Microsoft unveils Maia 300 and plans deployment of more than one million

what happened

Reuters reports that Microsoft on Monday unveiled Maia 300, the third generation of its proprietary chips for artificial-intelligence workloads. The company said it could deploy more than one million of the processors next year to support its internal AI services. Its predecessor already runs Microsoft’s in-house models and OpenAI models supplied through Microsoft’s Azure cloud service.

The new processor uses a collection of “chiplets” rather than the single large computing chip surrounded by high-bandwidth memory used in Nvidia’s accelerators. Taiwan Semiconductor Manufacturing Co’s 2-nanometer technology is used for key computing functions, while less expensive 3-nanometer technology handles tasks including moving data away from the processor and within the system.

Maia 300 processors will be installed in racks containing 96 chips and consuming about 167 kilowatts. Nvidia’s latest GB300 AI server racks, with 72 GPUs, use 142 kilowatts. Microsoft said Maia 300 would deliver about 40% more useful computing per megawatt than “leading GPU-based systems,” without identifying the systems used for comparison. It plans to begin “multi-gigawatt deployments” next year and expects capital expenditures above $175 billion this calendar year.

why it matters

The launch highlights efforts by cloud providers and AI companies to reduce their reliance on Nvidia. For markets, the key issues include the chip’s efficiency, manufacturing cost and the scale of Microsoft’s planned deployment.

instruments