How will we power the AI boom?
47:18 Share

Support for WBUR comes from MathWorks, creator of MATLAB and Simulink software for technical computing and model-based design. MathWorks, accelerating the pace of discovery in engineering and science. Learn more at MathWorks.com. I can say to my new Samsung Galaxy S25 Ultra, hey, find a keto-friendly restaurant nearby and text it to Beth and Steve. And it does without me lifting a finger, so I can get in more squats anywhere I can. One. Two.

Two, three. Will that be cash or credit? Credit. Galaxy S25 Ultra, the AI companion that does the heavy lifting so you can do you. Get yours at Samsung.com. Compatible with select after-pars Google Gemini account results may vary based on input check responses for accuracy.

This is On Point. I'm Meghna Chakrabarty. And today we're going to start off with a question for ChatGPT. I've got a window open right here in front of me, ChatGPT.com, and this is live radio, so I'm not 100% sure this is going to work, but we're going to try it out. So here's my query for ChatGPT. How much electricity does Chat—you guys are now finding out how slow of a typer I am—ChatGPT use per query—

compared to an incandescent light bulb with a B. How much electricity does ChatGPT use per query compared to an incandescent light bulb? Enter. It is thinking. Oh, here it goes. The amount of electricity ChatGPT uses per query is relatively low, but calculating it directly can be complex. It depends on a number of factors. However...

The bot says,

The extent of energy consumption, for one query, okay, here's finally the answer, is likely in the range of a few watt-seconds. For context, one study suggested that a large model like GPT-3 could consume 0.5 to 1.5 kilowatt hours for training. Okay, wow. So, this is not actually answering my question. It says here...

ChatGPT uses 0.01 to 0.1 watt hours per query, and a 60-watt incandescent bulb uses 60 watt hours. I mean, that's why it's called a 60-watt bulb. So ChatGPT says it uses much less electricity per query. Interesting. Okay, let me say how about—obviously, this is going to be maybe an even more strenuous comparison. How about an LED bulb? Okay.

Again, live radio, I have no idea if this is going to work. Blah, blah, blah. Okay, it explains that LED bulbs are even more energy efficient.

And it says, "Even an LED bulb, which is more efficient than an incandescent bulb, still consumes more energy than ChatGPT does per query." Oh, interesting. Okay, it's funny because we ran this totally transparent—we ran this same test yesterday, and ChatGPT gave us a completely different answer. It actually said it uses more energy than an incandescent bulb. So here's the thing.

ChatGPT's architecture, obviously, its hardware, its LLM is not public. So we cannot verify the accuracy of the AI's response. And obviously, we're also getting just non-repetitive, non-replicable answers here just over 24 hours. But Jesse Dodge, who's a researcher at the Allen Institute for AI, he has done his own independent estimate here.

And he found that one chat GPT query could use as much electricity as one LED light bulb needs to stay on for 20 minutes. Now, this is really interesting because it's completely different than what chat GPT itself claims. Huh. Oh, you know what? Maybe it's because it's not factoring in the cloud servers, the multiple layers of processing, storage, and machine learning. It looks like it's only looking at a part of its energy consumption. Yeah.

Well, anyway, the point is, is that it's not just chat GPT. AI is everywhere. And you've probably heard by now that it can do amazing things. And in order to do that, it actually does take a massive amount of energy. It will become, AI, will become one of the largest single sources of electricity consumption in this country.

By 2030, AI and the data centers that power it are expected to make up anywhere from 8 to 18 percent of total U.S. power demand. So there's been a big question in the world of AI, in the world of energy policy. How will we power this AI boom? And what are the environmental costs of the rapid rise of artificial intelligence? So that's what we're going to talk about now. And Evan Helper helps us

joins us to help us do that. He is a business reporter for The Washington Post where he covers the energy transition. Evan, welcome to On Point.

Hi, Meghna, good to be here with you. Okay, and I just want to repeat, this is live radio. So I actually did type in that query to chat GPT. And I'm being completely honest when I say it gave us a different answer yesterday. Yesterday, it claimed it used a lot more power. So how can we know, Evan, how much power AI is taking up? Or even, I don't know if you have the information down to like a single query, how much power does it take?

Yeah, it does make you wonder why we're putting all these resources into this miracle of AI if it can't even answer this simple question you're asking it. The metric we've generally used is one chat GPT query uses 10 times the amount of energy as a simple Google query without AI.

And that doesn't sound like that much when you compare it to a light bulb and you compare it to these other things we do with energy, it seems like a relatively small amount. But when you think about how many queries are going on in the internet at any given moment, I mean, you're talking about billions and billions of queries. So it just, it adds up to a huge amount. And of course you point out rightfully that this is not just about chat GPT queries, but everything else AI is doing and wants to do.

And the amount of energy that we're talking about needing, as you pointed out, could be as much as 18% of all energy consumed in the U.S. by 2028 or 2030. At the same time, energy use has been relatively flat in this country for years. And right now, data centers are only using less than 5% of the electricity available. It's an astonishing number when you think about it, that this one...

I mean, in all its forms, we'll just generally call all AI a tool, could account for nearing 20 percent of our energy consumption in this country in just a few years. So at this point in time, Evan, like, do we have an energy generation system and grid that's capable of meeting that demand? Yeah.

That's an easy one to answer. The answer is no. I mean, if we're really going to need as much energy as these companies are saying that we need, and the White House is now saying that we need, and everyone in this industry is saying we need, there needs to be a lot more generation built, and it needs to be built very quickly.

You know, the problem is it's really hard to build generation and it's especially hard to build it quickly, especially over the last few years with all the supply chain issues and, you know, other sort of permitting fights and landowner rights. And it's just been really hard to build any transmission and new generation for the power grid. And now suddenly we need a lot, a lot of it.

Okay, so transmission and new generation. These are obviously two very mutually important factors when it comes to power consumption, but I want to deal with them slightly differently. Let's talk about generation.

First and foremost, if these AI companies, these tech companies need to stand up that much more power in, well, now essentially is when they want it, but in the next couple of years, to your point, building new facilities due to the permitting, due to finding the land, et cetera, et cetera, is not the fastest way to go. So where are Meta, Google, OpenAI, all of these companies, where are they looking to create more power?

Well, they're going to places where there's a lot of generation available and there's a lot of capacity on the local or the regional power grid.

The problem is they're running out of those places to go. So, you know, they started, I mean, first we think of all the data centers in Northern Virginia, that's called, you know, data center alley, sorry, or Silicon Valley. A lot of data centers started there, but those places just couldn't handle the amount of generation that these companies needed. So they started spreading all over the country. We're seeing data centers pop up in Georgia, in Omaha, in

you know, in Wisconsin, you know, throughout the Midwest. And some of these places are equipped with the capability to bring on a lot of clean energy really quickly, wind, solar power. But in most cases, even where there is a lot of renewable energy available,

The AI companies are moving to those places because they have fossil power. They've got 24-7, you know, their gas plants and in some cases coal plants that can provide power around the clock.

And what the tech companies have been doing is they're using that power, but they're saying, no, we're building all kinds of renewable power and it's offsetting it. And it sort of becomes like this accounting scheme where they claim they're clean, but they're actually using a lot of this dirty fossil power. Yeah, it's an offset rather than an overall reduction, right, with replacing fossils with clean energy. You went to Omaha, right? Can you tell us more exactly about what's happening there? Yeah, the Omaha case is really interesting because you have a –

a power plant that has some coal-burning units in this community that's really a low-income minority community. And it had been promised that the coal plant would be shut down. And they would stop burning coal, and they'd at least switch it over to cleaner-burning natural gas, which still has emissions but not like coal. Then Meta came to town, and they built this just massive data center. And then Google also started building in Omaha.

And they just needed so much power and so much energy that the power company became overwhelmed. And they had plans to bring on all of this renewable energy, and they are bringing on some, but...

In the end, they needed to delay the closure of this coal plant in this environmental justice community. And the residents there are really upset because they have heightened levels of asthma and other chronic disease because they live in the shadow of this coal plant. And the coal plant was supposed to close, and now it's not closing. Can you remind me why that plant was scheduled to be closed?

Deployment was scheduled to be closed for a number of reasons. I mean, coal is becoming less economic also. So in a lot of cases, they want to move over to natural gas, but also it was scheduled to be closed for health reasons. I mean, it's causing problems in this community. And there was sort of an outcry and people wanted this thing shut down.

And that was the plan. It was going to be shut down. You know, when they looked at their forecasts a few years ago, they didn't need it. And then all of a sudden, the data centers needed so much electricity that now they do need it to keep the power grid stable in that part of the country. And they could and Meta and Google have many, many billions of dollars. Right. I mean, were they able to tell the power company or the operator that runs the plant, we'll pay you anything to keep this open?

You know, when you talk to these companies, they say, look, we didn't go to the power company and say, you need to keep this coal plant open. They keep their fingerprints off these fossil plants for obvious reasons, because in their business strategies and what they want to do, you know, they keep talking about we're going to be net zero by 2030. We're going to have our data centers be zero emissions. And it's like, well, how does this make sense? It's burning coal. And they say, well, no, we brought in all this, you know, the offset issue you talked about. We brought in all this solar energy and all this wind energy.

And they do try to put some pressure on the local utilities and local regulators to bring in more clean energy and close some of these coal plants. But at the same time, when the utility comes back and says,

you know, if we close this coal plant, we can't give you the energy you want. These companies, we're not seeing them put up a huge fight and say, no, absolutely, we'll delay the project. You need to close the coal plant. At the moment, they are in this arms race to develop AI as quickly as they can. And the feeling is, okay, we'll just deal with this energy now and we'll deal with the emissions later.

Well, Evan Halper is with us today. He has reported in depth about the energy needs for AI and how those needs are going to grow dramatically, even in just the next couple of years. So how will we meet those energy needs? And what are the environmental and health impacts of the fact that fossil fuels seem to be the place where that rapid energy need may be most quickly met? So we'll be back with more. This is On Point.

Support for On Point comes from Indeed. You just realized that your business needed to hire someone yesterday. How can you find amazing candidates fast? Easy, just use Indeed. There's no need to wait. You can speed up your hiring with Indeed.

Indeed is all you need.

Hey, it's Ben Brock Johnson, co-host of Endless Thread and director of digital audio at WBUR. WBUR, including yours truly, is headed back to On Air Fest, three days of podcasting that takes over Williamsburg, Brooklyn from February 19th through the 21st. Want to hang? This year's lineup has some sessions I can't wait to see, like CNN's Anderson Cooper talking with WBUR alum Adi Cornish, also In Polite Conversation with Anna Sale and Death, Sex, and Money, and

And because WBUR and On Air Fest are tight like that, WBUR has an exclusive 40% off discount code just for you. You can go to onairfest.com and use your code WBUR40. Don't miss out. You can grab your ticket right now. That's WBUR40 at onairfest.com.

On March 5th, tens of thousands of women will come together in one digital space for a national gathering of women supporting women. There is power in unity. And this Women's History Month, we invite you to join us.

Register now and gain access to some of the world's top women leaders and experts on career advancement, leadership, and personal development. Speakers include Oprah Winfrey, Padma Lakshmi, Gloria Steinem, Julia Louis-Dreyfus, Isabel Allende, and many more. WBUR is a proud media partner of this virtual event organized by the Conferences for Women, the largest network of women's conferences in the country. Learn more at maconferenceforwomen.org.

You're back with On Point. I'm Meghna Chakrabarty, and today we are talking about the massive energy demands that the growth of AI is going to put on the United States' power generation capacity and our transmission.

For example, there are estimates that by 2030, maybe even before that, AI, just AI, could consume almost 18% of all the power produced in the United States. So how are we going to do that? And what are the environmental and health and climate impacts of this technology?

quick ramp up regarding AI's very hunger, deep hunger for electricity. Evan Halper joins us today. He's been covering this for The Washington Post. He covers the energy transition. And Evan, we were talking about your reporting from Omaha. I just want to stick with that for another second here, because just to bring folks up to speed, it was in 2023 that Omaha Public Power District in Nebraska, of course, announced that the North Omaha Station was

would not stop burning coal as planned. You talked about that, Evan. And it's now going to keep burning coal through at least 2026. I guess that's next year. North Omaha resident Kay Carney told local TV station KMTV that she's worried about that. I also think it is a clean energy issue.

I'm concerned about the issue of climate change for my children and for everyone on this planet. And this plant's been around a long time, and I think it's time for us to move to cleaner sources of energy. So, Evan, about climate change and emissions, a lot of these tech companies, you had touched on this before, right?

in the past have committed to reducing their overall greenhouse gas emissions. I'm not sure how that's possible with their sudden and very sort of asymptotic need for energy right now.

Yeah, and if you look at the environmental impact reports that these companies are putting out, these are reports that they sort of disclose once a year where they say these are the things we're doing for the environment and this is our footprint on the environment. Their emissions are generally just on the rise and sharply on the rise. And they're saying, well, we're still going to figure this out and we're still going to get to net zero by 2030 or whatever year they have the goal for.

But the math isn't really working. When you look at the amount of electricity they're using and what they say they need and what they're building right now to power their data centers, all we see is more and more emissions. And it's kind of this magical thinking that there will be new innovations and they'll get new kinds of nuclear energy approved and there'll be all kinds of new energy sources soon enough and this natural gas will just be a bridge and eventually they'll be using much cleaner energy.

But that sort of ignores the fact that when you build a new natural gas plant, its life is usually 30 to 50 years. And you have what's called lock-in, where the plant is just going to be there. And even if the AI data center is not using it, something else is going to use it. And the investors in that natural gas plant are not just going to shut it down because the tech companies five years in decide they no longer want to use it. Yeah. Yeah.

Well, we're seeing here that Google, for example, says its greenhouse gas emissions have risen 48% since 2019. Microsoft says it's grown almost by 30% since 2020. And that's

It's basically all AI driven. So, Evan, hang on here for just a second because I want to bring Amy Myers Jaffe into the conversation. She's director of the Energy, Climate Justice and Sustainability Lab at New York University. Amy, welcome to On Point. Thanks for having me.

So let me ask you this. I mean, like, this is kind of a reality check moment here. While it would be preferable for all sort of new power generation that comes online to be clean or renewable, the truth is, is that due to regulatory constraints, due to, I guess you could say, how much capital it costs, due to the time it takes to build new things, that these companies are naturally seeking...

stuff that they can get online faster. And that happens to be fossil fuels right now. Is there any other way? Yeah, no, go ahead. Yeah. Let me stop you. I'm trying to provoke here. Go ahead. Okay. So fossil fuels are actually not faster.

To bring on a new combined cycle natural gas plant is going to take five years. To bring on new solar facilities, if you can get the permitting and the transmission set up, you know, is much, much faster. You know, could be, you know, 12 months or something like that. What about existing natural gas plants?

I think the point is, and so every utility in the United States has to file something called an integrated resource plan, right? And that is how they're saying they have to define load, which is what we call demand, right? So what's the profile for how much electricity is going to be needed in my service area? And here's how I'm going to meet it, right? So DOE last year, um,

said that across the entire United States, there'll be 20 new natural gas plants coming online for 7.7 gigawatts. So a giant data center could need something as much as a whole one gigawatt. So we have these seven, we have these, I'm sorry, 20 new natural gas plants coming online. And that was part of the planning.

That's a change in the sense that, you know, a few years ago we had no natural gas plants coming on. In 2023, six new natural gas plants came on. And that has to do with just specific locations. So sometimes you have a change in what the demand is going to be over time. I agree with the idea that if you build a natural gas plant recently,

the people who are building it intended to operate for 30 years. But what we do know in the United States is that there are coal plants and natural gas plants that did close. And they did close even though it hadn't been 35 years, but because those plants in that particular location were no longer economic to operate. So it really is site-specific.

And I think that these data center providers, these hyperscalers, are literally going around the country with a map place by place and trying to figure out where is there already extra electricity capacity. So where is there a place where all of today's electricity demand is not using up every resource on that particular grid? But to give you just an example, because I kind of feel like

We're so headed on this one theme and picking places where it's negative, we're escaping places where it's positive. And one of those places, which I think the listeners will find surprising, is that Texas is meeting its needs and not having brownouts anymore because they've used 9% or 8% of their current supply at peak is coming from batteries. Right.

You know, we happen to do a whole hour exactly about Texas. Right. And it's clean energy example that it's setting for the rest of the country. Correct. So my point to you is in 2024, literally 96 percent of the planned additions to the grid were from renewables. And I think the Trump administration faces this real challenge, which is that

The electricity we have in what we call the queue, like what we already had proposed to build...

because no one was proposing to build new coal. So what we had in the pipeline was mostly renewables. And then what's more recent is that you've got a couple of hyperscalers that made deals like Constellation Energy, which is where most of the data centers in the United States, you've got almost 34%

a trillion megawatt hours of demand coming from data centers just in Virginia alone. So in that case, they have done this deal, which you may have talked about, to restart Three Mile Island. Yeah, we're going to talk about that in a second. So let me just jump in here because

Because I definitely want to get to that. But so, Amy, there's one thing that you said pretty quickly that I'm going to pick out a little bit. You said if we can, you know, if permitting can happen efficiently and if we get transmission issues,

Right. Those are two enormous ifs. Right. Because, in fact, from that hour that we did about Texas and clean energy, one of the things that we learned was that right now the transmission issue is so huge in terms of like there is a backlog. Right.

Speaking of clean energy, there's a backlog of solar, wind and battery projects that want to connect to the grid. Right. They could provide something like, I don't know, I'm seeing one estimate here of 2600 gigawatts worth of capacity. That's according to the Lawrence Berkeley National Labs from the Department of Energy. That's twice the country's existing generation capacity. But the backlog exists because they cannot connect to the grid.

So that's a really big if. And from with that perspective, first of all, respond to that. And second of all, for as long as that transmission issue exists, I still wonder if it does actually make more sense for these tech companies to go to places where they're already connected to the grid. And some of those places are fossil fuel powered or generation facilities. So I can only go to a place that has existing fossil fuels and benefit from that if they're

is space for me. So again, it's region by region. I mean, North Carolina, for example, Duke did say they're taking a coal plant and they're converting it to natural gas, for example, because that was a quicker and cheaper way for them to get their integrative resource plan approved.

But it really, you know, depends on where, you know, the DOE and we'll see if, you know, those project funds stay. DOE greenlighted some transmission project, an independent transmission project to bring wind in Texas.

uh, West instead of, uh, back into Texas. Right. So that would be a case where you'd be bringing wind to New Mexico and Arizona, Arizona being another place where a lot of data centers are proposed to go in. So, uh,

So again, you know, it's very hard to sort of talk on the national scale. I think we're getting to a point in this country because remember, you also have new manufacturing going in a lot of places and maybe not out in rural areas, but you've got cities where electric cars are, you know, being used more and you have charging stations. So again,

There's a lot of different things that are shaping the load on the grid today. And one of the things that has not been used as much as it could be is that in places... So there's something called virtual power plants. So that's when I combine up batteries that maybe I could lease to someone to put in their home or in businesses. And I...

switch off their demand at peak hour to sort of make space for the load that has to be 24 and can't go off at all, which would be a data center. And so I was recently at a function, and I can't tell you which utility because it was Chatham House Rules, and they were explaining in this very important region of the country how they have this 15 hours of peak load they have to flex to

to be able to service their region. And it doesn't make sense to build a giant billion-dollar fossil fuel plant to do that. What makes more sense is to look across their whole system and see how they could use this sort of flexible software to lessen demand at particular times of day. Well, let me bring Evan Halper back in here. Evan, your response to the things that Amy's been saying.

So there are a lot of things going on, innovative things that these companies can do to make better use of the power that's already there or to use more renewable power. And definitely what Amy's talking about with these virtual power plants, you know, there's something there. And what the AI companies will often say is, you know, and you haven't even seen what we're going to be capable of doing once AI is further developed.

So if we put more resources into building these data centers, even if it means more emissions now, we're going to have better AI and it's going to unlock these solutions for better virtual power plants and more efficient power and everything is going to be better. But in the meantime, despite the availability of the solar power and these companies building lots of renewables, I mean, I don't want to discount that they –

what they're doing. They are using battery power. They're putting all kinds of money into the next clean energy innovations, but they're looking at what they want to build. They're looking at this AI arms race and these worries that China's going to get ahead of us

And they're mapping out how to get different projects in and how these data centers can be fueled. And they're still finding that even though they're willing to pay extra and they're willing to take all these additional steps to make the energy as clean as possible, their commitments to clean energy are just...

they're not this year what they were a few years ago. They have definitely changed course. I mean, I just got a response from Microsoft for a story I'm working on. I asked them some questions about what they're doing in Wisconsin with all of this gas power that the local environmental groups are saying, why are you using all this gas power?

And Microsoft has changed its tune. It's saying, you know, the realities of what we need to do for AI and to get all this power means we need to, we can't just think linearly and have only clean energy. We have to do other things. So, you know, I understand that there's all of these options out there, but I think these companies are looking at their balance sheets. They're looking at how quickly they want to move and they're feeling like they can't get it done unless they turn to more fossil energy. Yeah.

So you both will have to forgive me because when I get stuck on something that doesn't make sense, I get really, really stuck. And I need your help getting unstuck because I'm still focused on the fact that there's a lot of power out there that's simply not being able to be used right now. And again, I'm just going to repeat this fact from the Department of Energy's own lab, the Lawrence Berkeley National Laboratory, that says...

Right now, this backlog of power that's waiting to get into the grid is twice the country's existing generation capacity. Now, let's say in an ideal world, if we were able to get all that power into the grid, maybe only a tiny fraction of these new projects or these sort of rebooted fossil fuel plants would actually need to happen. But so we're talking about generation when it sounds to me one of the major unsolved issues is transmission, because I'm seeing here that since the year 2000,

So that's what, 25 years now? And we've built now no transmission. Yeah, one-fifth of projects requesting grid interconnection have been completed.

I mean, so yeah, go ahead. You have to let's deconstruct that because it's a whole series of problems. I mean, people are saying, you know, new Congress, they're going to look at this issue again. Of course, in the last year, there was a lot of effort to try to get a new permitting bill through that didn't succeed. And the president has said, you know, permitting new transmission and electricity is a national security imperative.

Right. So, you know, how do you do that? I think the problem we have in the United States, which, you know, we're kind of discussing here now, is that you have all these different regions of the country. And in some places, a state like Texas is in control of its whole grid. But in some places, no.

Like the Mid-Atlantic, you have regions where multiple states have to work together, and those states are not agreeing on what projects to approve and then how to disperse the costs. Remember, when we build transmission for the listeners, you know, when you look at your electricity bill...

There's not only the cost of the electricity you're using, there's also your share of the transmission. Amy, hang on for just a second. And Evan Helper, you as well. When we come back, I'll just have to live with being stuck on that issue because there's many more aspects of AI's power demand that we should touch on. We'll do that in just a second. This is On Point. On Point.

Support for AI coverage in On Point comes from MathWorks, creator of MATLAB and Simulink software for technical computing and model-based design. MathWorks, accelerating the pace of discovery in engineering and science. Learn more at mathworks.com.

and from Olin College of Engineering, committed to introducing students to the ethical implications of artificial intelligence in engineering through classes like AI and Society, olin.edu. Still getting around to that fix on your car? You got this. On eBay, you'll find millions of parts guaranteed to fit. Doesn't matter if it's a major engine repair or your first time swapping your windshield wipers.

eBay has that part you need, ready to click perfectly into place. For changes big and small, loud or quiet, find all the parts you need at prices you'll love. Guaranteed to fit every time. But you already know that. eBay. Things. People. Love. Eligible items only. Exclusions apply.

You know what's smart? Enjoying a fresh gourmet meal at home that you didn't have to cook. Meet Factor, your loophole in the laws of mealtime. Chef-crafted meals delivered with a tap, ready in just two minutes. You know what's even smarter? Treating yourself without cheating your goals. Factor is dietician-approved, chef-prepared, and you-plated. Pretty smart, huh? Refresh your routine and eat smart with Factor. Learn more at factormeals.com.

You're back with On Point. I'm Meghna Chakrabarty. And today, Evan Halper joins us. He's a business reporter for The Washington Post. He covers the energy transition and has been reporting extensively on the massive energy demand, electricity demand, that the growth of AI is going to be putting or is putting in right now on this country. Amy Myers Jaffe joins us as well. She's the director of the Energy, Climate Justice and Sustainability Lab at New York University. And as you've been hearing, one of the themes of this conversation is the fact that

Like it or not, a lot of the tech companies are looking to fossil fuels to meet their surge in demand for the data centers that power AI. Fossil fuels happens to be a great passion for President Donald Trump. And he spoke via video, a video link at the World Economic Forum in Davos, Switzerland, just last month.

Nothing can destroy coal, not the weather, not a bomb, nothing. It might make it a little smaller, might make it a little different shape, but coal is very strong as a backup. It's a great backup to have that facility, and it wouldn't cost much more money. And we have more coal than anybody. We also have more oil and gas than anybody. So we're going to make it so that the plants will have their own electric generating facilities.

President Trump last month speaking to the World Economic Forum in Davos. Now, we wanted to sort of get our plant our feet on the ground in a region where a lot of this growth is happening. And Evan, a little bit earlier, mentioned northern Virginia, which indeed is home to roughly 70 percent of the U.S. data centers today. And

And to keep pace with the growing electricity demand in Virginia, utility company Dominion Energy is building a 1,000 megawatt natural gas plant in Chesterfield County, just outside of Richmond. Dominion was saying, hey, we're going to build this on an unused site.

But, you know, as it turns out, they had this unused coal plant just kind of sitting there, which essentially, you know, means that they have all of the local permissions they need kind of baked into the property. Patrick Larson is a reporter with VPM News in Richmond, and he says it's not clear how much of the existing coal plant Dominion will have to tear down and how much it'll be able to retrofit.

This is what's called a peaker plant. So in theory, it would only be running like 40% of the time when, you know, electric demand is highest. So when people are getting home from work, for instance, and turning all their lights and appliances on. Dominion spokesperson Jeremy Slayton told us at On Point in an email that power demand in Virginia is expected to double in the next 15 years.

Slayton told local TV station ABC 8 News the company is not abandoning renewable energy sources. We're planning to build on those exponentially over the course of the next 10, 15 years with more solar, more offshore wind. But as I mentioned, offshore wind and solar are both intermittent resources. So the sun's not always going to shine. The wind's not always going to blow. We need to have that balance.

reliable, always available energy source. And that's the natural gas facility at Chesterfield. No gas plant. No gas plant. No gas plant. This is the group Friends of Chesterfield. It spent months protesting the plant. Nicole Martin is a board member with the group.

We've gone consecutively to the Board of Supervisors meetings for six months. We've had our own little meetings where we would meet at restaurants. We do not want this gas plant to be built in Chesterfield County. Martin is also president of the Chesterfield County NAACP.

And she says the new gas plant is planned for construction in a neighborhood with mostly Black and Brown residents. We have a local YMCA in that neighborhood. We have plenty of church groups and several schools. You know, the plant will have like significant impacts on the health of nearby residents. You

You know, a community that has already, you know, shouldered the burden of living in the shadows of a coal-fired power plant since 1944. And Martin says her group wants Dominion to explore solar, wind, and battery power instead, or explore it more seriously. The utility says it hopes to start construction on the new gas plant by 2026 and have it up and running by 2029.

So, Evan Helper, I mean, reflect on what the story of Chesterfield County says and how that sort of matches to the reporting that you've been doing. Yeah, we're seeing this all over the country. And Northern Virginia is an interesting case because what's happening there, I mean, it's sort of the canary in the coal mine. This is going to start happening a lot of, you know, other parts of the country where data centers are popping up. But neighbors have just, you know, in the last year or two, really started to organize and really express serious concerns about

all of the fossil energy that is coming in to fuel these data centers. And, you know, they're also concerned about the development of data centers themselves, which are starting to encroach on national battlefields. And, you know, there's just not that much real estate in a place like Northern Virginia, and they keep going further and further out. But, you know, this is a textbook case, and we're going to be seeing a lot more of it. And it just makes sense that the Dominion is trying to locate this thing

on a property where there used to be a coal plant because that's the easiest place to site it to get the permitting. The infrastructure is already there to connect to the grid. And so at the same time, those kinds of properties, as your report highlights,

are the places where they've been burning coal or they've been burning natural gas for decades already and people were sick of it and it's creating environmental problems and health problems. And here they are building new facilities that's going to continue that for decades.

Evan, let me just take a step back here because suddenly there was a basic sort of aspect of AI that I realized and energy that I realized I don't really know at all. Because we're talking about additional power generation and transmission in the United States, but people around the world are making AI queries, right? These tools are being used everywhere.

everywhere. So is, you know, Google, Meta, are they doing the same things in other countries? Yes, they absolutely are. And some places in Europe have actually passed, and Singapore, I believe, moratoriums on data centers connecting to the grid, like Ireland and

And got to a point where it was looking like something like 30% of their power was going to be used to power data centers. And the country was just like, we can't do this anymore. We can't afford this. They initially, I believe, welcomed all these data centers coming in. It's development, you know, grows the tax base. But it created so much stress on the grid that now they're putting restrictions in place. That's happening in other countries around the world also.

Wow. Okay. Amy Myers Jaffe, there's something that the Dominion spokesperson there said, which, again, I don't like to let these sort of statements fly without some checking here. He asserted that unlike...

natural gas or coal. When the sun doesn't shine and the wind doesn't blow, you can't really get power from wind turbines and solar panels, which I guess technically in that moment is true. But can't you store power generation in batteries and then have it available 24-7?

Well, so I think the question is, again, you know, it's like what time and to what scale? Okay, so typically in the United States, the time of the highest demand for electricity is between 4 p.m. and 7 p.m.

And so, you know, Dominion is not worried about how much electricity is needed at 9 a.m. or 2 o'clock in the morning. They're really all their planning and building really has to focus on this four to seven time frame.

And the question is, you know, how many batteries do I have to put in to accommodate that four to seven time period? Now, Texas has done a good job with that. California during the heat wave in 2022 batteries played a key role. You know, but California also had to call people through Twitter and other social media and say, hey, don't charge your car between four and seven. Don't use your hot tub between four and seven. Right.

your air conditioning not to the coldest setting between 4 and 7. And why is that? Well, the answer is, you can just know that living in daytime because the sun starts to set in different places between 4 and 7. And then wind tends to be windiest at night. But when you're talking about something like a data center where you really can't have the electricity go off,

Um, it's not okay if there's three days like happens in West Texas sometimes in certain places where it's just literally not windy at all. Um, and so you have to have some kind of

way to manage that. And I think the question that Americans aren't asking, because no one really understands how electricity works when they look at their bill, is would I be willing to pay less for my electricity if somebody told me between four and seven I had to turn a lot of things off? Am I willing to do that? Or would I rather they stick up

natural gas peaking plant. And I don't think we've had a good debate on that in the United States because people don't really understand how their electricity bill works. Right. And especially if you ask them, would I be willing to pay less, but I can't use as many appliances between four and seven, but it would keep my Google searches going.

Like, I think actually, like, I add that not tongue in cheek, but I mean, this technology has become integrated. Yeah, it's been integrated into our lives. OK, I am in a really pushing the envelope kind of mood today, Evan and Amy. So you're going to have to forgive me. Tell me if this is a stupid question. I mean, I love your forthright approach to this. So I'm ready for it. We're still thinking about power generation and distribution in terms of large scale grids. Is it accurate?

out of the realm of possibility to say, well, why don't these data centers generate their own power? Why can't they be off grid most of the time with their own solar, their own wind, their own battery, but then have the, you know, the more the public grid as as a backup?

So they're doing that. I mean, and they're talking about doing that. And, you know, you mentioned some companies are more green friendly than others. There's some companies that are looking at places that have what we called stranded natural gas. So they got a natural gas resource somewhere in, you know, Appalachia or in Texas.

where they don't have demand for that gas, so they could just pipe it to a plant that would be connected to a new data center. The problem is when I'm that data center and even a natural gas plant has to go down for maintenance or for renewables or whatever it is I'm having as my backup,

I was thinking the other way, right? So listen, if my grid is the backup, then I still have to have enough power because in the end, when does the grid go down? The grid goes down in bad weather. The grid goes down in a heat wave, right? So the data center backup,

Even if they're off grid, we're still worried about they also might need to come in on a crisis. So I think one of the interesting things that's also never talked about is most data centers have diesel generation on premises in case there is an emergency like you're describing where they just have to use their own power. And because of permitting and cost, they have that equipment there and they never turn it on.

So some people are saying, hey, listen, you know, we maybe need to rethink the whole backup system process for data centers because...

you know, should it be some other thing? Should it be cleaner? If we get to these new long form batteries that are being tested, long duration battery systems and they work, should we shift to that? So, and, and same thing with micronuclear or, or small scale, you know, people are talking about what are the technical solutions and,

And if I'm living in a place and the data center says, hey, what we call the phrase going behind the meter, so I'm going to not be connected to your grid. I'm just going to be here with my own generation. But I'll just connect to you in case there's an emergency.

You know, if I'm a user, let's use a test experience, and there is this emergency with the data center, and everybody's bill goes to $1,000 for that one day, and you want to charge me that, am I okay with that? So what kind of rate-based system do we have to have to make sure that doesn't happen? Yeah. No, I acknowledge there's complications, but I mean...

Come on, we got to be able to figure these things out. You know, you mentioned you and Evan both mentioned the word nuclear, which I feel like we should do a whole hour on in and of itself. But I do want to at least in the context of this hour, give it a little bit of a nod because it has been mentioned that last year, Constellation Energy announced that it will reopen a nuclear reactor at Mars.

Three Mile Island in Pennsylvania. It's a 20-year-long deal with Microsoft that will help power the tech company's AI systems. Now, of course, it was in 1979 that Three Mile Island was a site of controversy

the site of America's worst nuclear power accident, a partial core meltdown. And Constellation Energy CEO Joe Dominguez talked about the deal on Bloomberg Markets in September of last year. AI was the thing that just really drove the load projections to a point where you said, without nuclear, it's not going to happen. Evan, do you want to talk about this for a minute? Sure.

Yeah, absolutely. So basically, this Three Mile Island nuclear plant, there are two units in it. And the unit that had the partial meltdown, that's not the one that they want to reopen. The other unit had shut down because of economics. There wasn't enough the economic case for keeping it open without subsidies or additional subsidies. It just didn't work. The demand wasn't there. And then AI changed everything. Microsoft is...

is actually going to buy all the power from this reopened reactor if it actually does happen. You know, Constellation is very confident that it can reopen this reactor. It's never happened where you've actually shut down a reactor for good and then reopened it years later. But it's also happening in Michigan. There's a plant called Palisades, Western Michigan. They're trying to reopen that one. And they're talking about trying to reopen one in –

I think that's the extent of the plants that you can reopen. But, you know, it's the fact that Microsoft wants to take all the energy from a nuclear plant, which provides enough power for, you know, a medium sized city just tells you how much energy these companies need. Do you know, we sometimes just get stuck on things. What's so bizarre to me about this whole thing?

and it is a huge challenge, is that we're talking about AI. And in this country, we love to celebrate the tech industry as being full of the smartest minds in the world, right? I mean, like Elon Musk is there in Washington right now claiming that he knows more than all the civil servants in the United States federal government combined. Well...

it sounds like these smartest minds in the world are actually not pursuing the best long-term solutions, right? Like, why go straight to fossil fuels? Why try to restart nuclear power plants? Although, nuclear I'm on the fence about. Why

Why aren't they doing things like advocating for getting more of that sort of reserve power that's waiting there back onto the grid? I mean, Amy, we've only got a few seconds left, but are there solutions that aren't yet being rigorously advocated for or explored? I think that, again...

I mean, one of the new things they're looking at is whether you could do carbon sequestration with natural gas in certain locations and therefore make the natural gas cleaner. And people, I mean, these companies have worked for years

on battery and software solutions. Yeah, well, unfortunately there, I'm going to have to just take it back from both of you and hopefully we'll get closer to a solution in the next couple of years. Amy Myers Jaffe and Evan Helper, thank you both so much for joining us. This is On Point.