(upbeat music) - Hello, and welcome to "Prairie Pulse."

You know, joining us today is Joe Kennedy, an academic technologist at Concordia College in Moorhead, Minnesota.

Joe, thanks for joining us today.

You're here today to talk about the rise in data centers across America, particularly in the rural areas and how they affect, what they can have on the communities.

But before we get into that, tell the folks a little bit about yourself and your background.

- Well, I am an academic technologist at Concordia, and as such, I'm responsible for looking at all of the different technologies we use for teaching and learning.

And when OpenAI released ChatGPT to the public a couple years ago, that included AI.

So now I also have the responsibility of coordinating AI initiatives.

So that's become a substantial part of my job, which means that I have to be aware of the broad impact of AI, different ways that we use it, how it impacts teaching and learning, and how it impacts all the career fields our students will go into, which is why coordinator of AI initiatives was chosen as a title.

It also means I'm not an expert on AI.

Like, I'm not a computer programmer who built the large language models.

I'm not an engineer who helped develop the data centers.

But I look at how they're built, how they're programmed, how they're set up, and how it works in the way that we use it, both at Concordia College and in the fields we're preparing students for it.

- Well, with that, I mean, yeah, you're coordinator of AI initiatives at Concordia.

I think you said that your third title that you're working at, but what do you really do in that role?

I mean, what all is involved with that?

- So it ranges from the very granular, where a professor will be looking at a course they teach, say a professor in the health professions, and they want to help prepare their students for the way the students might need to use AI when they're actually working with a patient, working in a laboratory, working in a dentist's office.

So they want to redesign their curriculum to include that.

To the very macro level where we say, as an institution at the Evangelical Lutheran Church of America, we want our graduates to be thinking about issues that might seem esoteric or really broad or they won't deal with in 10 years.

So how do we help them think about the impact of AI on privacy rights and environmental issues and intellectual property?

So I help the faculty and the staff as they're designing ways to make sure that our students are accounting for all of that.

- Yeah.

How long have you been involved in technology and what made you interested in it?

- I was called to be a math teacher when I was in seventh grade, and I taught high school mathematics and programming for 16 years.

And then I moved to Concordia College in the role of instructional design because I use technology so often, and the role of an instructional designer is to help faculty take what they're already doing that's good, and make it even better, more efficient, deeper, broader, et cetera.

And I've been doing that for 14 years.

- Yeah, well, you know, before we get to the main subject, let's define some terms.

You know, AI, people I think understand, and I shouldn't say that, they know what AI is, artificial intelligence, but then you got generative AI and then predictive AI.

Can you define those and tell us the differences?

- Sure.

So AI scientists have been working on artificial intelligence certainly since the sixties and even a little before.

Generative AI is the newest one.

And honestly, when you just say AI in conversation, that's what people think you're probably talking about.

So other forms of AI rely on like a decision tree where you ask yourself a bunch of yes/no questions and then you make decisions based on that.

Except that those forms of AI, there might be millions of yes/no questions, but at the end of the day, it's very programmatic.

That's not generative AI.

Generative AI is called generative AI because when it gives you an answer, it's generating something that it wasn't told it had to say.

It didn't follow a series of yes/nos.

And it does that by looking at billions of pieces of data, and then looking for what they have in common.

So, for example, if I ask you, who was the first president of the United States?

Why do you say George Washington?

- Because that's what history tells me.

- Because you have knowledge.

It's something you actually know.

Generative AI doesn't know.

Generative AI just looks at all that data that it saw, and it realizes that almost every time it sees the phrase first president and the United States, right next to it is this name George Washington.

But if we fed it billions of pieces of data that said Abraham Lincoln next to that, it would tell us the first president in the United States was Abraham Lincoln.

Because it is looking for associations, which words or phrases or pieces of words appear next to other ones and in which contexts.

So that's why it's generative.

- Okay.

So what services does generative AI provide?

You know, how does it contribute to outcomes, well, in the medical field or any field, medical or whatever you wanna talk about?

- Well, can I start with like a way you might use it and not even realize it?

- [John] Sure.

- If you're writing email in either Google or Microsoft, you're typing and all of a sudden you see a phrase appears that's kind of shadowed out, that's generative AI at work.

And the way it's doing it is, it's saying, "I've been watching you type email for years, and a lot of times when you start a sentence this way, this is how you end it or it's very similar to this.

So I'm gonna suggest maybe that's what you want to say."

In the medical field, one of the ways that it's used so amazingly well is to serve as a voice transcriptionist and this free doctors up where they can take more limited notes because they have an AI that is listening to the voice.

And, again, the difference is, if you did it, you know what the words mean.

Generative AI doesn't.

It just knows when it sees this particular wave pattern that maybe is the word penicillin, that it probably is exactly the same as this text pattern that spells out the word penicillin.

And it's highly trained in the medical field, and doctors never rely on it as that's absolutely what was said.

They're always looking to see if anything is off, but it's really, really accurate.

And then the doctor has, say, five extra minutes with you when they're not taking notes where they can talk to you, get to know the context, ask more about symptoms.

Visually, it can do the same thing.

It looks for visual patterns.

And so it's very good at helping people take something that they wrote or said aloud and turn it into a picture in an infographic.

It's used that way a lot too.

- Well, we invited you here today to talk about AI data centers.

A hot topic in the news right now.

But before we get into sort of the finer details of it, let's define it.

What is an AI data center?

- Well, first of all, it's just a data center like we've had for a long time, but for a very specialized purpose.

So all those things we just talked, about looking for patterns.

If you think about when someone asks you a question, you're like, "Oh, I think I have the answer."

And then maybe you've trained your brain to make an association to get you to the answer.

That takes some mental energy.

You can feel it.

An AI has to do the same thing, but computers are far less efficient than the human brain.

So for an AI to start looking at all that data and to make the connection so that it can generate something, it has to use an enormous amount of processing power.

So thousands and thousands and tens of thousands of computers like the one you may have at home, and they generate heat, and they have to talk to each other very quickly.

And the closer they are to each other, the easier it is to talk to each other.

Of course, this is a simplification, but that's what's going on.

So a data center says let's take all of these processors, all of these machines, put them in one place, make sure it's really cold so they don't overheat.

Because every time they make a computation, they're using electricity, generating a little bit of heat.

So an AI data center puts them all in one place so that we can deliver electricity to one place, have backup generators, alternate forms of power, multiple ways to cool, because it's far more efficient that way than to have it distributed all over.

- Okay.

Do people use these data centers on a daily basis?

- If you use a search engine, you probably do.

Not all search engines are AI-enabled, but the big ones certainly are.

So you can't use Google search engine without using the Google Gemini AI behind it unless you go in and change the settings.

If you use Microsoft Word, and it's helping you correct your grammar, you're using an AI tool.

So somewhere some data center is processing that request.

If you go straight to ChatGPT and ask it a question or Claude or Grok, you're gonna be using the data center.

So a lot of our work on computers involves touching a data center, even if that wasn't our intent.

- How do these data centers differ from ones that were built 10 years ago or 15?

- They're bigger, and they handle more calculations even if they're not bigger, which means they're generating a lot more heat in order to do that.

And, again, that's because an AI tool that uses programmatic or algorithmic, the decision tree, the yes/nos, that's an either or decision.

But an AI data center is asking the computer to start making associations where instead of just two choices, there may be hundreds of thousands of choices around the central prompt.

And so that's gonna require a lot more computing power.

Then when you start looking at music generation, audio generation, pictures and video, right, a really good picture requires so much data.

Every individual pixel has a different color.

What happens when they shift?

What's the fading?

How is the light hitting the surface and bouncing off?

And so now you have to have exponentially more data.

So the short answer is they're a lot more complicated, they do a lot more computations, they generate a lot more heat, and they need a lot more land.

- Are there different types of AI data centers, and if so, what are they?

- Some of the AI data centers that are purpose-built.

So Microsoft owns a data center, and instead of using water to help cool the environment, it's using refrigerant.

So hundreds of miles of piping inside the data center that are just carrying refrigerant around.

And that's a different engineering than something that is water-cooled or is using what we would think of as super air conditioners.

- Yeah.

What companies are primary behind the construction of these data centers?

I mean, you mentioned a couple of names.

Is it big organizations like Google or Amazon or small outfits?

- I wouldn't say any of them are small, because even the smallest data center, which could be built maybe by a local company, but it is still a massive engineering in terms of the footprint.

So it's not necessarily small, but it is sometimes the big firms that we think of, right, Anthropic and OpenAI.

I don't know which arm of Elon Musk's companies owns the one in Tennessee, but it powers Grok.

But then sometimes you also have firms that are national or transnational who specialize in building and maintaining a data center.

And then they lease it to the large AI companies.

- Well, according to a Brookings Report from November of 2025, the United States has about 5,500 data centers as of June 2025.

Is there a reason why the United States has so many of them compared to other countries?

- We certainly have a lot per capita, but China definitely has a lot of data centers.

Russia is building many data centers.

India has data centers.

But the reason that right now we seem to have more is because we're one of the few countries that is leading in terms of computing and computing power and different ways of creating AI tools.

And so in order to do that, you need to have the data centers to power them.

But China is rapidly catching up.

China's come up with a couple different ways of thinking about data centers that are very outside of the box.

They just opened one that sits offshore, 35 meters under the sea, for instance.

- Well, you know, why has there been a boom?

I mean, 30 years ago, we always talked about hardware and software sort of leapfrogging each other.

Is it just a technology boom that we've got going on?

- Yes and, feel like improv here.

The technology boom is a subset of human knowledge boom.

As we learn more, it doesn't progress, right, like a line.

It goes like this.

So we've got a nice exponential curve, and that's happening with technology as well.

And then when OpenAI released ChatGPT, it was the first commercial, quasi-commercial, but certainly the first large scale way of a new way of thinking about a technology.

And so, of course, that immediately spurred a whole lot of other companies saying, "Well, if OpenAI can think about it this way, why can't our bright people and our smartest minds think about it this way?"

So that's gonna lead to a rapid explosion of development, just like it does with any tech.

- Where are the majority of AI data centers located in the United States?

- At first, they were wherever they could be built conveniently.

So there wasn't one easy way to classify it.

Now they're looking, the companies that are building them are looking far more towards rural areas.

Land is cheaper, fewer residents to get upset.

The problem is, oh, and rural areas with arable land are particularly attractive because if the land is arable, there's a water source close by and you need a lot of water for most data centers.

- Okay, well, there you go because that was leading into, you know, a talking point about them.

How much water is actually used in these data centers, and what's the environmental impact with that?

- Well, first let's be clear that an individual AI query doesn't use that much water.

It's less than a teaspoon, it's measured in milliliters.

It's just that we make so many of them.

But even with that, data centers consume less water than a couple other industries.

The biggest one is fast fashion.

Blue jeans take thousands of gallons worth of water to make one pair between growing the cotton, the shipping, and the dying, which is really intensive.

So fast fashion, which designs clothing to be worn for a year and thrown away actually uses more water as an industry than AI.

But AI is growing really, really fast.

So the individual use of an AI doesn't use a lot, but collectively it uses enough water that it can deplete a water table or an aquifer very quickly if it's not managed well.

- Why do you think some people have responded with skepticism toward data centers being built close to where they live, and yeah?

Why is that?

- Well, there's a general, especially in America, there's a general mistrust of government saying, "We're gonna do this big thing."

There's a distrust of AI in general.

There is no way to build a data center without extreme noise.

You know how your computer hums?

Imagine 25,000 of them all in one place, and then all the turbines that blow the air, and the water pumps to keep it cool, and the generators that are always running as backup generators in case something happens to the power grid.

So there's the noise.

People oppose it because it can take so much land that a local government may feel they need to use eminent domain.

We know that it uses a lot of water.

So there's a concern sometimes founded on data, sometimes just in our heads, that it's going to deplete the local water.

And then there are also many, many studies including from the Brookings Institution, that talk about the impact on your electricity rates if there is a data center built on the same power grid that you're using.

- Yeah, well, you talked about size a moment ago.

You know, how much space do they require?

An average I heard is a hundred thousand square feet, and with hyperscaled ones up to a hundred million square feet.

- Yeah.

And for most of us, we just think about it in terms of, how many football fields is it going to take up?

And they're very, very large.

They're larger than when Amazon comes in and wants to put a distribution center in town.

But, again, if you build in a rural area, it may not seem like it's so impactful.

And in fact, maybe there's one landowner who has enough land that they'd like to sell, and that can all be done with minimal disruption.

But you can't count on that.

- Well, overall, what are the long-term versus short-term cost of an AI data center for a rural community?

- So there is, oh, that's such a large question, John.

So the short-term cost are that there's gonna be disruption, there's gonna be construction.

You gotta put in power lines or build a power plant.

Lots of wind turbines or solar, however you're gonna power it.

There's all the lines that are gonna get the data in and out.

But the long-term cost tends to be, there is no way around it.

It's going to use more water than that area is used to.

So if the geological surveys are accurate and you've got reasonable predictions of what it will use, it could be that the long-term cost isn't that great in terms of water usage or electricity usage.

But if it's rushed into, and this is what the concern is in a lot of North Dakota communities right now is, the residents feel that the governments are rushing into agreements.

And they haven't done all of that research yet.

- I got too many questions.

We're running out of time.

Another topic when it comes to data centers, especially in rural communities, is about non-disclosure agreements or NDAs.

What role do NDAs play in the construction of data centers?

- Well, it's important that I have very little bias, but this is one I am very biased towards full transparency.

So whenever you say NDA, I immediately go, ah!

Non-disclosure agreements are needed to protect, like, oh, we have an innovative way of cooling the data center.

But a non-disclosure agreement by its very nature means that citizens don't get to know all the information about something that's being built.

And if it involves any tax breaks, any public land use, any diversion of water resources, or an impact on their electricity, citizens, of course, are going to say, "Why can't I know what's going on?"

So an NDA needs to be very carefully constructed to let citizens know what could happen to them without giving away proprietary information for the company owning the data center.

- Do AI data centers have positive impacts?

If so, what are they?

- Well, yeah, there are.

AI tools in general that have positive impacts when they're well tailored, like in the medical community, AlphaFold that has helped envision what every protein we can imagine looks like.

Like, these have been tremendous, and they can't happen without data centers.

So there are indeed positives.

I think when people get a negative view, they're just thinking of the AI slop on their newsfeed, right?

That's a data center powering that just like there's a data center powering AlphaFold.

- You know, is there a way for people to responsibly interact with generative AI despite some of the negative ramifications with the environmental impact?

- Yep.

And it's the same as every other technology we worry about.

If you have a car and you need to walk three blocks, walk instead of driving.

If you have access to an AI tool, but you need to look up the capital of Lichtenstein, use an encyclopedia or Google it rather than using an AI tool deliberately to do that.

That mitigates the use of the tool, reduces demand for the data centers, conserves the resources for the hard questions that only AI can really answer.

- We talked before we came on the air about research and things with AI.

Can you talk a little bit about that?

- Sure.

The last thing about opposition to AI data centers that the research is starting to reveal in only the past few months, is that some people are opposed to AI data centers because they're really opposed to AI.

They have a general distrust of this technology, but AI is like out there, right?

It's amorphous, you can't touch it, but a data center, you can impose that building.

So someone may be opposed to it for environmental reasons, and then five people are also opposed, eh, maybe for the environment, but really because it's more of a generalized antipathy towards a technology that's not always being well explained and not always being well marketed.

- Well, Joe, what do you think the future of AI data centers is?

- We're gonna see more of them for a variety of business needs, national security needs, geopolitical politics.

I believe we're going to see a slowing down of the growth, because I believe the AI industry in general is going to become more focused, tailoring itself to the problems it does best and less just let's throw AI at everyone.

- Well, we are running out of time, but if people are more interested in finding out about this, where can they go?

Who can they contact?

- For broad overviews, things like the Brookings Institution, other think tanks, the Heritage Foundation, the Electronic Frontiers Foundation.

For very specific topics, like if you're worried about the environmental impact, go to something like the Sierra Club.

If you're worried about the impact on your local business community, start with the Better Business Bureau, which is gonna connect you to national networks of research.

- It seems like there's an awful lot to think about on this one, Joe.

- It really is mind-staggering, John.

- Yeah, thanks for joining us today.

- Yeah, thanks for having me.

- Well, that's all we have this week on "Prairie Pulse."

And as always, thanks for watching.

(upbeat music) - [Announcer] Funded by the members of Prairie Public.

(upbeat music fades)