Key Takeaways
- The Microsoft Maia AI chip reached its third generation with the Maia 300, unveiled at the OCP Global Summit in San Jose on Monday.
- The chip is built mainly for inference, which means running trained AI models for everyday requests.
- Microsoft claims about 40% more tokens per megawatt than leading GPU-based systems but did not name the systems it compared.
- A rack of 96 Maia 300 chips draws about 167 kilowatts, while Nvidia’s GB300 rack of 72 GPUs draws 142 kilowatts.
- Microsoft plans multigigawatt deployments starting next year, with potentially more than one million chips.
- Nvidia and AMD processors will still matter because customers want different chips for different workloads.
The Microsoft Maia AI chip has a new generation. The company revealed the Maia 300 on Monday at the OCP Global Summit in San Jose. The chip is part of Microsoft’s plan to depend less on processors from Nvidia and AMD. This guide explains what was announced, how the hardware works, and what the numbers mean in plain terms.
AI chips can sound technical. The core idea is easy to follow. Microsoft wants to run more of its own AI services on hardware it designs. That could help the company manage costs and reduce its reliance on a single supplier.
Microsoft Unveils the Third-Generation Maia Chip
Microsoft introduced the Azure Maia 300 as the third generation of its custom AI chips. It joins the Maia 100 and Maia 200 as part of what the company calls multigigawatt deployments that begin next year. Microsoft says the Maia family will become the primary AI inference system for its products and services. Inference is the stage where a trained model answers real requests from users.
The Maia 300 is designed for jobs such as coding and carrying out complex business tasks on a user’s behalf. Its predecessor already runs Microsoft’s in-house AI models. It also runs OpenAI models that Microsoft delivers through its Azure cloud service. The newer chip builds on that existing work.
The announcement also fits a larger spending picture. Microsoft has said it expects to spend more than $175 billion in capital expenditures this calendar year. That figure gives context to the Maia 300 launch. Microsoft is investing heavily in AI infrastructure, and its own chips are one piece of that effort.
How the Microsoft Maia AI Chip Is Built
Nvidia’s AI accelerators typically use one massive computing chip surrounded by high-bandwidth memory chips. Microsoft took a different technical path with the Maia 300. Its central processor is built from chiplets, which are small specialized chips bonded together.
The chiplets that handle key computing functions use 2-nanometer technology from Taiwan Semiconductor Manufacturing Co. Other chiplets use a cheaper 3-nanometer process. Those chiplets handle jobs such as moving data off the processor and elsewhere inside the system.
A smaller nanometer number generally points to a more advanced manufacturing process. Microsoft uses the newest process for the central computing work. The 3-nanometer process is the cheaper of the two options. This split reserves the most advanced technology for the parts of the chip that need it most.
Racks, Power, and the 40% Efficiency Claim
Maia 300 processors go into racks that hold 96 chips. Each rack uses about 167 kilowatts of power. Nvidia’s most recent GB300 server racks hold 72 of its graphics processing units and use 142 kilowatts. The Maia rack holds more chips and draws more power.
Microsoft says the design still comes out ahead on efficiency. It claims about 40% more tokens per megawatt than leading GPU-based systems. Tokens are small units of text, images, or video that AI systems process. The more tokens a rack handles per megawatt, the more efficient it is.
Microsoft did not name the systems used in that comparison. The figure is therefore a company claim that outsiders cannot yet check. The company also says a 96-chip Maia 300 rack offers seven times the rack-scale token throughput of the Maia 200. Rani Borkar, president of Azure hardware systems and infrastructure at Microsoft, described the aim as “more intelligence for every watt, every dollar, and every byte of memory.”
A Million Chips and the Workloads Behind Them
Microsoft says it plans to deploy potentially more than one million Maia 300 chips next year. The chips are meant to back its internal AI services. The company describes the rollout as multigigawatt in scale. A gigawatt equals 1,000 megawatts, which gives a sense of the scale involved.
Microsoft says Maia chips will support workloads such as Microsoft 365 Copilot. The company also points to OpenAI models, its own MAI models, and open-source models. For AI features in Teams, Copilot, and Cowork, Microsoft reports 1.7 times more tokens per watt than other GPU-based systems. As with the 40% figure, this is a claim from the company itself.
Why Cloud Giants Keep Building Their Own Chips
Microsoft is not alone in this effort. Cloud rivals and AI companies such as OpenAI are also releasing custom chips. Their aim is to loosen Nvidia’s hold on the AI computing market. Microsoft, Amazon, Google, Meta, OpenAI, and Anthropic either offer custom chips already or are working on them.
Making a chip in house lets a company tailor it to its own needs. It also acts as a hedge against relying too heavily on Nvidia and AMD. Amazon and Google now rent their custom chips to outside customers. Amazon is also reportedly exploring sales of chips that customers could install in their own data centers.
Nvidia Still Leads as Its Biggest Customers Build Rivals
These moves could be a problem for Nvidia. The company regularly earns upwards of 50% of its quarterly revenue from hyperscalers, which are the largest cloud computing companies. Chips that Amazon and Google rent out compete directly with its business.
That does not mean cloud companies will stop buying Nvidia processors. Customers still want access to several chip types for different workloads. Microsoft and its peers will therefore keep stocking the most in-demand processors. For now, those belong to Nvidia.
Where the Maia 300 Leaves Microsoft
The Microsoft Maia AI chip shows how seriously the company treats the cost of running AI. The chiplet design, the rack-level efficiency claims, and the plan for multigigawatt deployments all point toward more control over its own infrastructure. The key performance figures come from Microsoft itself. Independent testing will be needed to confirm the 40% gain.
Even so, the direction is clear. Microsoft plans to rely on its own chips for inference while still working with Nvidia and AMD. Readers following the AI hardware race should watch how quickly the deployments begin next year.
Discover how AI is reshaping technology, business, and healthcare—without the hype.
Visit InfluenceOfAI.com for easy-to-understand insights, expert analysis, and real-world applications of artificial intelligence. From the latest tools to emerging trends, we help you navigate the AI landscape with clarity and confidence
