New to this? Read it in simple words
- Microsoft showed Maia 300, its own AI chip, on October 12.
- Microsoft wants Maia to run most of the AI in its products from 2027.
- It claims 40% more AI output per unit of power than leading GPU systems.
- Nobody outside Microsoft has tested the chip yet.
- Inference
- Running a trained AI model to answer requests, as opposed to training it.
- Chiplet
- A small chip that works together with others in one package, like parts of one bigger chip.
- Gigawatt
- A billion watts of power: roughly what a large power station produces.
An AI chip can train a model or run it. Running it, called inference, is what happens every time someone sends a request to Copilot or ChatGPT. Microsoft’s Maia chips are built for that second job.
Microsoft’s plan
Microsoft announced Maia 300 on Monday, October 12, in a post by cloud chief Scott Guthrie. “Maia 300 will begin production deployments in 2027,” the post says.
Microsoft will scale its Maia chips to deployments of several gigawatts from 2027, making Maia “the primary inference system for Microsoft products and services”. A Microsoft manager declined to give CNBC a timeline for that.
The older Maia 200 already serves Microsoft Copilot and runs OpenAI models, Microsoft’s own MAI models and open models. Microsoft says it is deploying in the United States, Europe and Australia.
Maia 300 as Microsoft describes it on October 12; every figure is Microsoft’s own, and none has been tested outside.
What is inside
Maia 300 splits its work across several smaller chips, called chiplets. The compute chiplets are made by TSMC on its N2P process.
Each chip has eight stacks of HBM4 memory, with 18 terabytes per second of bandwidth, within 950 watts. A full rack holds 96 chips within a power budget of 167 kilowatts, twice as many chips as a Maia 200 rack, CNBC reports.
Microsoft says Maia 300 will give 40% more tokens per megawatt than “leading GPU-based systems”. It also says a full rack handles up to seven times as many tokens as a Maia 200 rack.
What is still unknown
“Nobody has tested them outside of Microsoft yet,” Patrick Moorhead, an industry analyst, told CNBC. Microsoft does not name the GPU systems it compared Maia 300 with.
Unlike Google’s and Amazon’s AI chips, Maia chips cannot be rented by the hour by other companies, CNBC reports. For training models, Microsoft still works with Nvidia and AMD.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1Primary source · October 12, 2026Microsoft to expand Azure Maia system to multi-gigawatt deployments Microsoft
- 2
- 3Research · October 12, 2026Microsoft to rely more on homegrown AI chips, reducing dependence on Nvidia CNBC
- 4
- 5Research · October 12, 2026Microsoft unveils Maia 300, designed for multi-gigawatt deployments Techzine
- 6
The chip’s specifications and its speed and power figures are Microsoft’s own. No outside group has tested Maia 300, and the 2027 plans can still change.
Microsoft’s Maia 300 chip promises 40% more AI output per unit of power from 2027, though nobody outside has tested it. How we rate the mood