AI data centers are huge energy black holes, consuming as much energy as 30,000 homes – and their rapid growth is straining global grids. The numbers are astonishing: Sweden could see power demand from data centers roughly double over the course of this decade. In the UK, AI is expected to suck up 500% more energy over the next decade. And in the US, data centers are projected to use 8% of total power by 2030, up from 3% in 2022.
On today’s podcast, host David Gura and Bloomberg reporter Josh Saul discuss just what these insatiable AI data center power needs mean for local communities, energy prices, and efforts to switch to renewables to combat climate change.
Read more: AI is Already Wrecking Havoc on Global Power Systems
Bloomberg Audio Studios, podcasts, radio news.
Just west of Washington, d C. There's a pretty idyllic place with rolling hills and horse pastures.
It's really pretty driving around, really beautiful, lots of trees. Especially coming from New York City. I was like, Wow, this is so fun to get out of the office. Is really nice out here.
Josh Saul recently went to Louden County, Virginia. He covers energy for Bloomberg, and he says this area was mostly farmland until just over a decade ago. It was something that county touted in economic development videos.
One of the things I real egove about Louden County is it has rural parts and then urban hearts.
Quality of life is great here.
It's fresh air, there's biking mountains, the wineries, almost like a California scene.
Now. It has a population of about half a million people. It has the highest median income per county in the US, and it's home to Dulles International Airport and something we're going to see a lot more of in the years to come.
Louden County offers the easy access to power, tempered environment, terrific workforce to draw from, and of course connectivity, which is second to.
Nine easy access to power. Recently, Louden County has seen some new construction.
Josh says, big big buildings, like a couple of Walmarts stacked on top of each other, just stretching out. They look just like a flotilla of spaceships that kind of landed down right there.
These are massive data centers that now dot the Ladden County landscape. Each one is a big windowless warehouse.
It's pretty fun to walk around. They have server racks which are I guess about eight feet tall, little taller than me, which are just packed with servers which just look like the black box decks to your computer, but a little bigger and fancier with more blinking lights. And you know, they're just server racks. Kind of as far as the I can see.
One of these data centers, Josh says, can consume as much power as thirty thousand homes, and that electricity has to come from somewhere. But all these server farms cropping up haven't exactly been met with enthusiasm by Louden County residents.
I went to a community meeting where people were just mad.
At a certain point if we continue to approve all of these data center buildings. There will be a time when there won't be a park.
There won't be there were just so many data centers going up all around their homes, you know. And they live in a sort of kind of American suburban rural nice open fields and lots of trees and also strip malls and also gas stations and just kind of nice, leafy America, And all of a sudden, there were real big industrial look and data centers going in around around their homes and on their way to their kids' schools.
So we've really got to get a handle on what we're doing and think about where the power is going to come from.
As the AI industry takes off, there's a competition among places like Loud County for the economic development that comes with being a data center hub. But that distinction comes with some really big trade offs.
It just raises a lot of questions, how do we generate this much power? What does it mean for our climate goals? What does it mean for reliability and cost of power?
And that's becoming an issue not just in Northern Virginia.
It's the same issue that we're seeing all over the world. AI is really scrambling our power grids.
I'm David Gura, and this is the big take from Bloomberg News on today's episode, how AI's insatiable demand for energy is altering the landscape in hundreds of towns worldwide, putting a strain on local power grids and on commitments to reduce energy consumption to combat climate change. The promise of artificial intelligence has captured the global imagination, but that promise comes with a price. AI needs lots of power, way more than technologies that come before it. According to Bloomberg's Josh Saul.
The difference is that generative AI, instead of just storing your photos or you know, running very simple calculations or relatively simple calculations, can create new deductions, new inferences. And it's that creation, it's that inference of new things that requires so much processing power and therefore so much electricity.
And this is just the beginning right now. Some say AI is still in its infancy, and generative AI is expected to be a one point three trillion dollar market by twenty thirty two.
The demand seems to just be going up and up.
Globally, there are now more than seven thousand data centers with the capacity to consume a combined five hundred and fourteen tarawad hours of electricity every year. That's greater than the total yearly electricity production for Italy or Australia. They're just huge, massive energy black holes.
We had the CEO of a big energy company in last week and he told us that he for the first time got a request for a five gigawat data center. That would be you know, five gigawatts, that's five nuclear reactors. That's about the city of Miami, he compared it to. So that's just a huge amount of power.
As Josh mentioned, the power needs of AI are significantly different because of all that goes into training models. It requires way more computing power than what's used for cloud storage of data.
Generative AI really refers to inference. It refers to AI taking you know, data sets or information that it's absorbed and coming up with new things. So that could be as simple as you know, a silly song that I have it create to send to my wife, to new climate models, new business models, almost anything you can imagine.
And this development is of course not unique to loud In County. Data centers are cropping up all over the US. And around the world to keep up with demand.
I mean, in the UK, AI is expected to suck up five hundred percent more energy over the next decade. Here in the US, data centers currently use about three percent of our power. According to some estimates, it's supposed to be up to eight percent by twenty thirty. And Ireland is the place that really boggles my mind. They're using more energy for their data centers than they have renewable energy total. So it's basically, if you imagined, like all the clean energy in the country, it's all going just to their data centers. Ireland has attracted so many data centers from the big tech companies you know, Microsoft, Amazon, and others, that the data centers are forecast to consume a third of the country's energy by twenty twenty six. It's like you have houses, businesses, and data centers. It's just wild.
The issue is that this demand for electricity is not just burdening grids in places like Northern Virginia, it's also leading to some compromises in terms of where that electricity comes from.
Power companies have been happy to quickly add on or delay retirements in order to keep more power on the grid. That's something that's troubling because in a general sense, we've been excited to shut down coal plants, to not build new gas plants, for example, because both of those emit carbon, which is what's warming the world or causing climate change. But when you have big, new, one gigawatt data centers showing up, then you have the power companies or even their regulators saying, hey, that one gigawatt coal plant, let's keep that bad boy burning.
Coming up. The practical consequences of this surge and electricity demand from data centers in terms of the price you pay for power and the fight against global warming. As places like Laden County and rural Texas, Ireland and Malaysia struggle to keep up with how much power data centers need Bloomberg's Josh Saws as utilities are scrambling to get ahead of that demand. They're upgrading their systems, installing new transmission lines, constructing new power plans, and all of that comes with a cost. Are people footing the bill for all of this? I mean, are prices going up?
Experts tell us that the price of power does go up. I mean a simple supply and demand. Utilities like the one in Virginia are very focused on allocating those costs, with more going to data centers and less going to homeowners or to small business owners. But certainly it's the price of power is going up, and it's not.
Just the cost of electricity. All this demand on the grid is leading to delays of all kinds of other non data center related development here.
In the US. Even in other places, power companies are kind of stretching out the amount of time it takes to hook up to the grid. That's something that we haven't really seen that much before in the US, especially that you don't just get a power hookup when you want it. I mean, can you imagine building a house and the local power company says, hey, we might be able to hook you up to power until twenty twenty nine. Till then good luck, careful with the candles.
This is not just a problem in the United States. It's something happening in many other countries, and some places have decided that they need to slow things down.
It's really the issues as we describe to them in Virginia identically happening all around the world. You know, Ireland also had a moratorium. Malaysia also had a moratorium of several years on new data centers.
But a lot of these tech companies are now building data centers that they can't, at least as of yet, hook up to the grid. Existing infrastructure just isn't ready for it.
I mean without electricity, the servers are completely useless. I mean you could stack them up and kickflip over them, but you couldn't do anything with them without electricity. So certainly having wires to them, having a fishient generation, having a well balanced grid, all of that is absolutely crucial to the development of AI.
All of this is creating a conundrum for tech companies whose future rests on their ability to push forward generative AI technologies.
How are the.
Tech companies thinking through the realities of this. What are they saying about the challenge of getting the energy that they need to run these data centers.
Pretty much all of the big tech companies have set ambitious, impressive clean energy goals, you know, they almost to a company have said by twenty thirty, we're going to be one hundred percent clean energy for our data centers. But it's a difficult thing to do because data centers are on all the time, one hundred percent of the time. So to match that to clean energy is hard.
Do they the tech companies believe that kind of traditional green energies. Do they think that green energy is going to be enough to make up the difference that they need.
Well, everybody loves nuclear, I mean nuclear has gotten so hot. Like Joe Rogan talking about nuclear. Oh, it's all we need to have enough water for California. Just a bunch of desalination plants next to nuclear and we'll turn the whole valley green. Bill Gates talking about it now everybody. Yeah, everyone's very excited about New clear it's baseload power, but the truth is that it's extremely expensive and takes an extremely long time. The only big nuclear plant that has been built in the US in recent years was finished recently in Georgia, and it was the over budget and ten years behind schedule. I mean, it's just tremendously expensive and difficult to build, and that makes it really hard to count on nuclear as a big part of what's going to power these projects. But again, our grid as it was set up, even when I started reporting on this maybe a few years ago, the idea was that we would be shutting down the fossil fuel generation in order to lower carbon emissions, and when you add new load onto that, it just makes that job of shutting things down really really hard.
For now, in Loudon County, there's no sign the development of new data centers is going to stop or slow down, and that's made its story something of a cautionary tale for other counties nearby. Josh recently attended a community meeting in Prince William County, which is just next door to Louden, and it's already starting to see server farms crop up in its horse pastures.
You know, about a dozen people stood up in each of them saying real clearly, you know, we do not want to be like Louden County. Prince William is different.
Prince William County is in a hole. We are in a hole like Louden County, and we are in a hole like the state of Virginia and many other places throughout the country. At some point we need to take stock of how much infrastructure is going to be required.
We do not want this. This is a dystopian nightmare.
And we cannot and should not support any more data center expansion from what is already permitted within the overlay. Due to the crippling energy demands of the current data center infrastructure in this county.
Already, and the supervisors listened to the whole thing and then voted almost unanimously to allow the larger data center.
It's going to be the first vote in a domin no effect of data center sprawl that you cannot control, that balls out of your hands, and that the public rightfully blames you for because we will hold you accountable for it.
It reinforces this idea that even though AI is still in its early days, because of how incredibly fast it's growing and how much money is behind its development, it's very hard to stand in its way.
I think there's a huge amount of momentum for more data centers, and I think all of us vote for more data centers. Every time we have a more sophisticated medical procedure done at a hospital, every time we order something online, every time we send a Miley Cyrus meme to our wife. It's all running through the data centers. It's all running through our phones and our data centers and our computers and our employers. And I think it just means more and more data centers around the world and around the US, and on your drive to work.
This is the Big Take from Bloomberg News. I'm David Gura. This episode was produced by Adriana Tapia. It was edited by Stacy Vanick Smith and Seth Figgerman. It was mixed by Alex Sigura, who also fact checked this episode. Our senior producers are Kim Gittlson and Naomi Shaven, and our senior editor is Elizabeth Ponso. Nicole Beemster Borr is our executive producer. Sage Bauman is head of podcast. Thanks so much for listening. Please follow and review The Big Take wherever you get your podcasts. It helps new listeners find the show. We'll be back tomorrow