Why AI Is Making Things More Expensive

Electronics are getting more expensive—and the culprit has a lot to do with generative AI. Amid all the talk of AI taking over people’s jobs, automating people’s workflows, and doing people’s homework, we’re learning that actually scaling this technology is a lot harder, and more expensive, than it seems. On this week’s episode, “Galaxy Brain” host Charlie Warzel talks to his Atlantic colleagues Hana Kiros and Alex Reisner about why AI is such an engineering disaster, and how this memory shortage is impacting everything from MRI machines to smart fridges to kids’ access to technology in schools. Get more from your favorite Atlantic voices when you subscribe. You’ll enjoy unlimited access to Pulitzer-winning journalism, from clear-eyed analysis and insight on breaking news to fascinating explorations of our world. Atlantic subscribers also get access to exclusive subscriber audio in Apple Podcasts. Subscribe today at TheAtlantic.com/Listener. Learn more about your ad choices. Visit podcastchoices.com/adchoices

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  Tired of rage scrolling through AI slop and clickbait just to get a handle on today's news? Meet Inpress, the algorithm-free news app that makes it fun and rewarding to stay informed. Track your media literacy and rediscover reality. Download today at inpress.app. That's I, N as in news nerd, press.

• 0:19Speaker 01

  When was the last time you indulged your desires, felt true Range Rover refinement? The last time you felt total serenity, total confidence, no matter the terrain? Limitless, effortless, peerless? How far can Range Rover take you? Range Rover, designed for distinction.

• 0:35Speaker 02

  This era of dirt-cheap consumer electronics seems like it's at least temporarily over. I think that we might just enter sort of like a wartime era, like the way that during World War II, we had like meatless Tuesdays or something. You know, people might just be rethinking their tech consumption.

• 0:56Charlie Warzel?

  I'm Charlie Worsell, and this is Galaxy Brain, a show where today, we're going to talk about why electronics are getting more expensive and what all of that has to do with AI. Now, if you've been in the market for any kind of consumer electronics recently, you've probably noticed that the price tags looked a little different this summer. Apple has raised prices on its MacBooks and iPads. Game consoles like the PS5, they cost more today than when they launched over five years ago. Samsung has raised the price of its Galaxy smartphones. And Apple is expected to do the same with iPhones. All of this has analysts predicting that the primary smartphone market could decline by 14.8% in 2026, which would be a record drop. Supply chains and manufacturing are complicated, but in the case of these rising prices, the

• 1:41Charlie Warzel?

  culprit is actually somewhat obvious. Everything is getting more expensive because we are in the middle of a memory crisis. Now, you've probably heard of RAM, which stands for Random Access Memory. It's basically your device's short-term memory. And there are different types of it. There's dynamic RAM, there's static RAM. But what you really need to know for this is that it is an essential component of TVs, computers, smartphones, gaming consoles, even many modern cars. And the price of these components is going up sharply. The prices of certain memory chips have tripled, even quadrupled, in the last trimester. And the reason is the generative AI boom. AI companies are so hungry for memory that manufacturers can't keep up, meaning that there are production shortages and there's no real end in sight.

• 2:24Charlie Warzel?

  Carl Pei, the founder of the hardware company Nothing, summed it up very succinctly earlier this year. The era of cheap silicon is over. Now, if this is actually the case, it would represent a fundamental shift in how we use electronics. iPhone and Xbox buyers, but also huge corporations, all of us, we have gotten used to purchasing electronics in similar ways, upgrading every few years when our machines get sluggish. Manufacturers build their technology with the expectation that the bulk of consumers will update to devices that are more powerful, more efficient, and have better memory. And this may sound like a luxury to you, but the point is that it isn't. The modern computing paradigm is one of uninterrupted progress. The AI buildout, it might threaten all of that.

• 3:07Charlie Warzel?

  The AI memory crisis, or as some people are calling it, the RAMpocalypse, is an emerging story. You may not be seeing or feeling it yet, but there are chances that you will very soon because so many of the devices that we use every day are affected. Now, to better understand precisely why this is happening and to chart the ways that the memory crisis just could ripple through the economy and our lives, I've invited on two of my colleagues who've been reporting on elements of this story. Alex Reisner is a researcher, a programmer, and a staff writer here at The Atlantic, who investigates how artificial intelligence systems really work. Hannah Kiros is an assistant editor who's been writing about these price increases and the unexpected ways that they may be ushering in a new era of computing. And they both join me now to talk about it all.

• 3:55Charlie Warzel?

  Alex, Hannah, thank you for coming on Galaxy Brain. Welcome.

• 3:59Speaker 04

  Thanks for having us.

• 4:00Charlie Warzel?

  So we are here to talk about the memory crisis, the RAMpocalypse, give it a name, whatever you want to call it, and specifically how this is causing a lot of things to get more expensive. But Alex, I want to start at the origins here, very basic steps. Can you walk us all through how generative AI works? Like, what are these, what do these large language models require to operate?

• 4:25Speaker 04

  Yeah, so large language models require a lot of language to operate. They're trained on millions, tens of millions of books, tens of millions of research papers, all the text that these companies can find, they use to train these models. And basically, what they've discovered is that the more of that text they use to train the models and the more of that text the models store, meaning the larger the models are, the better they seem to work.

• 4:54Charlie Warzel?

  So talk to me a little bit about the mechanics of that in terms of, I think a lot of people are seeing news about data centers, how these data centers work. Can you talk to me about the nuts and bolts of the chips, the storage, the intensity of actual, like, hardware resources in creating these models and training them?

• 5:18Speaker 04

  Yeah, so because the model, because the industry's strategy has been to make the models bigger and bigger, they need more and more resources, right? Like, they require a particular kind of memory, this high-speed, high-bandwidth memory that can process just enormous amounts of data quickly. And they also need to increase the number of data centers dramatically. And so we're actually, what I've seen from other people's research is that they're planning to increase data center capacity by eight times what it currently is, which is just a massive increase compared to, you know, the previous 20 years.

• 5:57Charlie Warzel?

  So, okay, you have these data centers, you have these large language models, they're very resource-intensive. You wrote a story about the AI engineering disaster for us recently that everyone should go and read. Your story notes that these companies, quote, "may be purchasing 70% of the world's supply of high-end computer memory." Now, that's a problem in its own right. We're going to get into that with Hannah in a second. But this is also, as you argue in the piece, an engineering disaster. You write, "The problem with generative AI, in the industry's own jargon, is that it doesn't scale." What doesn't scale here?

• 6:31Speaker 04

  Yeah, so, I mean, over the past 30 years, we've seen other technologies rolled out, like major technologies. We now have kinds of devices that we didn't have 30 years ago, about smartphones, tablets. Generative AI is really unprecedented in that it is engineered in a way that it's like, it's like the polish hasn't been put on it in a way. It's kind of in the stage that usually products that are kind of still in testing would be in.

• 7:01Charlie Warzel?

  Why is it not polished? What about it isn't polished?

• 7:04Speaker 04

  I think the simplest thing to say is that it's really inefficient. And what I mean by that is two different things. So the first is it includes algorithms that just don't scale. And by scaling, what we mean usually is, scaling is what venture capitalists look for in startups, which is the ability to add new users and to grow really large without that costing a ton of money. And so they're looking for companies that can build products that, for every additional user added, it actually costs less to support that user. And generative AI is the opposite. The more users are on the system and the more input they have, the more resources and time and money they consume. So that's the first thing.

• 7:47Speaker 04

  And then the second thing is, as I said before, the industry has decided that these models should be as big as possible, that they're better and more capable when they're larger. And so they're just being given more input, they're made larger and larger. And so the combination of that with technology fundamentally that does not really get larger gracefully, that consumes exponentially more resources, is very bad.

• 8:10Charlie Warzel?

  So there's this chart in your piece from Epoch AI, this organization that tracks the operating costs of major AI models. And they use several public AI models to show the exponentially increasing costs of serving more tokens. Tokens are the words that users type to chatbots, you know, the number of words in a response. And when you think about the enterprise AI companies, or the enterprise companies that are using this AI stuff that are, quote-unquote, "token-maxing," right? They're trying to use as much as possible. It seems to me like this is, this is costing more and more and more money. Is there any way to, like, to solve for that?

• 8:56Speaker 04

  I don't know. I think, you know, one of the biggest and most important mysteries of the AI industry right now is how much it actually costs to run these systems. Like, there's subscription prices if you use these systems. There's like, you know, $20 a month or something like that. We have no idea what it's actually costing these companies to run the systems. Is there a solution? Potentially, but the companies have been working on this for years already. It's, you know, they're well aware of what I'm saying in the piece. This is not news to anyone in the industry, but it's an extremely hard problem to solve, and they've made small progress here and there, but they are, my understanding is that they're nowhere

• 9:38Speaker 04

  near addressing the fundamental problem of exponentially scaling algorithms.

• 9:44Charlie Warzel?

  So, Hana, Alex's reporting here has established these companies are building a technology. It's incredibly resource intensive, inefficient. Now we get into the second part of this, which is, what is this demand from AI companies done to chips and computer memory?

• 10:00Speaker 02

  Yeah, so for a long time, a company like Apple could sort of twist arms and get the best deals possible for memory because they were the first and most important customer in line. So, memory manufacturers would say, okay, you know, everyone's going to buy, you know, the new iPhone, and so I'll give Apple a really cheap deal on this part because I know I'll, you know, make a lot of money off of this. But now, the hyperscalers have come in and they're approaching memory companies and saying, basically, we'll pay as much as you want for as much memory as you'll give us. They just are a much more valuable customer than the consumer electronic companies that are building the products that sort of run

• 10:45Speaker 02

  our daily lives. So, what's happened now is more of that manufacturing capacity is going to the hyperscalers, and now only the biggest consumer tech companies are even able to get memory chips at the quantity that they need to fulfill their orders. And they're paying more for it because now, you know, there's, one person that I was speaking to about Apple described them to me as like an 8,000-pound gorilla in the supply chain, but now there's an even heavier gorilla in the hyperscalers, and they're setting the price, and now the price is just higher for everyone. And as those build costs go up, you know, that's being passed on to the consumer, and now the stuff that we buy is costing more, too.

• 11:26Charlie Warzel?

  Yeah, you wrote this great article for us in July, calling this essentially an AI tax on a lot of consumer electronics, right? Describe to me what's happening. Let's use the MacBook as an example of what's happening at a specific, very important consumer electronic.

• 11:41Speaker 02

  Yeah, so the cheapest computer that Apple offers, the MacBook Neo, is now $100 more than it debuted at quite recently for the same device. The base iPad model now costs 30% more. iPhone prices haven't increased, but the prediction is that the iPhone 18 Pro will cost $200 more than the previous model, and some analysts that I've spoken to even expect the iPhone 17 to cost more, even though it debuted, you know, last season for the same exact device. So, what we're seeing is just, you know, Apple was sort

• 12:23Speaker 02

  of one of the last companies to be hit, but we see this across like Dell, Lenovo, every single gaming console costs more. Some indie gaming companies have shelved plans to make consoles because the margins don't make sense anymore because memory prices have pushed build costs up so much. This has been going on for a while, but the Apple price increases sort of raised people's eyebrows because the Apple ecosystem is really sticky. So, like, parents that have iPhones raise iPad babies, and then they buy their iPad Babies Max when they go to college. And so, these like $100 price increases across all of these devices, like around Christmas time and back to school season is when we really expect people to start to feel it.

• 13:06Charlie Warzel?

  And Alex, you wrote in your reporting that hard drives that you bought just two years ago for $350 each are now, when you look, they were $800, right? And that these prices are just out of control. What are you seeing across the industry when you look at it?

• 13:23Speaker 04

  Yeah, I mean, I need those hard drives for my reporting. I bought a whole stack of them, and I, two years ago, and I'm now sitting on a small fortune in hard drives, which would be funny if I could sell them, but it's really bad because, you know, I may need more of them. Everything is going up in price. It's not a time to buy computers. And unfortunately, it seems like these companies, the manufacturers, are not really very close to solving the problem. It seems like the shortage is going to continue for years.

• 13:51Charlie Warzel?

  This is exactly where I want to go with this, because I can imagine people are listening and they're wondering in some sense, kind of what the big deal is, right? Why these companies can't like get it together? Why can't we just make more chips? I'd love, Hana, and then Alex, if you have something to add here. What in basic terms is the fabrication process like for this? Why can't they just make more chips?

• 14:17Speaker 02

  Well, reading about this, I was like, wow, humans are so amazing. We can do really cool things because the fabs that they use to make memory chips are like many orders of magnitude cleaner than like a hospital clean room. A speck of dust can ruin millions of dollars worth of product. So, like, these are just incredibly complicated manufacturing processes, and they're only, like, most of the world's chips are made by, memory chips are made by three companies. And the lead time for making a new fab is like three to five years.

• 14:52Charlie Warzel?

  Fab being a shorthand for fabrication facility, which is where these memory chips are made.

• 14:57Speaker 02

  And those efforts are like in place, but also memory companies, they're sort of cautious because, let's say, demand disappears, you know, the AI boom is a bubble, then they've spent billions of dollars building these fabs that, you know, now aren't operating at full capacity. So, the expectation is that until 2030, we won't have the capacity, like the manufacturing capacity needed to really drive costs down. And another thing that I've been thinking about is just that during COVID, prices went up because the world stopped and supply chains were really disrupted, and they've never really gone down. So, there's kind of a tendency for like, you know, line go up, like, people

• 15:41Speaker 02

  get used to paying higher prices, and like, even when that extra manufacturing capacity comes online, I kind of find it hard to believe that Apple will say like, oh, okay, you know, and they'll lower their prices. So, prices might be higher from now on.

• 15:56Speaker 04

  It's also made worse by what people call the end of Moore's Law. Moore's Law was this kind of observation that was made in the 60s that computer hardware was getting significantly faster and cheaper at a pretty steady rate. And that was true from the 60s until about 10 to 15 years ago. And components have gotten so small at this point that they're having a really hard time shrinking them any further. We got used to computers getting always continually faster and cheaper, and that's really not happening anymore, that kind of automatic progress that we came to expect.

• 16:39Speaker 00

  If you're trying to get news today, you'll find a lot of biased algorithms, rage bait, and AI noise. Thanks, social media. That's where InPress comes in. It's a social platform on top of factual journalism, not the other way around. Track your reading and earn points. It's like Strava for your brain. Plus, you can connect with other users in your city who share your genuine curiosity. Ditch the echo chamber for reality with InPress. Download it today for iOS or Android at inpress.app. That's I, N as in news nerd, press.

• 17:12Speaker 01

  When was the last time you indulged your desires, felt true Range Rover refinement? When was the last time you drove in total serenity, the outside world reduced to a hush? The last time you were confident on uncertain terrain? The last journey that made you feel at home, conveniently connected at every step? In a Range Rover, pioneering innovation meets leading-edge luxury to make every drive and occasion. Limitless, effortless, pureless. How far will you let it take you? Range Rover, designed for distinction.

• 17:43Charlie Warzel?

  Alex, you quote one AI engineer who told you that the idea that one must rely on massive foundational models trained for millions of dollars by some big corporation in order to achieve success on hard tasks is a trap. What is the trap here? Do these AI models need to be built this way? Is there a way beyond the LLM, you know, training, feeding, more, more, more paradigm?

• 18:05Speaker 04

  A lot of people in the industry, or some people in the industry, it's hard to know how many, but there's certainly a bunch of developers that are trying to build smaller models that scale better, that take fewer resources. None of these models that I've seen so far are really replacements for large language models, but they are good at solving certain problems and some of them have been put into use already in places where companies are using other kinds of AI models. So there is, there are kind of grassroots efforts to create different types of AI. If you look back historically, you know, AI was about trying to figure out how people

• 18:46Speaker 04

  use their own brains and translate that into code, trying to like simulate human reasoning that way. And those, the kind of old approach that was used in the 60s and 70s was something that scaled better for the most part. And there's some people now that say we should go back to that method or, you know, combine that method with the current language model method. But I think there is so much momentum behind the chatbot products essentially, that there is not much will within the industry to try to figure out how to do things differently. There's so much money coming in to support this approach, even though it's so bloated

• 19:29Speaker 04

  and inefficient, that I think the industry is just, is not that interested in changing course.

• 19:35Charlie Warzel?

  So Hana, so let's say we stay on the same trajectory here. Walk me through what this period looks like. How, how might this play out across different areas?

• 19:45Speaker 02

  Yeah, so I think it's, it's already hitting certain parts of society harder than others. So I've spoken with IT managers for like public schools and hospitals. And I was speaking to a guy who works with a school district in Missouri. And he, he was saying that basically they're starting this school year in the red because they are paying more for per student, per device. And, you know, they're getting less money from the Department of Education. So these, these places where you have to buy bulk tech and you can't just, you know, charge students more for their education, like it's, it's creating headaches for the school system.

• 20:28Speaker 02

  The school system, they give kids when they enter middle school and when they enter high school, like Chromebooks. And he mentioned that the school district is considering even just like, you know, maybe the kids don't need a laptop during middle school because we can't afford to do that for them, you know, if, if this, if this continues. In healthcare, things like MRI machines use a massive amount of memory. In my reporting, I reached out to an MRI manufacturer and they were like, yeah, we might have to charge hospitals more for these machines. If, if these prices don't change. A healthcare consultant I spoke to said that the, the hospital system she works with, they, they were installing basically these iPads, these tablets in, in hospital rooms that would allow people to like order their lunch and also see all of their vitals and

• 21:12Speaker 02

  like, you know, their health charts. And, and they stopped the project because, you know, those tablets require memory and now that costs, it costs way more to buy them, you know? And so in that way, you know, we're already seeing the effects of the memory shortage. But there's also this idea of like, we've talked a bit about this AI austerity, the idea that I don't like paying more for, you know, something that I could have bought a month ago for way cheaper. Right? So maybe I hold on to my tech for longer. Maybe I decide to buy, you know, something used, like after the Mac and iPad price increases debuted, there was a giant surge in looking at resale markets for those products.

• 21:52Speaker 02

  So, you know, I, I think that we might just enter sort of like a wartime era, like the way that during World War II, we had like meatless Tuesdays or something. You know, people might just be rethinking their tech consumption. This era of dirt cheap consumer electronics seems like it's at least temporarily over. And I think that people's habits will probably change.

• 22:13Charlie Warzel?

  I think when people think about the ripple effects, they don't often think of the ways that, that, that chips and, and memory is now just, just in everything. We're not just talking about MacBooks, Chromebooks for kids and, and iPads. And, you know, the MRI machine is a great example of this. Also, we've talked before offline about cars and things, things like that. You know, cars becoming basically just like computers on wheels, driving those prices up, changing, rippling, you know, through, through the market, affecting used car sales, all that stuff. So it seems like it is, it is something that is not just actually confined anymore to

• 22:56Charlie Warzel?

  what we would, you know, traditionally call the consumer electronics space. But Alex, I want to go back for a second here to the, the inefficiencies of the language models. Because you've reported and you alluded to earlier that, that some of these AI companies have found techniques for improving performance, but they've also not yielded significant gains. If it does improve, if they do find some ways to, to make some real significant gains in terms of the efficiency, do you think that that changes the resources, like the, the intensive amounts of resources that these demand, like put a little more slack in the system in terms of the, you know, like the, the supply chain of all of that?

• 23:33Speaker 04

  I mean, if they were to solve the exponential scaling problem, absolutely would change everything completely. I think AI would become a profitable business, but I, I think it's, it's sort of at this point, just a fantasy. That's a, that's a hypothetical scenario. There's no evidence that that's coming anytime soon.

• 23:49Charlie Warzel?

  Part of the problem here is also outside of the models themselves needing more computing power, our devices need more power too. Right? Like they are, they are integrating AI technology and the features inside of them. So Hana, is there a chance that our computers and phones will essentially have to, you know, dumb down in order to work if, if all this continues?

• 24:12Speaker 02

  When memory chips were, were dirt cheap, there was a big push to make everything smart. So like you can have a smart fridge and like a smart toaster. And it was just sort of a cheap way to justify charging more for your product. You know, I can imagine a dumb Renaissance, basically where we just, you know, like maybe your device doesn't need to be connected to the internet of things to be good. What's really being hurt by the memory price increases? These ultra cheap Androids that have razor thin margins that like the majority of people in like Africa, people in India use. Some Chinese companies that make those uber cheap smartphones, they've just gotten out

• 24:54Speaker 02

  of the business entirely because the, the margins don't make sense. But for a lot of people, like you access banking through your smartphone, like A groups during famines, they like identify who they reach through, you know, smartphones. So I think a lot of tech can be dumbed down. Like I don't, I don't need a, you know, a smart fridge. But I, I do worry about smartphones being, people having to go from smartphones to dumb phones involuntarily because of price increases, because a lot of those ultra cheap Android phones no longer make economic sense, given the memory crisis and, and the price increases we've seen.

• 25:31Charlie Warzel?

  I think it's a great point to note that this is, this will play out differently in different areas of the world. For some people, it will be an issue of, I really want to upgrade because I like, you know, the new camera on the iPhone, whatever, 18 versus I no longer can buy the phone I need to access a lot of, you know, very basic services that, that we might take for granted here in America or in Europe or someplace like that.

• 25:59Speaker 02

  We're kind of already seeing it. Global shipments of smartphones are down 11% and that's the lowest they've been since 2013. So like, there is already like a historic dip in smartphone buying. So I guess, you know, it seems like the price increases are marginal, but people are like reacting to them.

• 26:17Charlie Warzel?

  There's a brutal irony here, I think, which is that these AI companies are spending unfathomable amounts of money and they're trying to build what they believe is a super intelligence or just very, very powerful models. And yet this technology that they're building is, you know, theoretically driving up the price of the gadgets so that they may become both incredibly smart and less accessible to people. Like Alex, does this to you jeopardize the whole project that these labs are trying to, trying to infuse generative AI into everything, but the everything becomes harder to, to get?

• 26:56Speaker 04

  Yeah, you know, it's hard for me to tell like what their plan really is. I think we would all like tools that help us do our work and make the world better. I'm not really sure what these companies are doing. Right? Like they primarily want to make a profit. In order to make a profit, they are advertising generative AI as something that is a useful tool. I think if you were to look at the, the actual cost of AI, which we're really starting to see now in a, in a very concrete way and compare that to the actual benefits of AI so far, it's a very strange ratio. Again, I think sort of unprecedented in the history of, of the tech industry.

• 27:41Speaker 04

  It's just a really weird project that I think is motivated in part by just the idea of artificial intelligence and wanting to build an artificial human, which is something that people have wanted to do for thousands of years. I think there's an almost religious need to just continue that, that project. And if getting funding for that means saying, hey, this is really useful. Everyone should, should have this. We're going to put this in everything. I think, you know, that's, that's how these, these companies are making money. It's just by pushing, pushing the technology into everything.

• 28:22Speaker 00

  If you're trying to get news today, you'll find a lot of biased algorithms, rage bait and AI noise. Thanks, social media. That's where Inpress comes in. It's a social platform on top of factual journalism, not the other way around. Track your reading and earn points. It's like Strava for your brain. Plus, you can connect with other users in your city who share your genuine curiosity. Ditch the echo chamber for reality with Inpress. Download it today for iOS or Android at inpress.app. That's I, N as in news nerd, press.

• 28:55Speaker 01

  When was the last time you indulged your desires, felt true Range Rover refinement? When was the last time you drove in total serenity, the outside world reduced to a hush? The last time you were confident on uncertain terrain? The last journey that made you feel at home, conveniently connected at every step? In a Range Rover, pioneering innovation meets leading edge luxury to make every drive an occasion. Limitless, effortless, pureless. How far will you let it take you? Range Rover, designed for distinction.

• 29:25Charlie Warzel?

  Hana, AI companies are not exactly beloved by the general public right now. Do you think that this is going to lead, or have you seen already that this is leading to like a consumer or actual, you know, person on the ground backlash?

• 29:42Speaker 02

  Yeah, I think it already is. In like the subreddits that I lurk in, people are like, you know, I can't build a PC anymore because, you know, of the quest to build a machine god. I think back to school season is when a lot of people will feel this, you know, as they're trying to buy like new tech for their, for the kids going off to college.

• 30:06Speaker 02

  Or when, you know, the price increases hit the next iPhone and people are thinking about buying a new one. And I think part of why it feels weird is that when you talk to people about like, you know, will AI, you know, cure all disease and free us from labor like, maybe, but like, it just, it just feels very amorphous. And I think that when it hits like sort of the tangible digital stuff that helps run our lives like that, like, you know, people aren't going to be crazy about that.

• 30:33Charlie Warzel?

  I do think that that trade off is a bit radicalizing to some people. Alex, have you, have you seen or felt in your reporting anywhere that this feeling of backlash?

• 30:45Speaker 04

  I haven't talked to many people yet who are really aware that these price increases are as bad as they are. I have talked to a lot of people who are just confused about why they're supposed to be using AI and the differences between what the companies say it can and will do versus what they actually, the benefits they actually see from using it.

• 31:04Speaker 02

  In the US at least, I don't, I think it will be, I don't think it's necessarily apocalyptic. I think a lot of people are crafty. Like I found out about the memory crisis in March because I bought a Mac on Facebook Marketplace. And the guy I bought it from was like, you're getting such a good deal. You know, like the, these prices are going to increase. Like I'm increasing my used Mac prices. So I think that it's annoying and it, you know, it's painful. I think people sort of autopilot like, oh, I'll buy the next new thing. That might be disrupted for some people, but you can, even if used prices are going up, like you can get used electronics. Companies may also be crafty. They might, you know, come up with smarter ways to allocate memory.

• 31:48Speaker 02

  Apple really coolly has this thing called unified memory. That's quite unique that like allows like more tasks to be done with like less memory. So I feel like they're, I don't want to make it seem as if like Mac price goes up a hundred dollars, like now my family can't eat. Like I think that people will find ways to get a lot of the tech that they, they need. You know, I think, I think a lot more people will probably buy, buy used or just delay refreshing the tech and try to stretch their old tech out for longer.

• 32:21Speaker 04

  Yeah, although I, I watched the price of used smartphones pretty carefully. And I will say that that has been going up significantly in the last six months, which is something I've just never, I've never seen. And it's also important to note that like, it is the most affordable devices that are being hit the hardest by this. I actually am not sure that certain people are going to be able to continue having a computer in the house, you know, when their current one doesn't work anymore, or like the smartphone that they really want. I think that is really is going to cut off access to technology for a lot of people.

• 32:53Charlie Warzel?

  So this is a good segue into my next question, which is, what do you both anticipate that the future looks like here? Like, what are some possible visions for how all this plays out? And I'm talking about both the, the inefficiency of the hyperscalers as they build out. And then, you know, how that may or may not affect whether these prices just keep going up, whether, you know, the crunch on the, on the memory industry continues to intensify.

• 33:23Speaker 04

  I think, I mean, the word bubble is used a lot. And I think if we are in a bubble and as those in fact pop, you know, who, who knows? I think the AI industry is in a very weird position. It's trying to sell a product that may not be profitable. It may not have the resources even continue selling that product. Its own costs may go up so much that they, these companies can't continue. The whole AI thing could in some sense fall apart. It's a little hard to imagine companies of that size really tanking, but we have seen that in the past. I, I don't know. It's very hard to predict what's going to happen.

• 34:07Speaker 04

  But I think certainly everything can't continue going the way that it is. Like the path that we're on is not really sustainable. Something's going to have to change.

• 34:16Speaker 02

  I think something that sort of caught my attention is that GoPro told the federal government that the company was at risk of bankruptcy because of the memory crisis. And like GoPro is a name that I know. I guess, you know, they don't have enough pull to get the memory that they needed a price that's sustainable. And so I, I think what we might see happen is Apple, Samsung, Dell, like, like bigger companies that really buy in bulk and have giant markets. Like they'll be able to get the memory they need to fill orders. They'll be paying more for it, so stuff will cost more, but they'll like have what they need. Whereas smaller companies are saying that,

• 35:00Speaker 02

  you know, we can't even get memory companies to pick up the phone when we're trying to fill orders. So I think you, you know, we could get to a point where companies, I guess, B tier and below, I think if we put like GoPro in B tier, they just can't make their products anymore. And then we end up having sort of like a leaner

• 35:22Speaker 02

  group of gadget makers. Like I think of like indie gaming consoles. I feel like that's maybe a thing that doesn't make sense. Or like, just, just anyone but the biggest dogs kind of losing out.

• 35:33Charlie Warzel?

  That feels radicalizing to me again in this way that like, if this actually starts to keep going in the way that it does, right? If, if trends continue. I think consumers, politicians who are, you know, looking to latch on to something, watching these hyperscalers, these already huge, massive companies, driving other companies out of business, I think is a, is a really bad look for these companies. I think it is something that could be just really galvanizing, really tangible for, for lots of people, especially if they watch companies go out of business simply because they can't get these memory companies to pick up the phone. But I, I want to end here, which is that

• 36:16Charlie Warzel?

  reading all of your reporting, I'm struck by this possibility that we could really end up in a bit of a vicious cycle here. AI companies need to hoover up memory to run. If they succeed in scaling up and getting people to adopt the technology, then the demand is going to keep going up. This means the prices will keep going up. And so it feels like if companies, if these companies win, we get trapped in this paradigm. Right? And we, the consumers, ultimately end up losing, footing the bill for a lot of this. If this is a bubble and it bursts and the companies start to falter, then obviously we see prices go down. But that happens in tandem with what would be

• 37:01Charlie Warzel?

  a serious financial crisis, given the, the historic investment in AI. Is there a future here that doesn't feel really bad or look really bad to consumers?

• 37:11Speaker 02

  I think it's possible that like the, the era of dirt cheap computing that we've been in since the 1960s is like, you know, like gone. And that was cool. And the transistor bought, you know, brought that about. And it's also like, I think the optimist's take is that maybe this causes engineers to like get good. And there's like resource constraints and we design like systems, architectures and software and hardware that is, is more efficient. And we get, you know, whatever the next transistor is, you know, and we enter this era of like, you know, more efficient tech design.

• 37:53Speaker 02

  But I think that having a TV used to be a luxury, having a fridge used to be a luxury. Like maybe, I guess if we don't engineer our way out of this, or if, you know, memory capacity never catches up with demand, we might just be paying more for, for electronics and then we'll have to adjust our expectations and the way we live our lives.

• 38:16Charlie Warzel?

  Hana, Alex, thank you for coming on Galaxy Brain.

• 38:19Speaker 04

  Yeah, thanks, Charlie.

• 38:21Speaker 02

  Awesome. Yeah.

• 38:26Charlie Warzel?

  Thanks again to my guests Alex Reisner and Hannah Kieros. Before we go, though, a quick note from me. Talking to Alex and Hannah, I am struck by the ways that this memory crisis dynamic actually rhymes with a different issue in the tech world. This term that is coined by Cory Doctorow of "inshitification." Inshitification is essentially the idea that tech platforms entice people with free or very useful and efficient services. And over time, those services monetize, they get worse, and they exert more power over the people who become locked into that ecosystem. What's happening with memory is, in some ways, very similar. We've all gotten used to a style of computing that gets better and cheaper over time. In the case of smartphones, we now use them to replace all kinds of physical items

• 39:09Charlie Warzel?

  in our lives, from wallets to cameras to maps. We rely on them. And the companies who build these devices, they've locked us into their ecosystems over a long period of time: Apple, Android, you name it. But now, almost 20 years into the iPhone era, it's possible that we're starting to see the turn of the screw. Now, dependent on these products, they're becoming wildly expensive in order to help create the AI technology that is supposed to infuse these devices and make them even more powerful. You could call this "inchipification." And it's not quite clear how all this is going to play out. It seems plausible to me that the people who can will continue to pay, even if prices become astronomical. Those who can't, though, may be left out, stuck with devices that aren't powerful enough

• 39:54Charlie Warzel?

  or easily repairable enough to work well. This would deepen the digital divide, which is a technological form of wealth inequality that's already very real across the globe. Now, it's possible, of course, that things could snap back, that demand will change slowly over time, without a market crash or without a bubble bursting. But what seems really clear here is that the biggest, best-positioned technology companies and manufacturers in the world, they have this distinct advantage. They can hunker down, they can ride out this crisis. Others who can't afford to will not be so lucky. And so regardless of what happens, it is clear that the AI boom is drastically reshaping the world in all kinds of unexpected ways. The hyperscalers and boosters argue that what they are building is an unalloied good for humanity.

• 40:37Charlie Warzel?

  But what is inarguable is that whatever it is they're actually building, it's coming at a genuine cost. Okay, that's it for us here. If you liked what you saw, new episodes of Galaxy Brain drop every Friday. You can subscribe on the Atlantic's YouTube channel, or on Apple, or Spotify, or wherever it is that you get your podcasts. And if you want to support this work and the work of my fellow colleagues, you can subscribe to the publication, theatlantic.com/listener. That's theatlantic.com/listener. Thanks so much, and I'll see you on the internet.

• 41:16Speaker 00

  If you're trying to get news today, you'll find a lot of biased algorithms, rage bait, and AI noise. Thanks, social media. That's where Inpress comes in. It's a social platform on top of factual journalism, not the other way around. Track your reading and earn points. It's like Strava for your brain. Plus, you can connect with other users in your city who share your genuine curiosity. Ditch the echo chamber for reality with Inpress. Download it today for iOS or Android at inpress.app. That's I, N as in news nerd, press.