Microsoft details Maia 300, third in-house AI chip aimed at cutting costs on Azure

Microsoft announced on Monday (the 12th) the third generation of its in-house artificial intelligence chip, the Maia 300, according to a Reuters report published by the Economic Times. The company plans to deploy potentially more than 1 million units of the chip as soon as next year to support its internal AI services.

The Maia 300 is the direct successor to the Maia 200, which already runs both Microsoft's own models and OpenAI models offered via Azure. The new generation arrives at a moment when cloud rivals and AI companies like OpenAI itself are racing to develop their own silicon and reduce dependence on Nvidia in the AI computing market.

How Microsoft built the chip

Microsoft's technical approach differs from Nvidia's, whose AI accelerators use a single massive compute chip surrounded by high-bandwidth memory. The Maia 300 follows a different assembly logic.

The Maia 300's central processor is made up of "chiplets," specialized chips bonded (physically joined) together, which use TSMC's 2-nanometer technology for the main computing functions. Other chiplets in the same design use the cheaper 3-nanometer technology for functions such as moving data out of the processor and between other parts of the system.

This separation by function, rather than a single monolith, is Microsoft's bet to balance performance and manufacturing cost within the same chip.

Rack numbers: Maia 300 vs. GB300

Maia 300 processors go into racks with 96 chips, consuming about 167 kilowatts. That's more chips and more power per rack than Nvidia's most recent AI rack.

Microsoft states that the Maia 300 delivers about 40% more useful compute per megawatt than "leading GPU-based systems," but did not name which systems it used for comparison. Without the competitor's name or the benchmark methodology, the number needs to be read as the company's own marketing until independent validation appears.

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

Scale of the investment

Microsoft said it plans "multi-gigawatt" deployments of the Maia 300 starting in 2027. To put the scale of the bet in perspective: the company had already stated it expects to spend more than US$ 175 billion in capex in this 2026 calendar year, much of it on AI data center infrastructure.

This volume of investment is the backdrop that explains why in-house silicon matters: buying Nvidia GPUs at gigawatt scale carries a cost that no cloud company, however large, can sustain without pursuing its own alternative.

What changes for developers in Brazil

The Maia 300 is not a product for sale: it runs Microsoft's internal infrastructure, including the company's own models and OpenAI's models delivered via Azure. Developers don't choose to "run on Maia" the way they choose a VM instance.

The practical effect, if the promise of efficiency per watt holds up, shows up on the price side: lower energy cost and lower capex per unit of compute tends to be reflected, with a lag, in the price charged for the inference API on Azure. Brazilian startups that today pay per token for OpenAI models hosted on Microsoft's cloud are, in that sense, indirect beneficiaries of any real hardware efficiency gain.

There's also a point of attention for anyone thinking about multi-cloud architecture: silicon diversification among the major cloud providers and OpenAI itself reduces Nvidia's pricing power in the medium term, which historically pushes the entire market down, including providers that keep buying GPUs from Nvidia.

In short: the Maia 300 is still closed infrastructure, but it's the kind of behind-the-scenes decision that, if it works the way Microsoft promises, reaches the wallet of whoever pays for AI APIs in Brazil before it reaches any product changelog.

Source: ET Tech / The Economic Times (Reuters), "Microsoft plans huge deployment of its own AI chips, with an eye on saving costs" (https://economictimes.indiatimes.com/tech/technology/microsoft-plans-huge-deployment-of-its-own-ai-chips-with-an-eye-on-saving-costs/articleshow/134896014.cms)