Wisconsin Public Radio

What Are Data Centers? How Will They Change Wisconsin?

Will support AI. Just one project could use as much energy as all of Milwaukee.

By , Wisconsin Public Radio - May 12th, 2025 12:46 pm
An artificial intelligence data center is built on land once slated for development by Foxconn in Mount Pleasant on Wednesday, May 8, 2024, in Mount Pleasant, Wis. Angela Major/WPR

An artificial intelligence data center is built on land once slated for development by Foxconn in Mount Pleasant on Wednesday, May 8, 2024, in Mount Pleasant, Wis. Angela Major/WPR

WPR reporters are diving into the development of data centers around Wisconsin and the impacts they could have on the communities where they’re being built and the state as a whole. This story is part one in a four-part series.

Mount Pleasant, Port Washington, Kenosha, Beaver Dam — in recent months, Wisconsin’s data center tally is growing.

All four communities may one day host large, artificial intelligence-focused data centers for companies like Microsoft and Meta. Individual projects could occupy upwards of 1,000 acres and use energy loads equivalent to the entire city of Milwaukee. In response, lawmakers are already seeking to expand nuclear energy in Wisconsin while the Public Service Commission is considering a new electric rate for data centers.

But what are data centers? How are local governments dealing with them? How do they affect sustainability goals? And how do they impact existing communities? WPR has created a four-part feature series to explore the answers to those questions.

What do data centers do?

Data centers do the same two things that laptops or smartphones do: computing and storage.

That’s according to Vijay Gadepally, a computer scientist at the Massachusetts Institute of Technology. He used two extreme examples to illustrate the difference.

“On one end, let’s say we have a supercomputer,” he said. “Ninety-five percent of the equipment in that data center, 90 percent, is computing.”

That means halls filled with graphics processing units, or GPUs. These are powerful versions of central processing units, or CPUs, the part of the computer that executes commands, crunches numbers and runs programs.

Supercomputers may use GPUs to, for example, run physics experiments with many variables.

On the other end of the spectrum would be a data center that stores an iCloud photo library, or emails for an email service.

“The ratio is almost flipped,” Gadepally said. “In that situation, maybe 90 percent of the physical space is taken up by hard drives.”

Hard drives are a computer’s storage tool, keeping everything from audio files to PDFs.

What is an AI data center?

Data centers that power artificial intelligence programs often have a mix of storage and computing, Gadepally said.

Storage is needed for training AI models — feeding them huge amounts of public internet data, refining their ability to predict correct answers to users’ questions.

For example, he said, if the model is answering a question like “Who is the President?” it would predict “Donald Trump” based on its training with loads of recent internet data that says “The President is Donald Trump.”

“Of the million instances I’ve seen, 500,000 have ended this way, so I’m going to reproduce that, and that’s the data that’s stored in the model,” he said.

Computing is needed to answer users’ questions by sifting through and piecing together what the model has learned during training.

Gadepally said AI data centers can specialize on training or answering questions. He said the former are often much larger and built in places with cheap energy. The latter tend to be smaller and are meant to reduce response times by being built close to population centers.

But even when both computing and storage are present, computing — or question-answering — uses much more energy.

Again, Gadepally said that’s similar in household electronics like smartphones.

“When your battery dies quicker is when you’re doing something that’s compute-intensive,” he said. “You’re doing something with Photoshop, you’re on your phone and maybe you’re editing an image.”

Computing is more energy-hungry because it is “continuous in nature and executing billions of mathematical operations per second,” Gadepally wrote in an email. Meanwhile, storage is mostly passive, waiting for requests to access stored data.

Around 33 percent of the average data center’s energy use is for cooling, however, Gadepally, with that number potentially rising to 50 percent on hot summer days.

What are data centers made of?

Data centers include warehouse-like building materials, network equipment, power supplies and strong air or liquid cooling systems.

But — just like in a smartphone or computer — the key component of data centers are semiconductors.

Semiconductors are materials that can act as both insulators and conductors of electricity. Wafers of these materials are layered and etched with tiny patterns to create semiconductor chips, also called computer chips.

“The largest AI data centers now are including hundreds of thousands of [chips],” said Saif Khan, who’s researched semiconductor supply chains for the Biden administration’s Commerce Department and Georgetown University’s Center for Security and Emerging Technology.

He said that, if data centers keep growing, each one could cost their developer tens of billions of dollars in semiconductor costs.

Where do semiconductors come from?

The global supply chain of semiconductors hinges on one company — the Taiwan Semiconductor Manufacturing Company, or TSMC.

In 2023, Taiwan produced over 60 percent of the world’s semiconductor chips and over 90 percent of the most advanced chips, The Economist reported. Most of those were made by TSMC.

“The manufacturer does not design its own in-house chips, and instead, services third-party designers,” Khan said of TSMC’s business model.

Orders come in to TSMC from chip designers, many of them American, like Apple or Nvidia.

To build those chips, the company needs minerals like silicon, gallium and palladium.

“China is, in general, the big producer here,” Khan said.

In 2024, China mined 84 percent of the world’s silicon and 99 percent of its gallium, according to U.S. Geological Survey Mineral Commodity Summaries. They said Russia and South Africa lead in palladium production.

Then comes a manufacturing process with “well over 1,000 steps,” Khan said.

He said the most important steps are layering silicon wafers with other minerals, using a “lithography tool” to draw the shapes of circuits on the wafers, etching those shapes into the metal permanently and inspecting chips for quality.

That process involves some seemingly unexpected countries — 100 percent of the world’s supply of a certain lithography tool, Khan wrote in a 2021 report, is made by a single company from the Netherlands, ASML.

“They’re the only company in the world that can make them,” Khan said.

Both the Biden and Trump administrations have tried to bring parts of their supply chain back to the U.S.

The Biden-era CHIPS and Science Act created incentives to move semiconductor plants to the U.S. One is planned by TSMC in Arizona, with $6.6 billion in CHIPS subsidies.

Since his inauguration, President Donald Trump has ordered an investigation into U.S. import dependencies for minerals including gallium and germanium. A March order directs executive agencies to explore financing new mining operations themselves, and to prioritize mining on federal lands.

What’s the future of data centers?

Most of the large AI data centers proposed in Wisconsin aren’t off the ground yet. And some projects may be moving slower than originally predicted, as WPR previously reported.

But demand for AI will continue to drive growth in the data center industry, said Meredith Strong, an infrastructure consultant at PricewaterhouseCoopers, a global accounting and consulting firm.

“We don’t see that demand slowing down, and I think that AI is a major catalyst for that demand for data centers,” she said.

She said that includes non-AI data centers building AI capabilities into their existing facilities.

But if developers want to keep making data centers bigger and bigger, she said, they’ll have to confront limits on availability resources, including the electricity needed to power them.

“There’s a limited supply of power, there’s limited land and resources,” she said.

“That scaling is going to be hindered by how much of all of those resources are available, and power is certainly one of those resources,” she said, adding that limits motivate companies to find more efficient ways to build and power data centers.

In Wisconsin, making land and power available for data centers ultimately falls to governmental bodies.

One of them is the Public Service Commission, appointed by the governor. It’s considering a special electric rate for data centers, which could force developers to pay for infrastructure built to power their data centers.

The others are local elected council members and mayors. In recent months, officials have approved annexation deals and subsidies for proposed data centers, while residents in at least one community have signed a petition opposing the plans.

Listen to the WPR report

What are data centers? Wisconsin cities’ potential new neighbors, explained was originally published by Wisconsin Public Radio.

If you think stories like this are important, become a member of Urban Milwaukee and help support real, independent journalism. Plus you get some cool added benefits.

Leave a Reply

You must be an Urban Milwaukee member to leave a comment. Membership, which includes a host of perks, including an ad-free website, tickets to marquee events like Summerfest, the Wisconsin State Fair and the Florentine Opera, a better photo browser and access to members-only, behind-the-scenes tours, starts at $9/month. Learn more.

Join now and cancel anytime.

If you are an existing member, sign-in to leave a comment.

Have questions? Need to report an error? Contact Us