Hard Fork
Original podcast
“I'm told that Australian teens, in preparation for this ban, have been exchanging phone numbers with each other.”
14,615 words · 7 speakers
Your agents can talk to each other. They still can't think together. That's the gap Outshift by Cisco is closing. Outshift is Cisco's incubation engine, building the internet of cognition. Shared intent, shared memory, and guardrails that allow agents to work together from anywhere. Open source protocols and code. A system for intelligence that scales. Read the paper and experience the demo at outshift.com. That's O-U-T-S-H-I-F-T dot com.
The other day, my friend Kyle is out in the suburbs here in the Bay Area, and he's like in a parking lot. And he sees a car, happened to be a Tesla, and he notices the license plate. And do you know what the license plate said?
What did it say?
The license plate was Hard Fork.
No.
So Kyle listens to the show and, for whatever reason, assumed that this must have been your car.
Wow.
And so he walked up to the, the sort of, you know, passenger's side window, which was down, and he says to the man behind the wheel, "Oh, hey, do you know Casey?" Because he assumes this is you. And the man looks at him and is like, "No." And Kyle, you know, of course, becomes embarrassed and explains, "Oh, well, you know, like, there's a podcast and, you know, it's my friend host it." And the man looks at him and says this. Ugh, this is the second time this has happened.
Wait, that's amazing.
Yeah, so whoever is out there driving the Tesla with the Hard Fork license plate, we're sorry for the fact that you're just gonna be accosted now pretty much every time you, you know, park your vehicle.
You know how people put those bumper stickers on their Teslas that say like, "I bought this before Elon went crazy"? That guy should buy a bumper sticker that says, "I bought this before there was a Hard Fork podcast." Would solve all his problems.
I'm Kevin Russo, tech columnist at The New York Times.
I'm Casey Noon from Platformer.
And this is Hard Fork.
This week, the drama down under as Australia bans children under 16 from social networks. Will the rest of the world follow? Then blogger Andy Mazzli joins us to separate fact from fiction on AI's environmental impact. And finally, it's our first ever Hard Fork Wrapped. Wrap it up, Kevin.
Well, Casey, I miss you. I know you're in San Francisco this week. I am back in my hometown in Ohio, visiting some family. And it's cold here.
Yeah, and let me just explain this. You know, sometimes podcasters get in a huge fight and they just don't really want to be in the same room for a little bit. And one of them will go home to their family and just kind of cool off. So Kevin, thank you for taking a step back and, you know, hopefully cooler heads will prevail here.
Yeah, you're welcome. And I hate you.
No, actually, Kevin went home for the holiday and it's very sweet, you know, going there to be with your family.
It is weird to be like on someone else's like Zoom setup, though.
Okay, say more.
It feels very personal.
Yeah.
It feels like you're sleeping in their bed or something. You know, it's like, things aren't where they were, you know? My windows are everywhere. I've got one monitor over here, one monitor over here.
Well, with that in mind, maybe we should zoom in to what's happening in Australia.
Great transition.
Thank you.
Love that segue.
But yes, this is the big tech news this week. On Wednesday of this week, the children of Australia lost access to their social media accounts, many of them, as a new nationwide law went into effect that bars anyone under 16 from having an account on the most popular social media sites.
That's right, Kevin. And I think this is really big news because for more than a decade now, countries have been reckoning over what to do about social media. It seems to have a lot of secondary effects. On their people and their countries that they don't like. And yet, what to do about it has remained an open question. And people have tried to pass, you know, little regulations here and there. This is the biggest step we have seen a democracy yet take to simply say, hey, if you aren't 16 years old, you just actually can't use this at all. And I think that this is the first domino to fall in a move that could reshape how we all use the internet.
Yeah, it's a big deal. And if I were to summarize the sort of reaction to this law going into effect in Australia in a way that Australians might appreciate, I would, I might say that some people are mad as a cut snake. They think the government's made a dog's breakfast of this policy, which they think has bucklies of working. But others reckon it's fair dinkum. And that while improving teenage mental health may be hard yaka, you can't just spit the dummy and do nothing.
Am I right?
Wow, I didn't really get one word of that, but I'm gonna trust that ChatGPT has not led you astray there and that those are all actually Australian phrases.
It was Gemini. Thank you.
Gemini. Did you at least try to slip in one fake Australian phrase just to irritate our Australian listeners?
We have Australian listeners. They're gonna be so mad about this segment.
No, they're gonna love it. We're shining a spotlight on what is happening in their country, Kevin.
That's true. That's true. So yeah, let's talk about this because I know you have been writing about this. I have been watching it play out a little bit and I've been reading a little bit about it. But you know much more about this than I do. So why don't you like, give me the sort of capsule summary history of this new Australian law that went into effect this week?
Yeah, so the law says that if you are under 16, you cannot use the following 10 apps. Facebook, Instagram, Threads, TikTok, Twitch, Snapchat, YouTube, X, Reddit, or Kick. Okay, so, you know, you add all of those up and that is where Australian teens, just like teens all over the world, are spending the bulk of their time online. If you want to try to have an account there, you're going to need To try to verify your age in some way. The most popular way tends to be this age estimation video selfie that you see people do. Of course, this can famously be tricked by video game characters sometimes, but it also works a lot of the time. And so that's the move.
And it all came about, Kevin, after the wife of the premier of Australia's second smallest state read The Anxious Generation, a book by past Hard Forecaster Jonathan Haidt, and said, "Well, this seems really bad. We ought to do something about it." And that got some legislation going. And less than a year later, so this was last November, Australia passed the law. And the platforms have now had a year to kind of consider this and figure out how they're going to implement it. And this week, it finally went into effect.
Interesting, the power of books in this day and age. You know, we can, we can write them and sometimes people read them and pass laws based on them.
Absolutely.
Questions about this. One is, does every platform have to do some kind of proactive age estimation? Or are they allowed to just say, like, here's a dropdown menu, put in your birthday, and we'll tell you whether you can use the site or not?
You cannot do the dropdown menu. So these sort of fake age verification measures that have existed since the dawn of the internet are no longer going to cut it. And so that does mean that, yes, if you are an adult and you're just signing up for your first Snapchat account, you are going to have to verify your age via one of these means.
Got it. And you listed off a bunch of apps. Two that I didn't hear mentioned, one was Roblox, which we have recently talked about on the show. The other was Discord. So are there other apps that are popular with teens in Australia that are not included under this provision?
Sure. So, you know, it only applies to the time that I mentioned. Some folks that I talked to at the platforms are irritated that Minecraft isn't on this list. Steam, which is a popular way for kids to play PC games. But it's really Roblox and Discord that are generating the most eye rolls at the platforms who were targeted. Because folks there believe that this is an example of, as I've been told, the government picking winners. And losers and effectively saying like, hey, we're going to make it really hard for you. Meta, TikTok, YouTube, but Roblox, for whatever reason, is going to get a pass. Julie Inman-Grant, who is the e-Safety Commissioner of Australia and who is sort of overseeing this effort. When asked about the Roblox omission, basically said, yeah, we were concerned about them and we went to them.
And then they rolled out a bunch of new features, including age estimation technology. And so we feel like they're OK for now, but it is possible that in the future, they could get added to the list as well.
Got it. And what about YouTube? How are they handling that? Because there are lots of ways to watch YouTube videos that don't involve like signing up for a YouTube account.
Yeah, so this is a really interesting one. So as you just mentioned, Kevin, you can watch YouTube without logging in. Although, of course, YouTube will constantly prod you to log in because they can sort of make more money off of you that way. But it is possible to watch in what they call the logged out state. And interestingly, you can actually still personalize a feed even in that logged out state, right? So just by seeing what kind of videos you're watching, taking other cues, you know, what browser are you using? Where are you located? They can show you things that they think that you might like. And so this is another criticism that has been leveled at the law, which is that, you know, if you're Meta and you make Instagram, Instagram doesn't have that logged out experience, right? And so folks at Meta may feel like this feels like a giveaway to YouTube
because they're going to have a way to sort of continue to get engagement from their younger viewers. Of course, what I think is likely to happen is Meta is simply going to introduce a logged out experience for Instagram and Threads and Facebook.
Yeah. So speaking of how these companies may try to get around this or push back on it, like, what are the companies doing? I imagine they've been lobbying against this law. But what have they been doing in anticipation of this thing going into effect?
Basically just frowning and stamping their feet. You know, YouTube has said that they feel like it's very rushed. You know, because of course, we've only had 20 years of seeing what the internet does to young people's brains.
Interesting. What about kids in Australia? Do we know how they're reacting to this?
Yes, so kids are doing a number of things. Some smarties, and I have a feeling that you would be one of these kids, Kevin. They've had a year to prepare for this, so they created backup accounts where they set their ages to, you know, like 19 or 26, and so they're planning on using those. They're also downloading apps that aren't covered by the ban, right? Another pretty predictable response. That includes a bunch of apps that I'm not super familiar with, including Coverstar, Yope, and Lemonade. Do you have a huge following on any of those platforms?
Yeah, I met my wife on Coverstar.
Okay, interesting, interesting. So.
That's a Casey Newton joke for those keeping score at home.
So they're doing that, but Kevin, I was most interested by this thing that these kids in Australia are doing. They've gone back to using this ancient technology that I didn't know that if we'd ever see this again in my lifetime. But I'm told that Australian teens, in preparation for this ban, have been exchanging phone numbers with each other.
Come on, that's unrealistic.
Really?
Yeah, and so what happens is, so each of these kids, they have a seven-digit, or you know, maybe it's 10-digit. I don't know how many digits it is in Australia, but you have sort of a unique code that belongs to your phone, and you can give that to another person. And then if they put that in their phone, you can then send them messages and maybe even do a video call. So this is apparently seeing a huge surge of interest in Australia right now.
That's incredible. I love that. I'm so curious what the sort of second and third order effects of this will be. Like, I think right now, it seems like the primary people who would be benefiting from this would be parents and also maybe the teenagers, but I can imagine kind of some fears about maybe long-term competitiveness. If everywhere else, the teenagers are using social media and the internet and using that to not only like amuse themselves, but to learn things, like, do you end up in a situation where your kids and teens are behind in some ways? Or do they race ahead because they're the only country where the teens aren't glued to their Instagram feeds all day?
Well, and you're getting at what I think is maybe one of the most interesting aspects of this whole thing, Kevin, which is that we are now setting up a really interesting natural experiment in what happens when you take social media away from children. For the past decade now, I have been covering this endless debate about to what degree can we hold social networks responsible for some of the mental health challenges that the younger generation has. And those challenges, by the way, are enormous. We've talked about it on the show many times, but we have not had an island where everyone under 16 is being prevented from using these technologies. And I think it's worth asking, and I'm actually kind of curious to get a prediction from you on how big of an effect this will have.
You know, I was talking with somebody at one of the platforms the other day that was basically saying, "Look, if you think that this ban is going to go into effect and all the kids are going to put their phones down and go play in the park, you've got another thing coming," right? These kids have already grown up in a world where they are used to being online all the time, and we're not going to like throw back the clock to the 1980s in one fell swoop. On the other hand, if these kids don't have a million push notifications bothering them at all hours of the day and night while they're sitting in class, making them anxious about their physical appearance, sending them down eating disorder rabbit holes, I can't imagine these kids actually coming out on the other side of that better off. So it's a really interesting natural experiment, and I should add that Australia is actually going to study this.
They're putting together a panel of academic researchers that is going to essentially track what is happening to these kids once they're banned from these social networks, and they're going to publish that research. So we should actually, within a couple of years, have some really interesting data here. But what do your instincts tell you? Like, what do you think is going to happen to the children of Australia? Which, by the way, I'd love for somebody to clip that and just take that quote out of context and put it online somewhere. What do you think is going to happen to the children of Australia?
I mean, I don't know. I think the sort of first couple of years, we may not see like huge short-term effects. Like, I think, I think that part of the, the dream is probably unrealistic. But I can very well imagine that like five or 10 years from now, when sort of a whole generation or a micro generation has had time to like, go through childhood with this rule in effect, I can imagine there being some actually like, pretty steep differences between them and the generation just before them, who had this stuff all along. I think I've been like, radicalized a little bit by the debates around school phone bans here in the US. This is a topic that we've talked about a lot. And we heard
these sorts of same arguments on both sides in the sort of months and years leading up to those phone bans going into effect. Oh, it's, you know, how are kids going to communicate with each other? They're going to feel so left out. Like, you know, the parents aren't going to be able to get in touch with them if there's an emergency. And then these bans went into effect. And like, from what I can tell, they have been universally successful. Like, I'm sure there are some districts where people are unhappy about it. But like, everything that I've seen, every teacher or parent that I've talked to has said like, this has been a great thing for our schools. And so I am just very skeptical of these sort of, you know, worry wards who say, oh, if we take social media away from children, it's going to like, have all these bad knock-on effects. Because I think what we've seen so far is
like, sometimes this stuff just works. Sometimes it just produces a positive net result. But I don't know. How are you feeling about this Australian law?
Well, so, you know, I have talked on the show before about some of my concerns around implementing these sorts of bans, right? If you're a 13, or 14, or 15-year-old creator who has been building a business, and you're now going to lose that business, that sucks for you. If you're 14, and you're LGBT, and you have an unsupportive family, and you're about to be cut off from some online resources that made you feel less alone in the world, that sucks. And I do think that there is some number of kids who can just sort of safely use Instagram and TikTok, and their lives are basically fine, and they use it for entertainment. They are still able to like, you know, become self actualized humans and everything else.
But to use the public health analogy, Kevin, it's also true that a lot of people can smoke and drink without it ruining their lives. And we still made the choice that we wanted to restrict access to alcohol and tobacco until those kids were a little more grown up. And over time, that resulted in fewer people drinking and smoking. And I think most countries are better off for that case. So while I'm concerned about what some of these kids are going to be losing at the margins, I just keep coming back to the argument that I never hear anyone make. You know what that argument is?
What's that?
Looking at Instagram is really good for your kid.
Do you know, you know what I mean? Nobody is ever like, hey, you've got to get your 14 year old looking at TikTok. If he's not looking at TikTok, he's really going to be missing out. You never hear people say that. At most, this is like a tolerated distraction. And at worst, it does cause immense harm. And so I think particularly for the under 13s, it's just a no brainer to get kids off of this stuff. They weren't supposed to be there on the first place, but we know that they were there and the platforms took very few steps to get them off. I think it's slightly trickier for the 13 through 15 year olds, but ultimately not tricky enough that I am unwilling to let Australia run this experiment and see what we learn.
Totally. So Casey, let's wrap this up by making a prediction. Do you think this law will succeed? And what would success look like to you for this?
Well, I think what Australia is hoping to see is decreased rates of mental health issues with teenagers in Australia. Maybe higher test scores, right? Some sense that the, the mental health of teenagers in Australia is improving. And I have to say, if I had to guess today, my guess is that this would have some positive marginal effect on the teens of Australia. Again, I don't know that it's going to get them off of their phones. I'm sure that some of them are going to find their way into very dark corners online that are unregulated. And they're going to have, you know, issues similar to the ones that they're now having on Instagram and the other apps.
But on balance, I do think that we should expect to see some, at least some positive benefits from this. What do you think?
Yeah, I think so. I think the relevant time horizon is probably longer than people want. Like here, I don't think you can do a study 12 months in and say have the mental health outcomes improved and like have that be sort of an indicator of whether the law is working or not. So I think this is going to be more like a sort of, you know, five, 10 year project, but I think it'll be really interesting. And I actually don't think that other states, other nations are going to wait five or 10 years before they decide if this is something that they want to do. In fact, I think probably a year or two in, if there is like no indication that Australia's sort of teenage cohort is falling apart as a result of this law, I think a lot of other democracies are going to do something very similar.
Absolutely. You're already starting to see it. Denmark is pursuing a ban for under 15s. Norway also now pursuing a ban for under 15s. Malaysia has said it is essentially intending to copy the Australian law. So this is really striking for me, Kevin, because you know, you and I both covered social media for a long time. And I think particularly in the United States, after 2016, we get this big backlash against the social networks. And yet it felt like in America in particular, as if there were no consequences for these companies, right? Their, their brand value plummeted, right? Because these companies are widely disliked, but they continue to grow. They're making more money than ever. This is the moment where I was like, oh, every, all the bad choices that they made
over the past decade have actually caught up with them. They have lost this argument. A growing number of countries now simply are saying, we don't believe you anymore. We ran your experiment to see what would happen when we let anyone use your platforms. We didn't like the result. And so now no one under 16 can use your platform. That is what it looks like to lose an argument.
Yeah, I agree. So while some people in Australia may be carrying on like a pork chop, I think this one's a ripper.
You know, Kevin, sometimes I worry that your Australia jokes are going to boomerang back against you. You'll be in real trouble.
When we come back, is the AI water use narrative fake? We'll talk with press critic Andy Maseley about what we know and don't know about AI's water footprint.
Today's AI builders need infrastructure that scales with their ambitions. Irene is a vertically-integrated AI cloud platform, delivering data centers, compute, and software for AI training and inference, underpinned by its expansive portfolio of land and grid-connected power in renewable-rich regions globally. Irene's AI Cloud gives direct access to Nvidia GPUs, scalable from hundreds to thousands. For builders, that means direct access to AI compute, rapid scaling, and predictable pricing. Irene, AI Cloud for the next big thing.
Companies are pouring money into AI, but few leaders can explain how those AI investments work with the systems you already have. You know, the ones that actually run your business? Without Systems, you don't have to start over with AI. You can choose the projects where AI will have the most impact, and OutSystems gives you the platform to make it agentic on your terms, with complete control every step of the way. So you can trust AI with the mission-critical systems that run your business. Own your agentic future. Learn more at outsystems.com/NYT. That's outsystems.com/NYT.
Your sales team is at risk of missing quota. Don't panic, just ask Rippling AI. Built on your real-time people and business data, Rippling AI can pull metrics from Rippling and your CRM into a meeting-ready dashboard showing quota attainment, headcount trajectory, and revenue to quota by region. In seconds, you'll see what's behind your quota risk and fix it before it's missed. Head to rippling.ai/hardfork to get the only AI built to give you full visibility and take complex actions across your whole business. That's R-I-P-P-L-I-N-G dot AI slash hardfork. Sign up today.
Well, Kevin, you can lead a horse to water, but will there be enough in the trough to run one ChatGPT query? That is what some people have asked over the past year as questions about the water usage of AI systems has come to the forefront of many online discussions.
Yeah, this is something that I've wanted to talk about on the show for a long time now. In particular, because I've been observing that like after anything happens with AI, like any big scandal or drama or some piece of viral slop goes viral, like inevitably the responses to that will talk about water, right? It'll say like, why are we drying out the lakes and the oceans to create this technology? And so I just think that it's been a very fraught emotionally charged issue for a lot of people. There are a lot of people out there who see AI and all the infrastructure that we're building to support it and just think like, this cannot possibly be good for the environment. And so today we're going to have a conversation about that.
Yeah, and I'm glad we are because this is just something that I see all the time in memes on social media. Like the other day I saw one that was like, hey, can you get out of the shower? I need the water for ChatGPT. And it got many, many likes and reposts. And that is sort of, I think, the default understanding on social media of the relationship between water and AI, which is basically you can only have one or the other. And this has found its way into a lot of mainstream coverage. And there is basically one blogger out there who got a bee in his bonnet about it and launched a one-person crusade to try to change that conversation.
Yes, so today our guest is Andy Masley. Andy is not an environmental scientist. He is not an AI scientist. He's by his own admission, just a guy on the internet. He is the director of the Effective Altruism DC organization. And before that, he was a high school physics teacher. But where I really took note of his work on his Substack, which is called The Weird Turn Pro, is that he has just been incessantly responding to what he says is this sort of made up narrative about AI and water. Use he has been unusually persistent in kind of debunking and explaining these specific claims and statistics and his blog is just full of like these long explainer posts about
how he believes the AI water issue is essentially fake that we are sort of working ourselves into a lather about something that is not a problem on the individual level. He has written about putting the AI water use numbers in the context of things we do every day, like watching YouTube videos or eating meat or relying on a car to get us around. And it has been really helpful, his work for me in trying to figure out where to place my own concern within the many things that you could be concerned about with AI.
Yeah. And as you'll hear in the conversation, Andy is not here to give a permanent free pass on environmental issues to AI companies. In fact, he says that as AI becomes a larger and larger part of the economy, there are areas where we do want to pay more attention. But in his view, if we continue to focus on the water issue, we may miss the bigger picture and the real issues to focus on. So I thought that was a perspective that is worth airing out because it does kind of fly in the face of the way that AI is often talked about online.
So let's get into it with our guest, Andy Maseley.
Andy Maseley, welcome to Hard Fork.
Longtime listener, first time caller. Great to be here. Yeah, thank you.
So you have been spending the better part of 2025 fighting with people on the internet about this issue of AI and its environmental impact and specifically within that, the water use issue. And I wonder if we could start by just sort of getting into why you decided to pick this particular fight. What made you interested in writing and talking and thinking about this topic?
The main way I started writing about this, honestly, was I was just bumping into a lot of people like at parties and stuff. I think we've all had. Maybe similar experiences where we talk about like using chatbots or chat GPT. And someone will be like, oh, like, how can you do that if you claim to care about the environment? Or whatever people would bring up the like now kind of defunct statistic that like, oh, a chatbot uses like 10 times as much energy as a Google search. And my first reaction to that at the time was like, that's not very much energy, you know? Like if I said I had done like a thousand Google searches in a day, that on its own like wouldn't really stand out. You know, like you wouldn't think, oh, Andy doesn't care about the environment so much. So basically I had started to dig into how this was being covered. I'm finding I was disagreeing with a lot of it, going a little bit crazy.
I started to go down more and more rabbit holes. And like now I've ended up here as like one of the main commentators on the issue, despite being basically just some guy who's doing this as a hobby.
I'd love to like situate you in the landscape a little bit for our listeners who may not be readers of your Substack. Like, you are not employed by any of the AI companies. You consider yourself an environmentalist. Am I correct in that?
Yeah, receiving no like financial incentive to do this outside of like my loyal Substack subscribers. Basically, I'm motivated to do this mostly by just bumping into a lot of what I think are pretty simple misconceptions. And honestly, as like a long time environmentalist and physics teacher of seven years, I'm pretty worried that a lot of people just have these really wildly off misconceptions that are distracting them from what I think of as more serious issues. Like one story that happened recently, a friend was in a pizza place and he'd overheard this conversation where someone had said like, oh, this person uses chat GPT. And so she can't consider herself to be an environmentalist. And then immediately after went and ordered like a meat lover's pizza. And like, you know, even just there, I'm like, oh, this person just really doesn't understand the different trade offs
involved in different things, like one is just so much worse than the other. And so a part of this is also motivated by just worrying that a lot of the environmental conversation about this stuff is a little bit off base and people are losing track of the big picture and like where the actual bad guys are.
Got it. Well, maybe we should just dive into one of the claims that you made, Andy, that you say you're very, very confident about, which is that an individual prompt of a chatbot like chat GPT. Just does not have much impact on somebody's overall water usage. Make that case for us.
Yeah, a lot to say. So most importantly, there's this common wisdom floating around that I think is pretty conclusively wrong, which is the idea that an individual chatbot prompt uses an entire bottle of water. This is based on a misreading of a Washington Post article, which itself I thought was pretty strange, where the article implied that writing a single email with ChatGPT uses a whole bottle of water. And for that to be true, the article seems to assume, among other things, that you're sending like 10 to 20 ChatGPT prompts to write this 100-word email. Like, I'm a power user of chatbots myself, but I don't use them that much. It's also assuming a lot, like, a lot of the water that it considers is actually in, like, off-site lakes dammed by hydroelectric plants, and it's considering the water evaporated on those lakes by the sun.
And then even there, after that, it's also assuming that these models are not any more efficient than they were in, like, the year 2020 specifically. So I would really like that talking point to go away. Basically, like, that's just a horrible misreading of a kind of, like, inconclusive article. Our best current data on this, I think, comes from Google, which seems to imply that the onsite, like, in a data center, water cost of a chatbot prompt is about 0.26 milliliters. And if you add the off-site cost, like, the water cost of generating the electricity, that might rise to, like, 2 milliliters or something. And obviously, like, there's a range of uncertainty and different prompts use different amounts. But I want to be able to at least start by saying, like, well, let's look at the median prompt. And so, like, putting that number in context,
if you look at, like, 2 milliliters and ask, how much does this contribute to my daily water footprint? My best understanding is that the average American's total daily water footprint that they consume is something like 1,600 liters altogether. So what this means is that this is about 800,000 times as much water as a prompt uses. And so to raise your total daily water footprint by even 1%, you'd have to send something like 8,000 median prompts. And if you're spending your entire day doing that, just like constantly poking and sending these prompts, like it's very likely that you're not doing a lot of other things that also use a lot of water. And so it's actually very likely that using AI on net might actually significantly reduce your total water footprint, like over the course of the day. Like the numbers are just so small that it's very comical. It's something like, if you don't
buy like one additional pair of jeans, that's like two million prompts, which is probably more than I'll do in my lifetime, you know? So like, I think there especially, people just have a lot of wild misconceptions about how this is impacting their water footprint.
Yeah, I mean, got it. And Kevin actually hasn't bought a new pair of pants in a year. So that actually is probably going to give him a lot more ChatGPT credits than the average person.
So that's the, that's the kind of micro argument here, but more often these days, and I still do see these like statistics about, you know, each ChatGPT prompt uses a bottle of water. I still see those floating around, but more what I see is sort of the industrial, the sort of macro view of this, which is, look, we're building all these AI data centers. They're using a lot of water. Now, using is sort of a broad term. It can include, as I have learned from your writing and other writing, some of this water just gets recycled back into the water source after it's used to cool the data centers, but some of it is evaporated or does sort of get taken out of the water, sort of ecosystem. So there are lots of different kinds of water usage going on. But what do you think
about the argument that the sort of scale of these data centers, the scale of the water needs of these data centers is something that people should be paying attention to relative to other ways that we use water?
Yeah, I mean, first, I just want to flag that I'm very happy that the conversation has shifted in this direction. I think that whenever we're talking about climate, it's almost always a distraction to focus too much on your personal footprint. And we should be talking about like the big systematic things in society and what impacts they can have. So yeah, there's a lot to say about the total industrial use of water by AI and data centers more broadly. It's definitely a lot by the standards of an individual, you know, like you can read about like millions of gallons used every day and think like, oh, you know, I don't use that much. That's crazy, you know? And I think the first thing I want to get across to people is that literally all industrial use of water in society sounds crazy from the perspective of an individual. If you report something as using millions of gallons per day, like this could either apply to like AI,
it could apply to like berries that we grow, it could apply to the US steel industry and things like that. There are just so many different ways that these huge numbers can pop up when we're talking about this stuff. So again, like instead of talking about big general numbers or numbers of gallons of water that AI uses, I try to contextualize it just using either the total amount that like America uses altogether, or just comparing it to other things. Another complicating factor here is that the last year we have really good data on this is 2023. And obviously AI has grown a lot since then, so I can kind of start with like a 2023 background and say, my best guess for like data center water use, like how much water data centers were using in America specifically in 2023, was something like eight times the water that my hometown used.
I'm from a town of about 15,000 people. And so like AI altogether was effectively like building eight new towns like that across the country. And like, obviously, that's not nothing. That's like a lot from the perspective of an individual. But it's kind of something that I would expect the water system to be able to absorb and to manage pretty well.
Andy, tell us where that 2023 number comes from and maybe why we don't have more updated numbers.
Yeah, I mean, the reason we don't have more updated numbers is that AI is just very fast moving and it's hard to understand like exactly how much water AI companies are using because they're not always totally transparent. This number comes from the Lawrence Berkeley National Laboratory. They have a really great general report on data center energy and water usage. I think that's probably the best single document if you want to start doing deep dives on this yourself. They're just like broadly respected. This is the most comprehensive general report that we have, so a lot of my writing has been based on that. But yeah, because 2023 is the last year we have good data and there's a lot that's changed since then. I'm starting with that, but flagging that obviously like a lot's grown since then, but we also need to make some reasonable guesses about how much the growth has happened.
Got it. And that that 2023 data that you have, I wonder if you could put it in the context of like another industry. Like, how does the AI industry compare to like other industries that we have in America when it comes to their water use?
Yeah, it's a great question. Um, by 2030, it could be as much as about 250 square miles of irrigated corn, which is about 1% of America's corn. So again, I'm trying to frame this as like You know, we should be about as worried about this as America's irrigated corn growing by about 1% or something. And again, like this really isn't nothing like that's a huge amount of water. But the reason I'm comparing this to other industries like agriculture, other things to like even just household leaks use a lot more. The reason why I'm making these comparisons is more to say like, should we expect the current water system to be able to absorb this? And where should we expect to actually see problems? And I think just based on how much water we use in most other contexts, we shouldn't actually expect this
to be a huge disruptive disaster for water use. Like, there could be problems in the way that there are problems with other large industries. But I kind of expect this to be like one of many just normal industrial uses of water going forward. So I've come away thinking this really isn't as much of an issue as people think.
Yeah, but obviously, there are issues with the AI infrastructure build out.
We've talked about some of them on the show. One of them is that there's just like not enough power to power these data centers. And so because it's very hard and time consuming to build new renewable sources for these data centers, like solar and wind farms and things like that, and nuclear companies have been doing things like fracking or relying on, you know, older sort of less energy efficient, less environmentally friendly forms of power generation to power these data centers. So are you worried about those things? Are you saying it's all overblown? And we shouldn't be worried about any of the environmental impacts of AI?
Definitely not saying that is overblown, especially I basically think the water case seems kind of open and shut to me right now. Whereas the power case is a huge question mark. One way to think about this that I think is useful is that as a proportion of our total system, the increase in electricity going to AI will be in order of magnitude more. Like, from what I can tell, it's going to rise from like 0.5% to like 5% by 2030 of our national electric grid. Whereas for water, it might only rise to about like 0.5%. So like, there's an order of magnitude difference there. And there's just a lot to think about. And there have already been, I think, like, pretty bad effects on the environment from data centers. Like, there's been a lot of instances of like coal plants staying open. And so this is the main thing that I am really
Really quite worried about. And I think that like the answer to this is going to involve a lot of different things like it might involve in some cases data centers not getting built. Like I'm definitely not coming in saying like, oh, data center should always be built everywhere or whatever. It's just that I do want the full number of trade offs to be considered. And there are also potentially a lot of positives about the build out for the energy system. Like a lot of AI companies are, you know, committing to buy more like renewable energy as a proportion of their total energy than a lot of other industries and stuff like that. So that could at least lead to like economies of scale in renewable energy. And that might not cancel out the negative effects either. I kind of just want people to be aware of the full picture here. But yeah, I have to admit, I'm really agnostic about the future of AI and energy stuff.
And like, this is a place where there are just so many unknown unknowns that it feels really hard to say anything concretely. So my ask isn't that like, we assume it's fine. I just ask that like, we weigh the pros and cons and consider the trade offs rather than just saying like, oh, any new energy usage is bad, or like if AI causes any new emissions, it must be bad. Because ultimately, any new industry is going to come with some additional emissions.
That is a very intellectually honest and principled crusade. It has also been my observation that your arguments are beloved by the AI accelerationists. David Sacks, the White House AI czar, just reposted you on X.
And I think frankly, a lot of people, when they read your stuff, they think, gosh, this is great. I can finally relax about the environmental impact of AI and I can maybe ignore some of the other stories that I was seeing. Now, I know that your intention is just to give people the facts about water in particular. And now you're also sharing some information with us about data centers. But I wonder how you feel about the ways in which your work is being used, maybe by people who actually don't care about the environment at all.
I mean, not, not great, to be clear. Like, I think the general weird thing is I'm diving into this huge contentious debate about like one of the largest new industrial build outs ever, period. Definitely in my lifetime personally. And so like some weird allies are going to pop up. And yeah, I'm not especially excited when people use this as a cudgel against environmentalists. Like, I definitely identify very strongly as an environmentalist. I do really worry about anybody who's concerned about climate and the environment, worrying at all about these really tiny levels of emissions in their day to day lives. Like, it makes me really sad, honestly. I feel like back when I was in college, there was just a much stronger consensus that like, personal emissions are like, okay to worry about, but the main game, like, the main thing to worry about is the green energy build out.
And like, I feel like if I could take myself forward in time to the present and see like, oh, now everyone is kind of litigating like, equivalently, like, whether they watch YouTube for a few extra seconds a day, or something like, that would just kind of depress me. So, I do think that like, from an environmentalist standpoint, I do kind of want to snap people out of this incredibly tiny concern in parts of their lives. And if I make some allies with people who hate environmentalists, that's unfortunate, but like, I do want to get the message out there. I think more broadly, though, I'm becoming more and more convinced that like, bad arguments and bad arguments against AI are mostly just drowning out the good ones. Like, I think there are just a ton of reasons to be worried, even for like, non-weird sci-fi scenarios that like, AIs that currently exists can be harming people a lot. And I think that to get a full sense of what's going on,
like, the average person should probably at least poke around a chatbot a little bit and just have some general idea of like, what can these models do? Like, you don't have to use them in your life, ever. Like, I'm not saying like, you are obligated to use AI. But I would at least like people to understand that these are very different from like, a roulette wheel that just lands on random words and can't possibly produce any value, because a lot of people still seem to be motivated by this. And like, they seem to be motivated in part to not use them by these environmental arguments. I do actually, I have worried about this a lot, where I'm like, I'm worried that I'm coming off on like, one strong side of this and just helping the pro-AI side, which I don't really want to do. But I have to admit, for the most part, just like, the general understanding of this stuff is so bad. And I would like my side to at least be like, armed with a correct understanding of the world.
And this just feels like, honestly, shooting fish in a barrel, like, so many misunderstandings are still floating around everywhere, that this all leads me to the direction of thinking like, okay, I'd like people to actually understand that this one thing is incorrect. Let's focus on the more serious stuff instead, including like, things that I think a lot of AI critics would otherwise like, agree with me on.
Yeah, I mean, I think if I try to like steel man the argument that you may face or put myself in the shoes of someone who's very worried about AI water use, I would just say like, look, whatever the actual statistics are, however much water AI does or doesn't use, the issue is like, we didn't sign up for this as a society. We didn't ask these giant Silicon Valley tech companies to start putting up data centers all over the country, or stringing the electrical grid, or just diverting a huge percentage of the economy into building out these powerful AI systems. This all happened very suddenly and without any real like democratic participation. And it just feels bad to see this transformation of our society into something that's seems
designed to support this technology that I don't like, that doesn't seem that useful to me, and that, you know, I was never consulted about. So like, what's your response to that version of the argument?
Yeah, a lot of things to say. I guess I want to respond to it with a lot of respect, first of all, because I'm trying to write for an audience where it doesn't matter whether you love or hate AI. Like, you know, I'm very worried about AI overall, but also a very big fan of AI tools as they currently exist. Like, I use chatbots pretty regularly. And this does bias my writing to an extent where I do think, you know, like, I have a lot of positive impressions of AI as it currently exists. And I want to try to reach people who don't, basically. If you think AI is bad, then obviously you should also like generally oppose it using water for the same reason that you should oppose anything else bad that's using water. But that opposition should also come with a sense of proportion of like where the actual water issues are in the country. And like, there are a lot of other things I think are bad that use a lot of water, specifically animal agriculture.
And like there, I do actually feel a little bit more justified in focusing on that a lot, just because it's using such a huge amount of the nation's water. Something like a quarter of your total dietary footprint for water comes from like the food that is grown to eventually feed animals and stuff like that. And so in thinking about this, I do want people to Not not be upset if they don't like AI, but I do want them to at least prioritize like, okay, if I care about water, will it actually make much of a dent at all if we shut down all AI data centers like tomorrow? Like, how much will this actually help America's water problems? And my basic claim is that it's not really going to make much of a dent compared to a lot of other issues. The very final thing is that I do worry that in these debates, a lot of the potential positive aspects of data centers for local communities
are getting drowned out. And obviously, there are a lot of trade offs, there are a lot of negatives, too. But data centers also just often add a ton of tax revenue for sometimes very poor communities. They can provide a lot of utility revenue for utilities to upgrade aging infrastructure. And in a lot of places in America, the main water access issue isn't necessarily the raw amount of fresh water. It's like aging infrastructure that might have dangerous amounts of lead, or just make it more expensive to deliver the water in other ways. And so like, a new very large buyer can actually help water access here. And again, I don't want to say that makes data centers good all the time, but I do want to say that we need to at least consider these trade offs, and not just potentially steamroll something that might in very specific instances help communities rather than hurt them.
Hank Green, a past Hartford guest, made a YouTube video the other day about this AI water use issue. And one of the points I thought he made that was interesting to me was that there's just bad data and misdirection happening on both sides of this argument. Basically, you know, there are some AI critics out there who are using bad data to make unsupported claims about water use. But the AI companies themselves are also guilty of sort of making their environmental footprints look smaller than they are by doing sort of fancy footwork with the data, doing things like telling us how much a single prompt uses when most of the environmental cost of these AI models happens during training, not inference, and ignoring the fact that a lot of these AI models now have these sort
of long reasoning chains that output like a bunch more tokens than a standard chatbot query, and they're not giving us like those numbers for the most part. So what were your thoughts on his argument? And are there any places you disagreed with it?
I mean, I love the video personally. I think I've been going a little bit crazy by reading a lot of what I thought of as very bad coverage of the issue that left people less informed. And like Hank's video It was just phenomenal. I thought like throughout it, I was like nodding along thinking like, man, this is actually just going to give people such a better understanding of water in America, like where AI fits into this. It's weird for me to like defend AI companies here. Um, I will say that for a while, there was a lot of demand from a lot of different places for AI companies to like release data on the per prompt cost of chatbots. And like, we're not being told the full picture here. And now that they do that, people will be like, oh, well, this distracts from the industrial use and the total use. And I'm like, well, you know, like, I would rather have this information about the individual prompts and we can look separately at the total amounts.
But like, we kind of have to hold them to account for one thing at a time. And here I think they've done an okay job, basically. And then, yeah, there are a lot of like invisible costs, like the cost of training, which is a little bit nebulous. And then the offsite cost and things like that. I guess when like Hank says things like that, or when other people say things like, we're not being given the full picture here. I usually want to clarify that if you actually dig into that full picture and try to draw error bars around like, what is the absolute most water this thing could be using? It still just never comes up to a point where I would personally worry about this adding to my personal um, water footprint at all. And so like, I, I did find one place I disagreed with Hank was more in his tone where he was just like hedging a ton and being like, oh, there are all these different complexities.
And I usually want to push people when they're doing that to say like, yeah, the complexities are real, but I still want to acknowledge that we have this like reasonable range of uncertainty. And within that range, like I'm just not finding a place where this is a serious concern on the individual level.
Now, Andy, I know you said that in general, we shouldn't focus on our, you know, individual impact on the environment. But I wonder how you think we should think about the fact that Kevin flies a private plane from Oakland to San Francisco every week for our taping.
That's unchill.
Yeah, yeah, yeah, a little bit unchill. Yeah. I mean, at the extremes, I, I think this is actually like a good lead in to like, if you are going to focus on your individual impact. There are just some things that just use hundreds of thousands of times as much energy and water as other stuff where like, they're just such crazy, easy wins everywhere. So definitely like consider that, but just like hold in your heart that like the main challenge here, the main fight that matters is like uh, helping with the green energy transition.
So Andy, as you think about the year to come, how are you expecting to blog? Are you going to continue shooting fish in a barrel, as you say, and sort of find more instances in the mainstream media, people sort of, you know, making these arguments about water? Or do you feel like you've sort of concluded that debate and you're moving on?
I don't know that there's too much else to say. Like, it would be exciting to write about that elsewhere. But I think for my blog itself, like, um, my main pieces on this are just pretty comprehensive. So I would definitely recommend if people want to see my full argument, you can read them there. Yeah, I'm honestly just very, like, personally tired of the topic. Like, every time I type data centers and water now, I just feel myself getting tired, honestly. So there's a lot of other stuff to say. Would definitely like to cover the electricity stuff going forward. Like, I think that's just the most interesting, complicated question here with a lot of unknown unknowns in both directions. And it's really cool. It like incorporates a lot of like climate ethics and questions about like, you know, grid stuff, which I've been really excited about since I was a teenager. Honestly, just like learning about the grid is really cool. And then eventually would also like to write more about like other AI related stuff.
Like, I do worry that I've kind of pigeonholed myself a little bit in like writing so much about how like, oh, the A, like, environment stuff is really overblown that I think people are sometimes missing that like, oh, I'm quite worried about this for all these other reasons. So we definitely like to write more about how like, oh, hey, like, what will be the effects on labor? Like, how will this enable like government surveillance and stuff like that? Like, what is the X risk case like and things like that? So definitely like to pivot to that more over the next year.
Andy, thank you so much. Really appreciate it.
Yeah, it was great. Here's a blast. Thank you, guys.
Well, Casey, that segment is H2 over. When we come back, we'll send off the year that was with a new segment called Hard Fork Wrapped.
Your agents can talk to each other. They still can't think together. That's the gap Outshift by Cisco is closing. Outshift is Cisco's incubation engine, building the Internet of Cognition, shared intent, shared memory, and guardrails that allow agents to work together from anywhere. Open source protocols and code, a system for intelligence that scales. Read the paper and experience the demo at outshift.com. That's O-U-T-S-H-I-F-T dot com.
Today's AI builders need infrastructure that scales with their ambitions. Irene is a vertically-integrated AI cloud platform, delivering data centers, compute, and software for AI training and inference, underpinned by its expansive portfolio of land and grid-connected power in renewable-rich regions globally. Irene's AI cloud gives direct access to Nvidia GPUs, scalable from hundreds to thousands. For builders, that means direct access to AI compute, rapid scaling, and predictable pricing. Irene, AI cloud for the next big thing.
Companies are pouring money into AI, but few leaders can explain how those AI investments work with the systems you already have. You know, the ones that actually run your business? Without systems, you don't have to start over with AI. You can choose the projects where AI will have the most impact. And OutSystems gives you the platform to make it agentic on your terms, with complete control every step of the way. So you can trust AI with the mission-critical systems that run your business. Own your agentic future. Learn more at outsystems.com/NYT. That's outsystems.com/NYT.
Well, Kevin, the end of the year is upon us, and that means it is time for Hard Fork Wrapped. Yes, and what is Hard Fork Wrapped? Well, as you may know, since 2015, Spotify has been giving users a recap of their music and listening habits for the year. And so this year, we thought, why don't we... Just steal that idea and we will tell listeners a little bit about what they've been listening to all year. And beyond just giving you a few stats, we can also give you a few updates on some of our most talked about topics and segments of the year.
Yes, we can do that. But first, can I just say what my favorite piece of the Spotify rap discourse has been this year?
Please say it.
So this year, people started noticing that like some of their Spotify rap was like made by AI and people started like posting about it online, very unhappy that like Spotify was using AI to do their rap. And like, the backlash was like, you thought they were just doing this by hand?
You thought there was like a guy in Spotify's headquarters, just like scribbling down your own personal thing. Anyway, I derive pleasure from that.
Yeah, I thought it was that that AI DJ, I thought he was the one who was taking all the notes on my listening.
Yeah, he's the author.
With that, Kevin, I think we should start by sharing a few stats with listeners about the year that we've had here on the Hard Fork Show.
What a year it's been. So Casey, so far in 2025, we have done a drumroll, please, collective 3,524 minutes of Hard Fork. That's about 59 hours or about two and a half days.
Two and a half days we spent talking to each other this year? Actually, that's an undercount because we only run about a third of what we record, so.
Yes. And according to our analytics on Spotify, which we get as creators, we saw that we had collectively 1.4 million hours of people listening to Hard Fork on Spotify, which I'll say it, it's too much. That's 160 years of continuous listening that you did this year.
No, you're thinking about this the wrong way, Kevin. That's 1.4 million hours of going on a run, of doing those dishes, of folding that laundry. Every single minute you spent listening to Hard Fork and doing something else was a minute that you also spent doing something else, and I salute you.
Yes, I would love the breakdown in the Spotify rap of like, actual time that people were listening and, you know, there's some way to like, use AI to determine whether you were just tuned out. You just left it on in the kitchen, forgot about it. I want to know that, too.
Just a few more fun facts. We've done 52 episodes this year. I think we're planning on 56 by the time that the end of the year is over. You will probably guess our listeners' top artist, a sort of pretty famous pop star by the name of Taylor Swift.
Oh, yeah.
Yeah. And you may also have guessed our listeners' top audiobook, Abundance by Ezra Klein and Jared Thompson.
One final set of stats about the year that was, Kevin. Our producer asked Notebook LM, which we have been training on a transcript of every episode of the show, what it thinks our audience's listening age was. This was a new Spotify wrapped feature this year. It told you how old it thinks you are based on the music you listen to. It told me that my listening age was 86. That is not a joke. Here's the good news for-
Wait, really? That's incredible. Wait, I have to say, what in your listening history made it think that you were 86 years old?
I listened to The Beatles, and I'm not even joking. But that's the problem, was that I like The Beatles, and they were like, "Are you 100?" I was like, "No, I like pop music." Unbelievable.
Wait, that's amazing. Mine was only 18.
Wow. Well, good news. You can still listen to Spotify in Australia then.
I think it's because my kids' like, blippy music got mixed up with mine. So it sort of took the average.
I see. Well, here's the good news for Hard Fork listeners. Your listening age is estimated to be 35 to 45. And we actually got some demographic data back recently that says that actually that is exactly right. Statistically, you are between 35 and 45 if you listen to the show. But, you know, we also have a lot of younger listeners as well, and even some older ones. So, you know, Hard Fork, a show for the whole family.
Whole family.
But what we really wanted to do, Kevin, with this wrap segment is to put a bow, if you will, on some of the most discussed topics and stories of the year so far. And so we came up with a couple of cute categories, not unlike Spotify. And first up, in a year that has been dominated by our discussion of artificial intelligence, we have an update that we're calling the AI Regulation Regulation of the Year.
Okay, I'm listening.
Now, as you know, states, including California and Colorado, have started to pass laws regulating AI. California, for example, passed a law requiring that AI labs publish transparency reports about new risks that their models create and how they plan to handle catastrophes. And the accelerationists in the Trump administration hate this because they say it's going to make life too difficult for the AI labs, it's going to make us less competitive against China. And so this week, President Trump said in a post on Truth Social that he will sign an executive order to try to curb these state laws from being passed. At the time of this recording, we don't have the final language, but in a draft executive
order that circulated last month, Trump directs the US Attorney to sue states to overturn AI laws, particularly if they think they can prove that they infringe on interstate commerce. And they also plan to tell federal regulators to withhold broadband funds and other funding to states that pass these AI laws. So Kevin, what do you make of our AI Regulation Regulation of the Year?
So this is the most enraging part of the AI regulation debate, is that there is this whole group of people, mostly in the White House and around the White House, who have just sort of bet the house on what they call federal preemption, right? This is the notion that instead of having a bunch of different state laws governing AI, the federal government should step in and make its own plan, which on its face, I think would be a good idea. The problem is, when people ask, okay, so what's your federal plan? And like, how are you going to get it through Congress? Basically just start like hemming and hawing and try to change the topic.
Yeah, I think this one is going to face a lot of trouble in the courts. You have to remember, there was an earlier effort to pass a moratorium on state AI regulations. The Senate voted against it 99 to 1, right? So this is an idea that both Democrats and Republicans really, really don't like.
That was only one of the attempts. They also tried to attach it to the defense spending authorization bill. And that failed as well. So like after trying to sneak this thing in through several different means, they have now decided, well, we're just going to do it by executive order.
Yeah, and based on the reading I've done, that does not seem particularly legal. Although, of course, you know, what will the Supreme Court decide? If it gets there, who knows? But this remains one to watch because, as you say, Kevin, if states are truly forbidden from passing any AI regulations over the next year, it just completely changes the landscape of what these labs are and aren't allowed to do in ways that I think might be bad.
Totally. All right, what's next on our Hard Fork Wrapped?
Next up, well, you know, Kevin, we talk about many different nations on Hard Fork over the course of the year. But when it comes time to doing wrapped, there's really only one country that qualifies in the category of other country of the year.
Is it Australia?
You know what? It came very close. It was a very close race this year. But for 2025, according to Hard Fork Wrapped, our other country of the year is China.
China.
You know, China is our other country of the year for many reasons. I think it starts way back when DeepSeek's R1 model came out and we saw this temporary freak out over whether China was on the verge of overtaking the United States in AI development. That idea has since loomed over all efforts to regulate AI here in the United States. And while DeepSeek, I think, hasn't lasted in the public imagination in quite the way that we thought it was going to, it has continued to release some new updates and maybe those are actually worth talking about for a minute.
Yeah, so DeepSeek, first of all, I can't believe that was this year. If you had asked me, what year did the DeepSeek thing happen? I would have guessed 2023.
But yes, that was just earlier this year that the stock market reeled from the news that this Chinese AI company appeared to have trained a model that was competitive with our state of the art American models for much less than American companies were spending. I think that was always a little bit questionable as a narrative. I always, I never really bought the figures that were thrown around. But I think this was a big moment in terms of people realizing that China was not actually, you know, that far behind the state of the art. And I think they've continued to show that. They released some new models recently that show that, that maybe they're not right at the frontier, but they're, you know, a couple of steps behind.
Yeah, it is a good model. DeepSeek makes good models. There has been reporting that the number of free open source models used by American companies that are Chinese models has been on the rise all year and that now a significant percentage of US companies that are building with these models are using Chinese options. DeepSeek is a big part of that story, even if they are not as good as I would say the leading three frontier labs. What else went on in China recently, Kevin, that gives it the designation of our other country of the year?
So the big story this week has been about whether Nvidia can sell these chips, these H200 chips to China. These are some of the most advanced chips on the market. And for obvious reasons, Nvidia wants to be able to sell them to companies in China. Jensen Huang, the CEO, has been lobbying hard for this. Currently, they are limited in which chips they can sell to Chinese tech companies. And a lot of people in the US national security establishment and at the frontier AI companies have been very worried about this. They've basically been saying, look, China is our biggest strategic adversary. We know that AI is a huge priority for them. Why would we allow them to have these very powerful chips and sort of erode the lead
that we have over them when it comes to creating these frontier models? Models. So a lot of folks have concerns about this, but in the Trump administration, they have basically said, we don't think this is going to be a strategic advantage for China. And so Trump this week said that Nvidia can now sell these more powerful H200 chips to China, and the government will take a 25% cut of these sales. Now, the one interesting wrinkle here is that it's not at all clear yet whether the Chinese government actually is going to let Chinese companies use these chips. They have been building their own chips domestically. Companies like Huawei have been developing their own chips. But I think it's fair to say that they will find some use for these advanced chips if they are allowed to buy them.
Well, here's my case that this gives China a competitive advantage, Kevin. There was a story in the information this week that said that DeepSeek is developing its new AI model using several thousand Nvidia Blackwell chips, which are Nvidia's top of the line chips, which remain banned for export to China. And apparently, DeepSeek was able to smuggle those chips into China via third-party countries, essentially, you know, having some other country like Malaysia, you know, make the purchase happen there, get it into China. So, you know, if DeepSeek thinks it's worth running a smuggling operation to get these top-of-the-line chips, my guess is that China is actually going to find it useful to get these H200s into China.
Yeah, I mean, this to me just feels like a complete victory for the Nvidia lobbying machine, which has been pushing hard for these, you know, the relaxing of these regulations for some time. I don't understand why anyone in the Trump administration thinks this is a good idea. I've heard their arguments about how the entire world's AI stack needs to be sort of American. But I think that allowing your biggest geopolitical adversary to have access to the key ingredient in building powerful AI systems at a time when American companies are racing to stay at the frontier, I just can't see any good reason for that, other than we got worked by Nvidia's lobbyists.
Well, Kevin, that brings us to the final item in this year's Hard Fork Wrapped, and that is the most podcasted about podcast. And this year, our most podcasted about podcast was our interview with Roblox CEO, David Bazooki.
Yes, this was a wild one. We haven't actually gotten the chance to catch up about this because it aired on the show a few weeks ago, and then we immediately went into another episode that we had been planning in the holiday break. But Casey, this was by far the most response that we got to an episode all year. If you have not listened to that episode, you should definitely go back and listen to it. What feedback did you get, Casey?
I mean, I think we honestly just got a lot of really nice notes from listeners. I think that listeners like us get very frustrated with the world of tech. And as journalists, we get the rare opportunity to actually sit across the table from people who run these companies and try to ask them the questions that are on the minds of a lot of their listeners, like, why did you let strangers contact children on your platform for 20 years before you thought to do something about it? And I think the response just really shows that people are craving some kind of accountability for the decisions that these platform executives are making. What kind of feedback did you get?
Yeah, I mean, people were really upset about how Dave Bazooki answered our questions. There were some other podcasts, as you mentioned, that sort of either followed up or covered this in some way. Some YouTube creators took it upon themselves to make videos about this situation. There were lots of articles in like gaming publications. This became like a news story, and I was surprised, not by the sort of reaction, I think in the room with Dave, it felt very tense and confrontational and like it had some explosive potential. But I think I was sort of just surprised because I don't think of us as a podcast that like brings on CEOs and kind of ambushes them with like hard hitting gotcha questions, like-
Obviously, we're journalists and we feel a responsibility to ask tough questions of powerful and important people. And leaders at these companies certainly are powerful and important. But I was not ready for a fight when we went into that interview. And I think he was just so defensive and so taken aback by our line of questioning in a way that was frankly, kind of wild to me because, you know, his team had been fully aware that we planned to talk about child safety and yet he seemed totally unprepared for those questions. So I was just, I was sort of surprised and I guess I was encouraged by the feedback. We had lots of parents writing in saying, you know, this, this sort of changed the way I think about Roblox or I'm not going to let my kid use this now.
And to me, that just suggests that like doing journalism has consequences. And I feel good about what we did and how we showed up that day.
Yeah, we also just got a lot of emails from people whose kids or whose family members had had terrible experiences on Roblox. And that was actually quite eye opening for me because I think as somebody who doesn't cover Roblox that much, I just had not been aware of just how many kids were having trouble on this platform. You know, I'll say, Kevin, like setting aside the kind of, you know, tougher questions that we had planned to ask him, just because they had been very much in the news and they seemed like they deserved answers. At the end of the day, I think the answer that he gave that surprised me the most was when he said essentially that, yes, he did think that we should consider putting prediction markets into these games for children, like essentially onboarding them into the, you know, a vibrant and dynamic world of gambling.
Like that was the moment, like sitting here in the studio where I was just sort of like, I can't believe we're talking about this. So that would just really surprise me and I think will maybe stay with me as long as, you know, anything else that happened during that interview.
Yeah. So one theory that I heard in the days after that episode that I wanted to ask your opinion of is like, there's this kind of universe now of podcasts, of shows where CEOs can kind of go to get kind of softball questions from friendly interviewers. And someone I know was saying like, this has basically left the CEOs sort of unprepared for encounters with real journalists like like taking your like, you know, your, your dog and like expecting it to live in the wild after it's been domesticated. It's like these CEOs just, they've spent so much time going on these friendly, uh, you know, non-confrontational podcasts and, and doing these sort of interviews with these, these creators
that they have sort of forgotten what it is like to be interviewed by a real journalist. Do you, do you buy that at all?
I mean, I do think you can just observe over the past five years, CEOs in general have been spending less time going on podcasts hosted by journalists and more time going on podcasts hosted by creators. And I think there's a lot of stuff in that, that happens in that world that is like undisclosed and frankly, kind of shady. Like I'm aware of podcasters who are accepting like $25,000 to, to agree to interview somebody, right? Like there's a lot of pay for play that's going on out there. And that is, I think, kind of maybe dulling some of the senses of these, these CEOs. But, you know, at the end of the day, I think these are people, these are, these are literal billionaires, right? Dave Bazooka is a billionaire. Um, these people have profited
massively from the success of their platforms and they should just be able to answer basic questions about safety and not feel like that is, um, like a vicious attack on their, their character. So yeah, maybe that is a part of the dynamic here, or maybe Kevin, it's just that ultimately the CEOs of these social platforms in particular have just never wanted to talk about this stuff. Yeah. I mean, I feel like that's how it's been with a lot of the CEOs that I cover. They've always hated talking about the safety, the policy questions, the stuff that involves difficult trade-offs, like cases where there were harms on their platform. They have really always shied away from talking about it. And, you know, the more that you shy away from talking about something, probably the worse you're going to be at talking about it. So I think there's a lesson there.
Totally. All right, Casey, that is Hard Fork wrapped for 2025. Uh, what have we learned about our show and the topics that we cover?
I think based on what we just covered, what I've learned is that Hard Fork is one of the most interesting shows in the entire world. And I want to give credit to you and the entire team for a great, uh, 52 episodes. We will continue to be making fresh Hard Fork episodes for you through the very end of 2025, and that's the Hard Fork promise.
We never rest, just like Santa. Actually, he does rest.
We're 996, baby.
Yes. At least Santa gets like the summer off, you know?
Santa is asleep at the wheel and we're grinding.
Your agents can talk to each other. They still can't think together. That's the gap Outshift by Cisco is closing. Outshift is Cisco's incubation engine, building the internet of cognition. Shared intent, shared memory, and guardrails that allow agents to work together from anywhere. Open source protocols and code. A system for intelligence that scales. Read the paper and experience the demo at outshift.com. That's O U T S H I F T dot com.
Today's AI builders need infrastructure that scales with their ambitions. Iron is a vertically integrated AI cloud platform, delivering data centers, compute, and software for AI training and inference, underpinned by its expansive portfolio of land and grid-connected power in renewable-rich regions globally. Iron's AI cloud gives direct access to Nvidia GPUs, scalable from hundreds to thousands. For builders, that means direct access to AI compute, rapid scaling, and predictable pricing. Iron, AI cloud for the next big thing.
It's time to innovate at the speed of AI without compromising control. That's OutSystems. Take a look at what happened with one global retailer during a period of massive expansion. They used OutSystems to build AI agents to automate invoicing and translation processes, shifting 95% of their daily store operations to a new digital app, saving thousands of hours of manual work. OutSystems gives you the flexibility to deploy how you want and the governance to do it safely. Own your agentic future. Learn more at outsystems.com/NYt. That's outsystems.com/NYt.
Casey, before we go, let's make our AI disclosures. I work at the New York Times Company, which is suing OpenAI and Microsoft over alleged copyright violations.
And my boyfriend works at Anthropic. Hard Fork is produced by Rachel Cohn and Whitney Jones. We're edited by Jen Poyant. Today's show was fact-checked by Will Peichel and engineered by Katie McMurren. Original music by Marian Lozano, Rowan Nemestow, and Dan Powell. Video production by Sawyer Roque, Pat Gunther, Jake Nickel, and Chris Shott. You can watch this whole episode on YouTube at youtube.com/hardfork. Special thanks to Paula Schuman, Puywing Tam, and Dahlia Hadad. You can email us at [email protected] with your water usage for the year.
You can't undo everything in life, but you can undo prediabetes. More than 2 in 5 adults in the US have prediabetes. Most don't know they have it. But with early diagnosis and action, you can delay or even prevent type 2 diabetes. Visit doihaveprediabetes.org to take the one-minute risk test and start undoing the things you can. Brought to you by the Ad Council and the Centers for Disease Control and Prevention.
Page updated October 12, 2026 · transcript and summary are machine-generated
