Running The Numbers On AI Energy Use

IRA FLATOW: You can’t help but notice that AI is increasingly everywhere, right? But all that number crunching needed to run AI models comes at a cost in many different ways. Joining me to talk about that and other stories from the week in science is Casey Crownhart, Senior Climate Reporter at MIT Technology Review. She’s here in our studios in New York. Welcome back.

CASEY CROWNHART: Thank you so much. Always great to be here.

IRA FLATOW: Nice to have you. OK. Let’s talk right about this. What are some of the costs that we’re talking about here?

CASEY CROWNHART: Yeah, I think a lot of people have heard this, that AI is this big energy suck. You’ve heard wild projections about it. It’s going to use as much electricity as small countries.

But the reality is that there’s actually still a lot that we don’t know. A lot of those estimates are educated guesses because companies don’t tend to reveal a lot of that information. So my colleague James O’Donnell and I set out to really put hard numbers to that. So we worked with researchers to actually measure the energy costs of AI, which turn out to be pretty significant.

IRA FLATOW: Let’s talk about that. For example, when I use ChatGPT or any of those things, we don’t notice it. But it costs energy to do that, right?

CASEY CROWNHART: Absolutely. So we looked at open source models. So ChatGPT, like I said, those tend to be very secretive. But for open source models, doing something like asking for recipe advice or a few jokes, it might not be very energy intensive at all. Based on our calculations and our work with researchers, it might be like running the microwave for a tenth of a second or something really small.

But there’s a big range. So if you’re generating an image using an AI model, it might add up to 8 seconds in a microwave. If you’re doing something like generating a five second video, it can be much, much more, running the microwave for maybe over an hour. So there’s a–

IRA FLATOW: Over an hour!

CASEY CROWNHART: –big range. Yeah, absolutely.

IRA FLATOW: Does it depend on how tough the question is, how much research it has to do on its own, or something like that?

CASEY CROWNHART: It can. Certain kinds of reasoning models that will go through a thought process a little bit more like what humans do can be a lot more energy intensive. We found that it tended to matter more the length of the question you’re asking and the length of the answer that you’re outputting, in the case of text.

IRA FLATOW: When you’re scaling up, though, AI is everywhere these days. It all must add up, I mean, all these little things. Right?

CASEY CROWNHART: Absolutely. And as we see AI get incorporated into all these products, increasingly, you don’t have to go looking for it. It’s in Instagram. It’s in all of these apps that we use.

One estimate from a national lab estimated that I could use 326 terawatt hours of electricity annually by 2028. That’s about the same as 22% of US homes today being powered for a year, just for AI.

IRA FLATOW: And does it matter where the energy is coming from?

CASEY CROWNHART: Absolutely So that’s another thing that we looked at in our analysis is where the data centers are will really matter. So somewhere like California, where there’s a lot of solar on the grid, a lot of wind power, hydropower can be kind of half what’s called carbon intensive, or basically half the pollution per electricity, as somewhere like Virginia or West Virginia tends to be more coal and natural gas heavy.

IRA FLATOW: Do they move data centers to where they think they can get cheaper electricity?

CASEY CROWNHART: Yeah. So when companies either build their own data centers or contract them out for use, that’s definitely one of the things that they’re looking at. And some of these companies do have climate goals. So some of them are also looking at that, trying to find cleaner sources of electricity, as well as just what’s cheap.

IRA FLATOW: OK. I’m an individual consumer. I want to cut down on the overall load on the grid or using all this energy and producing it. What should I do? Should I just cut back on my ChatGPT or perplexity usage? Or how should I help?

CASEY CROWNHART: I think of this a lot like how I think about climate action, which is that there are things that we all do that contribute to this, but ultimately, it’s kind of a bigger systemic issue. And I think the biggest thing that we can do is push companies to be more open about the energy usage of these models. If there is one thing, though, I would say maybe we could avoid video generation. Because that was, just in our analysis, so, so much more energy intensive than any of the other things that we looked at.

IRA FLATOW: So corporate or government, that has to go to there eventually?

CASEY CROWNHART: Ultimately, yeah. And especially, as I said, it’s everywhere. It’s almost impossible to avoid and will be in the future.

IRA FLATOW: Yeah. Well, let’s move on a bit. Because I know there was news this week about COVID vaccines. And the FDA has guidance, new guidance, on the vaccines. Tell us about that.

CASEY CROWNHART: Yeah. So this week the FDA announced that it will limit access to COVID-19 vaccines going forward. So only people 65 and over and those that are high risk will be able to get boosters.

IRA FLATOW: Did they give any reason for that?

CASEY CROWNHART: Basically, they say that they just want more evidence. So they want these manufacturers to submit new clinical trials to prove that there is a benefit to using these boosters for people who are younger and relatively healthy.

IRA FLATOW: And how do we compare to other countries?

CASEY CROWNHART: Yeah. This is something I didn’t really realize. In the US, I feel like a lot of people have started getting a COVID booster around when they get their flu shot. That’s what I’ve been doing, just at the same time, same arm, knock me out.

But a lot of folks, in the UK, for example, this is what they’re already doing. So in a way it is the US going along with a lot of the rest of the world and how they’re handling COVID boosters at this point.

IRA FLATOW: Sticking with these microorganisms, let’s talk about something really fascinating, not that this wasn’t fascinating. The research out this week about an unusual bacterium found on China’s Space Station.

CASEY CROWNHART: I know.

IRA FLATOW: I’m hearing the music.

[SINGS “TWILIGHT ZONE” THEME]

Doo doo doo doo beginning to play already.

CASEY CROWNHART: Seriously, outer space life, kind of. Yeah, so Chinese astronauts found this new bacteria. It’s never been discovered or it’s never been observed on Earth before in the Space Station. It was on cockpit controls, living in microgravity up there. According to China’s national broadcaster, the astronauts swabbed the Space Station in May of 2023, froze the samples, brought them back down to Earth. And they did analysis and found this previously undescribed species of bacteria.

IRA FLATOW: Now, before everybody goes crazy about some weird bacterium, this is not unusual, right?

CASEY CROWNHART: No, absolutely not. So this is very similar to other species of bacteria that are found on Earth. And it’s actually very, very possible that this bacteria does exist on Earth, and we just haven’t found it yet. There are billions of species of bacteria, by some estimates, that we haven’t classified, so very likely that it just happened to be found there first.

IRA FLATOW: Yeah. But it’s not something that– well, maybe it has. Is it evolving in the Space Station to the microgravity and changing?

CASEY CROWNHART: That’s a really good question. It’s hard to say, just having observed it. There are some characteristics about it that make it seem like maybe it’s adapted. For example, it can deal with radiation damage. But we’re not really sure exactly without knowing where it came from.

IRA FLATOW: Yeah. That’s just always fascinating stuff that we hear in space news. And other biology news this week, something related to one of my favorites, hummingbirds. I just put my feeder up, hoping they’ll come and drink out of my feeder. And hummingbird feeders are important in this story.

CASEY CROWNHART: Yes, absolutely. So scientists found that hummingbird feeders have perhaps been an evolutionary force for Anna’s hummingbird across the western US. So these researchers basically look back into the archives and looked at specimens of hummingbirds in museums. And they found that when feeders started to become more popular after World War II, these birds started, their beaks started to change shape in response, it looks like.

IRA FLATOW: That’s pretty quick, isn’t it? I mean–

CASEY CROWNHART: It’s so wild to me.

IRA FLATOW: –just a matter of decades.

CASEY CROWNHART: Yes. Between 1930 and 1950 there was a noticeable difference, which is about 10 generations of birds. They only live a couple of years.

IRA FLATOW: And it’s because I have a hummingbird feeder.

CASEY CROWNHART: Yeah, they want their beaks to be longer and wider so they can slurp up more food more quickly.

IRA FLATOW: Well, if I were a hummingbird, I probably would like that also.

CASEY CROWNHART: It makes a lot of sense to me.

IRA FLATOW: Let’s move on to something that’s not as humorous, and that’s the tough time we’re having for all sorts of federally-funded research projects in recent months. But there’s a story about how this could be especially noticeable in bee research, moving from hummingbirds to bees. Tell us about that.

CASEY CROWNHART: Yeah. This was a really wild story from Joanna Thompson at The Atlantic. So about 90 commercial crops in the US are pollinated by bees– apples, squash, almonds. And we’ve seen in recent years, there’s often every year, a pretty significant die off of those commercial honeybees. This year it was absolutely a record. So about 62% of commercial colonies died between June of last year and February of this year.

And so typically, the USDA plays a role in the overwinter where they help beekeepers do a lot of analysis. Look at was it parasites, was it bacteria, was it viruses and help them set up the new colonies the next year. The problem is now with all of these cuts, funding cuts, layoffs, there’s real concern that program isn’t going to work quite as well. And it is kind of on a delay this year.

IRA FLATOW: Because bees are very important to our agriculture, right?

CASEY CROWNHART: Super important. Especially for almonds, I didn’t really realize how significant it can be. Because they’re not self-pollinating. You need a pollinator. And so typically they truck in a ton of commercial honeybees to help.

IRA FLATOW: Yeah. I wonder how the natural bees are because all of these bees are European bees, descendants of European bees. They’re foreign bees. And we have natural bees living in the wild.

CASEY CROWNHART: Yeah. And a lot of species of bees are in trouble. It’s not just the commercial honeybees. We’re seeing those native pollinators in a lot of trouble as well. But this is just somewhere that there’s a very clear kind of economic impact when this happens.

IRA FLATOW: And then maybe somebody will notice once there’s an economic impact.

CASEY CROWNHART: Exactly.

IRA FLATOW: It’s not enough that the bees are dying. They have to have an economic impact.

CASEY CROWNHART: Unfortunately.

IRA FLATOW: One more bee-related story, and that’s about their interactions with flowers. Tell us about that.

CASEY CROWNHART: This was kind of funky research. And I’ll say it’s early, presented at a conference. But scientists say that now plants can hear the tiny wingbeats of insects and respond to them.

IRA FLATOW: Wait a minute. Wait a minute. Plants have ears?

CASEY CROWNHART: Not exactly, but they can detect the vibrations is the idea. A team of entomologists, sound engineers, and plant physiologists played recordings of buzzing sounds that are made by a small bee at snapdragons. And they found that these snapdragons, in response, appeared to increase sugar and nectar volume.

IRA FLATOW: So it’s a vibrational thing, as you say?

CASEY CROWNHART: Mhm.

IRA FLATOW: And because we feel vibrations in our bodies without our ears, maybe we can use that somehow.

CASEY CROWNHART: Yeah, it’s interesting. I think one of the key questions that the researchers have is whether it works in reverse. Are the plants, in turn, making some noise to help attract the pollinators or interact with them the other way? In that case, you could maybe see that if you could play that noise and attract pollinators, you could help the relationship.

IRA FLATOW: Wow. So you could maybe genetically engineer these plants to vibrate somehow.

CASEY CROWNHART: Wow. You’re on the next step.

IRA FLATOW: I’m always way out there, Casey.

CASEY CROWNHART: You’re always ahead of the curve. Ira.

IRA FLATOW: Casey Crownhart, thank you for taking time to be with us today.

CASEY CROWNHART: Thank you so much for having me.

IRA FLATOW: Casey is Senior Climate Reporter at MIT Technology Review.

Copyright © 2025 Science Friday Initiative. All rights reserved. Science Friday transcripts are produced on a tight deadline by 3Play Media. Fidelity to the original aired/published audio or video file might vary, and text might be updated or amended in the future. For the authoritative record of Science Friday’s programming, please visit the original aired/published recording. For terms of use and more information, visit our policies pages at http://www.sciencefriday.com/about/policies/