Microsoft plans huge deployment of its own AI chips, with an eye on saving costs | Reuters
Microsoft intends to deploy potentially more than a million of its custom Maia 300 AI chips next year, mainly to cut what it pays for computing compared with Nvidia-style GPU systems. The pitch rests on the company's claim that Maia 300 delivers about 40% more computing per megawatt than leading GPU-based setups.
- Microsoft could deploy more than 1 million Maia 300 chips next year.
- Each Maia 300 rack holds 96 chips and draws about 167 kilowatts.
- Microsoft says Maia 300 gives about 40% more computing per megawatt than leading GPU systems.
Microsoft plans to deploy potentially more than a million of its own Maia 300 AI chips next year, according to the Reuters piece. The push is aimed at cutting what Microsoft pays to run AI workloads, which matter a lot to a company that sells cloud computing and AI services at huge scale. Microsoft's pitch rests on efficiency: the company says Maia 300 delivers about 40% more computing per megawatt than leading GPU-based systems. Power is the real bottleneck in data centers right now, so that figure is the number that counts for cost.
The hardware details help explain the bet. Maia 300 racks hold 96 chips and draw roughly 167 kilowatts, so the density is high and the power budget per rack is substantial. Scale that to a million chips and you are talking about an enormous buildout of electricity, cooling, and floor space, not just a purchase order. Microsoft is essentially wagering that its own silicon, tuned for its own software stack, can undercut the Nvidia GPUs it currently relies on heavily, while reducing its dependence on a single supplier.
Keep in mind that "potentially" is doing a lot of work in that headline figure, and the 40% efficiency claim comes from Microsoft itself rather than independent benchmarking. The gap between a planned deployment and a delivered one can be wide in chip manufacturing, and vendor performance claims often shrink under outside testing. The useful thing to watch is whether Microsoft's cost savings show up in its cloud margins or pricing over the next year or two.