

Microsoft said in a blog post Monday that Maia will be “the primary AI inference system for Microsoft products and services.”

Beyond Microsoft, Amazon.com has a range of custom Trainium chips, and Alphabet’s Google is now on its eighth generation of AI chips, which it calls TPUs. Meta Platforms and OpenAI have only recently started getting into the game.

Microsoft expects Maia 300 to compute 40% more tokens—the measure of inputs and outputs in an AI query—per each megawatt of power than leading GPU-based systems. It also has industry-leading memory bandwidth, the company says.

A more efficient AI system is a selling point for Microsoft’s Azure cloud unit and Foundry AI software platform, which are competing with providers like Amazon Web Services and Google to host companies’ AI workloads.

“Azure Maia is a foundational component of our approach to expand access, accelerate innovation and help customers capture more value from AI,” Microsoft said.

Analysts polled by FactSet expect Azure revenue to grow 43% in fiscal 2027, which ends next June, up from 41% in 2026. That kind of growth is critical if Microsoft wants to justify the projected $189 billion it is investing this fiscal year to fuel its AI ambitions.

Write to Nate Wolf at nate.wolf@barrons.com

Copyright ©2026 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8