Key facts
- Microsoft unveiled its third-generation custom AI chip, the Maia 300.
- The company plans to deploy over one million Maia 300 chips next year.
- The Maia 300 uses chiplets built with 2-nanometer technology from TSMC for computing functions.
- The chip aims to deliver 40% more computing per megawatt than leading GPU-based systems.
- Microsoft expects to spend over $175 billion in capital expenditures this calendar year.
Microsoft announced on Monday the next generation of its custom AI chip, the Maia 300, and revealed plans to deploy more than a million of these chips next year to support its internal AI services. This move is part of a broader industry trend where cloud computing rivals and AI firms are developing their own chips to reduce reliance on Nvidia.
The Maia 300 chip utilizes a "chiplet" design, where specialized chips are bonded together. Key computing functions are handled by chiplets built using Taiwan Semiconductor Manufacturing Co.'s (TSMC) 2-nanometer technology, while other functions, such as data transfer, use TSMC's 3-nanometer technology.
According to Microsoft, the Maia 300 processors will be housed in racks containing 96 chips each, consuming approximately 167 kilowatts of power. The company claims these chips will deliver about 40% more useful computing per megawatt compared to "leading GPU-based systems," though specific systems used for comparison were not named. Rani Borkar, president of Azure hardware systems and infrastructure at Microsoft, stated that the optimization across the entire system delivers more intelligence for every watt, dollar, and byte of memory.
Microsoft anticipates "multi-gigawatt deployments" of the Maia 300 chip starting next year. This initiative aligns with the company's significant capital expenditure plans, with an expectation to spend over $175 billion in capital expenditures this calendar year.