Microsoft’s new 'Fairwater' megafactory aims to revolutionize AI training — how the world’s most powerful datacenter works

Microsoft has unveiled Fairwater, its largest and most advanced AI data center to date, located in Wisconsin.

Fairwater is the first of several identical datacenters under development across the U.S., as well as new projects in Norway and the UK.

These sites represent tens of billions of dollars in investment and hundreds of thousands of GPUs. Microsoft says the aim is to link them together into a single, global Azure AI data center, effectively creating a distributed AI supercomputer.

These AI datacenters are purpose-built, very different from traditional cloud facilities. A typical cloud data center runs many smaller jobs at once, like hosting websites or email. An AI data center instead focuses all of its resources on one enormous task — training massive AI models. Rather than servers working independently, everything is tied together to act like one giant supercomputer.

They run on NVIDIA’s GB200 GPUs, with GB300 already on the horizon, offering up to 10 times the performance of today’s fastest supercomputer. These GPUs are tuned for nonstop calculations, performing trillions of equations every second.

The scale here is remarkable and a bit mind-blowing. Each rack can process 865,000 tokens per second. Tokens are essentially the tiny pieces of text, images, or video that AI learns from during training. By combining memory across GPUs, the system avoids bottlenecks and keeps performance consistent.

Energy costs for facilities like this must be enormous, and it’s something I worry about as more of these sites come online. It’s also a shame nuclear energy was never more widely adopted, since fears around instability kept it from playing a bigger role in powering projects like this.

Whilst environmental concerns do exist, Microsoft is using a closed-loop liquid cooling system with zero water waste. Instead of relying on air conditioning, Fairwater uses sealed pipes of liquid that continually cycle through the servers and are reused again and again.

Fairwater also runs on one of the world’s largest water-cooled plants. Around 90% of its capacity is cooled without using any new water at all, with outdoor air doing most of the work to keep temperatures down.

As for storing data, Fairwater is built at an exabyte scale, which is billions of gigabytes. It can handle millions of read and write transactions per second, ensuring the data center keeps up with the demands of AI training.

Cooling is a big part of the conversation, but the bigger concern is electricity. These AI datacenters consume more energy than hundreds of households use in a year. Microsoft has also purchased 3.5 million carbon credits to help offset emissions from AI, partnering with Re.green in January 2025, securing these credits over the next 25 years.

The big vision

Image 1 of 4

It’s an impressive vision from Microsoft, not just to build these vast AI datacenters but to link them together into what it calls an “AI Factory.” Instead of the site in Wisconsin handling jobs on its own, it can pool resources with facilities in the UK or Norway. In theory, this makes the system more reliable, easier to scale, and far more capable of training huge AI models across several regions.

Microsoft likens this new AI era to the industrial revolution. Back then, factories powered economic growth by producing goods at scale. Today, it is pitching AI datacenters as the same thing, only instead of producing goods, these facilities produce “intelligence.”

It’s an impressive feat, but I can’t help thinking about the long-term impact. Growing up, I was always told to worry about my own footprint on the environment and do what I could to reduce it. Now, seeing the scale of these projects, it’s hard not to worry about the environmental cost. Renewable energy can help offset some of these concerns, but I also hope companies turn more seriously to nuclear power, which remains one of the safest, most reliable, and cleanest energy sources we have.

Follow Windows Central on Google News to keep our latest news, insights, and features at the top of your feeds!