Microsoft Azure Maia 300 prototype chip on a blue-lit platform
Microsoft Azure Maia 300 prototype chip on a blue-lit platform
+ AI News

Microsoft unveils Maia 300 to cut its reliance on Nvidia for AI inference

Microsoft plans Maia deployments at multi-gigawatt scale from 2027. Its 40% efficiency claim implies about 29% less power for equal token output, with caveats.

Microsoft has announced Maia 300, its third-generation in-house AI chip, with plans to make the Maia family the primary inference system for its own products and services. Production rollout starts in 2027, The Verge reports. The plan would reduce Microsoft’s reliance on Nvidia and AMD chips for running those AI services.

Inference is the work a model does when it answers a request. Microsoft is targeting that work with a system built for long-context reasoning, conversations that need fast responses, and agents that carry out sustained tasks over multiple steps.

Two performance claims with different baselines

Scott Guthrie, Microsoft’s cloud executive vice president, says the company expects Maia 300 to deliver 40% more tokens per megawatt than leading GPU-based systems. In a full configuration of 96 accelerators, he says it delivers up to seven times Maia 200’s rack-scale token throughput. The Verge quotes both claims.

The power-efficiency comparison uses GPU systems as its baseline. The rack-throughput comparison uses Microsoft’s previous Maia generation. Readers need to keep those baselines separate when judging the size of the improvement.

Microsoft plans to expand the Maia family to deployments at multi-gigawatt scale starting in 2027. Maia 200 already runs production workloads for Microsoft Copilot, GitHub Copilot, Microsoft Foundry and Microsoft’s own AI models, according to The Verge.

A chip designed alongside Microsoft’s models

Maia 300 uses a chiplet architecture that separates compute, memory controller and input/output functions. The Verge reports that it uses TSMC’s N2P process. Microsoft’s hardware team says the design gives more silicon area to compute and lets each function use technology suited to its role.

Microsoft AI chief Mustafa Suleyman describes co-designing Microsoft’s models with Maia as a focus for his team. That connects the hardware plan to the models Microsoft develops itself, as well as the services it operates. Microsoft’s announcement page also presents the chip, server and rack system and identifies Microsoft 365 as a Maia use case.

Sources

The AI Feed Desk

The AI Feed Desk

Editorial desk

The AI Feed Desk tracks AI provider updates, model releases, agent tooling, and enterprise adoption, turning fast-moving announcements into source-linked context for builders and operators.

Noticed a typo, incorrect information, or translation error?

Tell us so we can fix it.

Help Improve This Article

Related Articles

Two rocket demo panels labeled Standard at 17.0 seconds and Ultrafast at 6.4 seconds

GPT-6 Astra Ultrafast arrives on Nvidia Blackwell with up to 8x faster output

Nvidia says Astra Ultrafast is available in the API and to eligible ChatGPT Work and Codex users. Its 8x claim measures token generation against Standard.

The AI Feed Desk

By The AI Feed Desk

Open model blocks move through a scanning lane into secure managed GPU racks

Hugging Face models reach Microsoft Foundry Managed Compute

Hugging Face and Microsoft are putting curated open-weight models on Foundry Managed Compute with weekly refreshes, Azure-staged weights, and scanned runtimes.

The AI Feed Desk

By The AI Feed Desk

Satya Nadella and Jensen Huang standing beside a display of Windows laptops

Microsoft and Nvidia launch RTX Spark PCs, with laptops priced up to $7,000

RTX Spark laptops start at $2,599. The $7,000 model is an Asus, while the buying case against a MacBook Pro depends on memory and the software you run.

The AI Feed Desk

By The AI Feed Desk

Satya Nadella and Jensen Huang beside a display of Windows laptops

How Microsoft is changing Windows for Nvidia's RTX Spark laptops

Microsoft is tuning Windows for RTX Spark's local agents, shared memory and Arm apps. Copilot Plus features and some game support arrive later.

The AI Feed Desk

By The AI Feed Desk

A large AI factory rack sends green revenue tokens toward a smaller cloud node

NVIDIA turns AI cloud capacity into a revenue-sharing model

NVIDIA's new AI cloud model pairs revenue sharing with credit support, giving emerging cloud providers a way to finance AI factories while tying NVIDIA to downstream token demand.

The AI Feed Desk

By The AI Feed Desk