Microsoft Unveils Maia 300, Its Third-Generation AI Chip, with Plans to Deploy over a Million Units Next Year in a Push to Loosen Nvidia’s Grip

Redmond: Microsoft pulled the cover off the Maia 300 on Monday, the third generation of an AI chip it designs itself, and said it may deploy more than a million of them next year to run its own AI services. It is the clearest sign yet that the biggest cloud companies no longer want to rent all their computing muscle from Nvidia.

Efficiency is the whole pitch

The Maia 300 is built from chiplets — smaller pieces of silicon bonded together. The heavy computing work sits on TSMC’s 2-nanometre process, while the parts that move data around use cheaper 3-nanometre technology. Microsoft says the result should produce 40 percent more tokens for every megawatt than leading GPU-based systems, and that a full rack of 96 accelerators will manage up to seven times the token throughput of the Maia 200 while drawing about 167 kilowatts. With electricity now the binding constraint on AI expansion, that efficiency is not a technical footnote. It is the business case.

Nvidia is not going anywhere — yet

Chief executive Satya Nadella described the chip as the company’s biggest bet yet on its own silicon. The Maia 200 is already doing real work, serving Microsoft’s own models and OpenAI’s models through Azure. But production deployments of the Maia 300 do not begin until 2027, and until then Microsoft will keep leaning on Nvidia and AMD hardware. The direction of travel is set. The timetable, as ever with custom silicon, is the harder part.