Joey Politano on the AI Investment Boom and Trends in Economic Growth

Joey Politano is an economist and a commentator who writes a popular Substack newsletter on economics. Joey is also a returning guest to Macro Musings, and he rejoins David to talk about the AI investment boom and broader economic growth trends. Specifically, David and Joey also discuss generational differences in economic perspectives, the increased demand for nuclear energy, the importance of AI in driving scientific research, and much more.

Check out our new AI chatbot: the Macro Musebot!

Read the full episode transcript:

Note: While transcripts are lightly edited, they are not rigorously proofed for accuracy. If you notice an error, please reach out to [email protected].

David Beckworth: Joey, welcome back to the program.

Joey Politano: Thanks for having me back. It's always a pleasure, David.

Beckworth: Last time you were here, we were talking about your business, your newsletter. You're very entrepreneurial. You're living the American dream of sorts. Tell us about that. How's it going?

Politano: I am fully self-employed, for people who don't know. I write the newsletter. The newsletter is my, basically, entire source of income. It's all from people who pay for whatever paywall pieces I put in the newsletter. We just passed two years of that being my full-time job, which is crazy to think about. I remember when I started this— it's very funny. I started it as like, "Oh, I'm going to write these things, and then someone will see it, and then hire me for an RA position. They'll think my writing is so good.” Now, it's like there's actually hundreds of people who think that my writing is so good, which is very flattering.

Politano: It's been really interesting. Now, I got into this writing a lot about capital and macro stuff, like inflation, labor markets, the Federal Reserve, interest rates, and I still write a ton about that, to be clear. But it's been interesting to try to broaden horizons a bit more, I think, as you start out. I talk a lot more about industrial policy stuff, or industry trends, or international policy, and I think that that's been very rewarding, to be able to just say, "Hey, I think this thing is very important. I'm going to show it to a lot of people. I hope they also think it's very important." So, it's been a real blessing.

Beckworth: Yes. I've been a subscriber for some time as well, and you've had a spate of fantastic articles out. We're going to talk about some of them— the AI investment boom, productivity growth, potential tariffs we might see. I think all of these articles that you've been doing lately are very topical, because we've had an election, in part tied to our relationship to China, globalization, the world economy. So, for some economists like myself, we're a little nervous about what to expect. We're used to globalization, the beauty of markets, and maybe we won't have that quite as much on the margin.

Beckworth: So, you're going to walk us through where we stand on those issues. Now, Joey, something else, though, that you're at the forefront of, and that is the mass exodus from X, formerly known as Twitter, into Bluesky. I've gone there too, but you're at the cutting edge of this, and I just see you engage in that space. So, I've got several questions on this. One, what is your feel for this? Is this a sustainable exodus? Are we still going to see people at both platforms? Where do you see this all settling out at the end?

The Competition Between X and Bluesky

Politano: I think one thing economics teaches us is that network effects are really hard to break. For me, especially when I was starting out, the most valuable part of Twitter was the ability to post something and then have someone reply with something insightful, adding on or contradicting what I was saying. It was a very many-to-many social media platform. And it’s kind of become, in my mind, a more one-to-many. The relationship that, if you think, a YouTuber has to their audience— the YouTuber creates a video, but very few people in the audience are then going to make a video responding back to that.

Politano: It's really a broadcast channel. It just happens to be that there's a billion channels instead of three. That's the direction I feel like Twitter has been going in over the last two years. The big thing is, if you post something, lots of people can see it, but very rarely am I getting, "Oh, have you actually seen this data point that's really interesting? Have you actually read this paper that's very interesting?" Moving to Bluesky, it's a lot more many-to-many in that way, where I post a chart, and somebody links something very interesting related to that.

Politano: Then, possibly, you're going back and forth in a discussion about what is correct, or, possibly, you're going and learning, "Okay, maybe I need to reinforce this argument. Maybe this argument's wrong. Maybe I misunderstood this phenomenon." But it feels very different than on Twitter, where it's like, I post a chart, people see it, and that's it. That's the experience now. So, I've really been enjoying it on Bluesky. I think it is a little— The greatest problem is it's a little too tech-brained. I'm very tech-brained, but they're like, "Oh, you should set up your own feeds to follow."

Politano: You know how on most social media apps, you have a following, and then a “for you.” It's algorithmically determined, and then [there is] a list of people that you follow. They're like, "Oh, you can add the ‘EconSky’ feed, which is anything that has keywords based on economics, or you can do the “Catch Up” feed, which is the most popular post of the last 24 hours." Those are all very interesting, but your average person doesn't interact with them, because they don't— It's like asking them to assemble parts of the car from scratch, and they're like, “I just want this thing to drive, buddy.”

Politano: So, I think they're going to exist in parallel. I think the very interesting thing is the— Max Read, if you know his newsletter, he writes an “about the internet” newsletter, and he called it the “Email Job Blob.” People who spend all their time— they work in tech, they work in finance, they work as journalists like me, and so they have to spend a lot of time on the computer, and they want to ingest a lot of information. A lot of those people are moving to Bluesky. Conor Sen is a great example. I was talking to him about this, and he's like, “I’m enjoying the conversation more, even though Twitter is still good for that one-to-many broadcast.” The other thing that I'll say is kind of tough and is very personal— but it's really hard to post articles on Twitter now, because they very heavily downgrade links, and so what I do is I basically write a thread of just the charts. Basically, I rewrite the article as a thread, and then post that on Twitter, and then link the article.

Beckworth: Oh, that explains so much, Joey.

Politano: Well, in one sense, I understand why they do that. It's because they don't want people to leave the website and not come back, but on the flip side, it then makes it annoying for me, because the reason that I'm on Twitter is partly as advertising for the actual serious long-form work that doesn't fit in 200 characters, and it's hard to share that.

Beckworth: Yes, well, our own podcast producer, who's in the room with us now as we're recording, Sam Alburger— he posts material from the podcast, and he's been doing that very same thing. He'll post a video or some text, and then you go several posts down, and then you get the final link to the article. So, hopefully, you get the attention with that first— I want to call it tweet— and then later on. I didn't catch on to that for some time. “That's great. Joey's sharing all of his articles.” I don't need to go— even though I'm paying for it— I don't need to go over there and actually read it.

Beckworth: So, it all clicks, and I found out the hard way, providing a video with a link directly to the piece. No, it's not going to get the engagement that you want it to. What this tells me, then, is folks like me need to continue, and you need to continue engaging on both sides and cross-post to both places. So, you have that impact on X or Twitter, and then Bluesky to get the engagement, the conversation that is largely missing from the original big, important social media companies. So, when you cross-post, I guess, help me out here. Do you literally have to do it two times, or do you have some app that allows you to automatically do it?

Politano: There are apps. I am a little particular, so I just do them separately, because the app sometimes frustratingly will mis-link things in ways that I don't trust. But it's more often things that are happening in parallel, where I'm going to post something on Bluesky, and then it gets certain replies, and then based on the replies, I'm going to respond. Or the number one way to engage on any of these platforms is the quote post, quote tweet, whatever. And so, obviously, I can't— If someone posts something on Twitter, I can't then quote tweet it on Bluesky.

Politano: So, there's parallel conversations happening. It's interesting. I think Noah Smith actually wrote something very interesting about this maybe two years ago, about the fragmentation of the internet. The point being that people are using— stuff like Telegram is very big now. Discord is very big and getting only bigger, these parallel, not quite algorithmic social media feeds that people are— Instead of using Tumblr or Twitter or Reddit as, “This is the community, we don't need a parallel engagement,” it's like, okay, you have this thing on Discord that's the community, and then, occasionally, it spills over onto the public internet.

Beckworth: Okay, well, I guess we will both continue to cross-post and engage policymakers, other journalists, and those in our policy space who care about these issues. That's what I need to do. You need to make money as well as engage with the people in your audience. Alright, one last question before we get to your pieces, though. I wanted to ask this question, because I think, again, you know people in this space well, and technically you're a Gen Z-er, although you're more mature, maybe a little more seasoned. Are you a Gen Z?

Politano: I'm 26, and I'm Joey, and that's--

Beckworth: You're somewhere between the two.

Politano: But yes.

Beckworth: Okay, so my question is this. If I think about a millennial, they had their baptism into macro and finance during the great financial crisis. Gen Z-ers, some of them may not even remember 2008, or if they do, it was an uncomfortable time or season. For them— where they cut their teeth— their baptism was the pandemic and inflation. In fact, we went through an election where many people are saying inflation was more consequential than many of us anticipated. So, is it fair to characterize that these two different generations will see macro issues differently— the millennials, and those who went through it. Even myself, I'm not a millennial, I'm a little older than that— we still think about the zero lower bound where the Gen Z-er's are like, “Man, inflation is a real beast. We need to get that under control.”

The Generational Differences in Economic Perspectives

Politano: Yes, I think there's a lot to that. It's very funny that you bring this up, the idea of not having memories of 2008, because there was a trend that was viral on TikTok for a very long period of time where the explicit framing of the joke that was mocking the fact that you, as a kid, at like eight years old, wanted something silly, like you wanted the new Pokemon game or whatever, and then in the other room, your parents are fighting because their finances are falling apart because it's 2009, and they've shielded you from this.

Politano: So, you had no idea, and now you're looking back and you’re like, “Wow, that was crazy of me at eight. I almost ruined everything by asking for Pokemon Platinum or whatever.” I do think that when you look at a lot of these long-term views, there's a lot of research that people who lived through the 70s inflation, for example, in their, let's say, 20s to 30s, retain that as a long-term memory that informed how they engaged with future policy in a lot of ways. They remained more worried about inflation over the long term. They remained more sensitive to gas prices, is another good one, because the 70s had so much focus on gas prices.

Politano: I think that there's definitely going to be an element of that, where the millennial generation is a little more employment-focused, and the Gen Z generation, to the extent these are discrete variables, are more inflation-focused. And I think you may have seen a bit of that in the election. We talk about young people swinging towards Donald Trump by historic margins, and I think some of that is— even though, on a lot of metrics, people hold very liberal views, they're also much more sensitive to inflation in a way previous generations of young people weren't.

Politano: The broader thing I worry about is pervasive zero-sum thinking. This is something that you see in research on people who grow up in Italy, kids who grew up in Italy. Italy has, historically, very poor economic growth, especially over the last 20 years, and so a lot of people who grew up in that environment inherit this framework where growth is not normal, and so you're fighting over shares of the pie more than trying to grow the whole thing. And even though we've had a lot of growth in the US over the last five years, relative to other countries like Italy, I think that that sentiment exists, and I think that sentiment is actually very exacerbated by inflation.

Politano: This feeling of, “I only have a small part of this economy, and it is threatened by external forces that are trying to take it away,” and there's not a, "Okay, well, holistically, I expect—" that idea that if you're going to go onto the street in New York and find somebody who's 25 and ask them, "What do you think your future's going to look like?" They're like, "Oh, well, I'm going to get a great job, I'm going to go buy an amazing house." That's just not been the mentality for a long time, and I think that that that contributes to— it's self-fulfilling in a little way. People don't believe that growth is possible, so they're not prioritizing growth, so they vote for candidates that give me a larger share of the pie, or candidates that have very zero-sum mindsets on trade or other parts of the economy.

Beckworth: So, what you're saying is, yes, inflation is a defining moment or even characteristic of this generation, but maybe more importantly is they're slowly incorporating a zero-sum mindset, and you know what? Throw on top of that the inflation concern, and let's just be inward-looking, and we're going to talk about some trade policy, some tariffs, which is a reflection of that thinking. It's us against them. Minimize the number of immigrants coming in because they're taking our jobs, whereas you and I know well from the research that, actually, it's a win-win situation, typically, is the case.

Beckworth: There may be some transition costs involved. So, that'll be interesting to follow and to see. Maybe what we're going to talk about next, the AI investment boom, will be the dose of optimism that can break this growing zero-sum mindset. So, let's transition to that, and before we do, these are all based off of your articles. Tell the audience the name of your newsletter, since I will probably mispronounce it if I attempt it.

Politano: Yes, don't worry. The mispronounce-ability is a core part of the branding at this point, but it's apricitas.io, so it's a Latin word. It's A-P-R-I-C-I-T-A-S.io. I picked that word because it means sunniness or sunshine, and so the idea was trying to be illuminating through data, and I think that that metaphor eventually broke down once. But that's where I post everything. It's, as with almost all Substacks, also an email newsletter sign-up.

Beckworth: And then, they can follow you on X as well as Bluesky. I recommend that you do. I will say this, Joey's newsletter is probably one of my favorite ones that I pay to subscribe to. It's one that I look forward to. You put a lot of hard work into it, and it's evident. So, let's jump in and talk about some of the pieces that you've done, and you have one recently called, *The AI Investment Boom,* subtitled, “AI demand is driving skyrocketing US investment in computers, data centers, and other physical investments.” This is a good news story I want to read, so tell me about it.

Breaking Down the AI Investment Boom

Politano: Yes, I think there's a really interesting shift, in the first part, is that we've had a tech industry in the US that is world-leading for probably 30 years at this point. But if you look at the tech products that were very big and important of the 2010s, it was all very lightweight software network products, and the examples that I was talking about were— it was wild when Facebook bought Instagram for a couple billion dollars, but Instagram had like 10 employees. There was nobody. It was an incredibly valuable company, very productive, [and] added a lot to the US economy, but it didn't need this giant industrial backing the way a car manufacturing plant does.

Politano: The very interesting thing to me is now that we've shifted onto this era of artificial intelligence and generative artificial intelligence like ChatGPT, all of these image creators, and a lot of the stuff that is one layer removed from consumer products— This is the other thing that I think people forget, that autonomous vehicles are also business-to-business AI stuff that you don't see, but is very important and productive. All of that requires a lot more physical hardware than Instagram does.

Politano: You can't run this kind of a company— If you're running an AI company, you can't— You can hire 20 people, but then you're also going to need access to a massive warehouse of GPUs in order to train or design or use inference on your AI products. And so, the point of that piece was, here's all of this crazy amount of physical investment going on over the last two years within the United States, basically to enable the training and deployment of all of these AI models. Just to give you an example, if you look at data center construction in the US, that has been like hitting record highs every month for two years plus.

Politano: It's now at the point where we spend more on data centers than we spend on almost all retail construction in the US, if you think about restaurants and shopping malls and all of that stuff. And we're closing in on $30 billion a year, annual rate. If you asked me to guess, I think we're going to make that before the end of this year. That's just the construction. Those things have not finished yet. The second part of that is that that construction is only reflecting the physical building that the data center houses, right? The actual valuable parts of a data center are not the physical building. It's the computer wiring inside.

Politano: That stuff is incredibly in demand right now as anyone in the industry will tell you. And so, that shows up a ton if you look at, for example, US imports of large computers, so excluding laptops— up a ton and at record highs and keeps breaking those record highs. If you look at net imports of computer accessories— keep in mind, a massive warehouse of computers needs tons of fans and tubes and server racks and all this stuff— also at record highs. Increasingly more of those imports are coming from Taiwan, which is the number one producer of all of the very leading edge computer parts.

Politano: TSMC is a manufacturer for almost all of the best stuff that is used for AI training. So, that's the first thing. Look at all of this investment that's going on, from first principles, to actually run these AI systems. Secondarily, you need investment to run the things that are running the AI. And so, you've seen an amazing surge in deals from companies like Amazon or Microsoft or Google to build electric power resources in areas where they're trying to build massive data centers, because they are industrial facilities that consume industrial amounts of power.

Politano: And we're recording this from Arlington, Virginia. So, we're like a stone's throw away from the largest computing cluster on planet Earth over in Loudoun, and there's actually big agglomeration effects to data centers. You want them to be located near each other and near large pieces of infrastructure, cables and power and weather stuff, that are valuable for this training. And so, there's lots of building going around here to try to keep up with data center demand and then also to service all of these data centers. So, it was a very interesting swap from, okay, first, you had this very employee-light and also hardware-light company. This is Instagram, right?

Politano: Over time, those companies have gotten much more employee-intensive. Facebook was hiring thousands of people a year. They're still not the largest hire— and they're not a labor-intensive company the way Amazon warehousing is. Amazon Web Service is more comparable. But now, they've shifted a lot of that energy into hardware, physical investment spending. Then, the counterpoint to that is if you look at tech hiring, that's down a ton. So, if you look at software publishers— this is one of the couple of charts I made in that piece.

Politano: Software publishers— if you look at their hiring over the last year, it's basically a goose egg. It's zero, and it was negative for most of the year prior. But if you look at their physical investment, it's skyrocketing. It's double or triple what it was just a few years ago. And so, that's a very interesting change, and this is something that I expect is going to become even more of a big issue. This is something where the US leads, and we don't want to give up that lead for obvious reasons, and there's a lot of people in San Francisco and in DC who are convinced by this idea that AI is a generationally transformative technology.

Politano: We can't lose it from an economic point of view and also from a security point of view. And so, we're going to do whatever we can to service it. I would also expect that a lot of this investment is also partly downstream of the CHIPS Act. It's kind of funny looking back that the CHIPS Act was designed for this world where we thought the best chips were going into mobile phones and that we're just going to be making those for a while. I expected and continue to expect, regardless of the election results, that there will be some attempt to craft a second CHIPS Act that would be more concretely addressing data center chip demand and data center construction more broadly. This is something where the investment is really picking up steam just in general.

Beckworth: Joey, that is a fantastic story, development for the US economy, for the outlook, on a number of dimensions. I want to go back, though, to the initial grounding that you provided for that, and  that was comparing this AI boom to, say, Facebook and Instagram. I bring that up because I think it's very useful to illustrate the critique that Larry Summers had surrounding secular stagnation when he first made this argument. He said, "Yes, there's a lot of consumer surplus. Yes, we're happier because we have these technologies, the internet, social media, but we're not investing in capital. We're not building. Therefore, we're going to have low interest rates, low rates of return, and expect low rates, a depressed economy." Now, it's the opposite of that. We are investing massively. We are seeing real rates go up. So, maybe we have left behind the world of secular stagnation. What do you think about that?

Have We Left Behind the World of Secular Stagnation?

Politano: I think I would generally agree with that. I think you can still very clearly see— it is kind of funny that we talk about [how] we've left the world of secular stagnation with interest rates, where they are, as the US is running a massive budget deficit. Realistically, part of the reason that interest rates are holding so high is because we're spending so much, fiscally. But it is a very different world than it was 10 years ago, where you could plausibly say— there was an argument of like, "Okay, this is a structural long-term factor that is going to be totally suppressing growth, interest rates."

Politano: “Man, we're stuck at the zero lower bound. Maybe we're going to be stuck there permanently like Japan is.” Then, there was like a second set of arguments that was like, "Well, if we did X, Y, Z, A, B, C—” varying between the economists making the argument, “—we could actually exit secular stagnation. Then, we would achieve higher interest rates, higher growth, et cetera." I think that we— collectively, the second class people won in the sense that there were things that we did that overcame this structural change.

Politano: Now, the 2010s look a lot more like the Great Depression than they do at the beginning of a new era that is subsuming everything, where you're like, okay, the Great Depression was a monetary policy error. It took, depending on how you want to make the argument, between 5 and 15 years to exit that error. But once you did, it wasn't like people in the 1960s were looking back and [saying], "Well, because of the Great Depression, we're stuck in this era of secular decline." No, they looked at that more as a lesson to avoid a very horrible business cycle than a structural issue.

Politano: We did much more loose monetary policy in response to the 2020 recession. We did a ton more fiscal policy. And if you had the very simple test of like, "Okay, what would you expect of that?" You'd expect trend or above trend growth and probably a decent amount of inflation. That's what we got. Obviously, there's a ton of nuance within that story, and I spent three years writing about all the nuances in that story. But at some level, that prediction held in a way that looks a lot better. I think the specific story about AI investment is important, but also, there's lots of investment going on throughout the economy.

Beckworth: Yes, as I sit here thinking, I think there's two parts to escaping the secular stagnation story. One is, we did apply very aggressive aggregate demand policy. Now, again, maybe it wasn't intentional. We're just, in some sense, making up as we went along during the pandemic through everything, including the kitchen sink, and hope something sticks. We added $5 trillion. Were we excessive? Yes, but maybe we should also be grateful because it snapped us out of a funk. Then, we're snapped out of the funk, and yes, maybe it contributed to Trump being elected.

Beckworth: But we're snapped out of the secular stagnation funk, and it just so happened that the AI boom takes off right at about that time. So, we have policy that gets us out, but then we also have a sustainable foundation to keep it going. That leads me, I guess, to the next question, and that is the timing of this. So, looking at your charts, I'm guessing you think a key part of this timing, and maybe I'm wrong, is the public revealing of ChatGPT. At least, in the time series, you move along, and there's a spike, so everyone gets excited. Investment demand goes up. Is that a fair telling of what happened?

Did ChatGPT Kickstart the Next Chapter of Growth?

Politano: I think so, and it's actually a really interesting story to look back on, because I was one of the people playing with GPT-2 a lot. There was a great product— I wish I could remember the exact name, but it was like a Dungeons and Dragons game that you could play against ChatGPT-2. The fun of that game was that GPT-2 was so stupid that it didn't understand what you were trying to explain to it. And so, you would try to be like, "Okay, I want to move over here." Then, it would forget what room you were in. Then, it would say that you walked into lava or something.

Politano: It would repeat itself. You'd say that you defeated the dragon, and it would say, “A dragon shows up.” You're like, "That's not what we're doing." So, it was kind of radical to me when I saw GPT-3 come out and I was like, "Wow, this is very cool. It's a lot better." Then, I didn't use it a lot, because it removed the fun of the AI being bad at games that you were trying to play. Then [with] ChatGPT, the really interesting thing to me is that GPT-3 had been out for a while. ChatGPT was the user interface. It was the chat part. Restructuring it in this way that it's conversational, that you're talking to the robot instead of just providing text prompts was actually really important.

Politano: And I just think that that clearly demonstrated to so many people that, okay, this technology has a lot of potential, and we need to do something to seize on that potential, and it was also that there were very, very clear first-mover advantages for OpenAI. I think you could make a really strong argument right now that Claude, which is Anthropic's main product, is better than ChatGPT, even the more recent versions that are updated and whatnot. But the user base is much smaller, because there were a lot of people who just— you started using ChatGPT and then, now, there's an entire layer of products that are built on top of it to integrate it.

Politano: And so, yes, there was a ton of movement from companies like Facebook and Amazon and Apple and Google to like, "Okay, here's our moment. We have to catch up here." A lot of the investment story is kind of— in one sense, it's these vertically integrated companies that are doing the construction and the training and the development and the deployment. In another sense, there's some of these companies that are shovel sellers. They're not developing products. They're developing infrastructure, and then they sell that infrastructure to startups that are developing products.

Politano: I'm optimistic about AI stuff. I think it's tough. It has a tough sell politically, in part because a lot of the stuff that is very valuable about it gets deployed at the back end of some business in a way that you'd never see, and a lot of the stuff that's very annoying about it gets deployed at the front end, on the internet, where you have to see it all the time. I took one of the autonomous, Waymo vehicles when I was in San Francisco, and that's here now. They are doing 300,000 rides a month in California. That's a real taxi service level of product, and that's within one state and functionally within parts of one city, and I fully expect that to be deployed more widely. If you think about just that product, driving has got to be one of the most common labor tasks in the world.

Beckworth: In San Francisco of all places.

Politano: Well, maybe not in San Francisco, but I'm just saying, in general, how many labor hours are taken up by driving in the world?

Beckworth: Oh, definitely.

Politano: That's a very clear and demonstrable case where it will--

Beckworth: It's a clear sign for sure. 

Politano: Right.

Beckworth: I say San Francisco because they have all those hills, its downtown traffic, and if AI can handle that— because that's always was the critique is, okay, it's one thing to talk about driverless cars out in the country. There's not many challenges. But when you get into cities, sometimes you have to even break the rules to get somewhere, so to speak. So, if AI can master that, you should be expecting great things.

Politano: Yes, I will actually say, Tim Lee has a great newsletter called Understanding AI, and he has written a lot of pieces about— He's a very big autonomous vehicle advocate, and he's written a lot of pieces about this. The most interesting thing to me was it's almost the opposite, the way you think, because, you think, driving on a highway, it's flat and empty, and there's just cars in front of you. What could possibly go wrong? Driving in a city, there's pedestrians, there are people breaking traffic laws, and there's a whole bunch of stuff that's happening at any given moment, [so] much more potential for disaster. Almost exactly the opposite, where driving in a city is much easier, because the vehicle can confidently just crawl forward if it doesn't know what to do.

Beckworth: The speed is much lower.

Politano: Right, the speed's much lower, and so if they reach a situation where they're like, "That cyclist is not supposed to be there. What do I do?" They just pause. They're like, "Okay, let me think this through." They can call home, basically, and have someone take over. But if they're on the highway, they can't, because if they pause, it's going to be a really big problem.

Beckworth: Big pile up on the I-65, yes.

Politano: It's actually very interesting that you brought this up, [because] they're now deploying on the highways in San Francisco. They're doing test drives, and they're talking about [how] they're going to be— They've had the legal authority to deploy on the highways for a while, and they're talking about doing it for real sometime very soon. That's another big progress. They had this challenge, that it was difficult to get autonomous driving on highways because of the speed risks, and now they're more confident that they can handle that kind of thing, and that's only in the span of two years or so.

Beckworth: Yes, it's more data for AI to process, to learn from, so good stuff. I was talking to a friend, and he told me that he told his young daughters, "You know what? You're never going to have to have the problem that I had with my father, where I had to take his keys away from his car." He goes, "I'll just have a car that will drive me everywhere I go," which may be true. But circling back to AI in general, before we move on to some of your other pieces, I think two fascinating things about it is, one, this increase in demand for nuclear technology. This is something that we've put to the side after Three Mile Island in the US in the '70s, and a lot of environmental regulations. We've just really been shy and timid, and now we have the profit motive. We have the right incentives to go back into it, as well as, I think, climate change interest as well, but I think this AI may be what gets us past that threshold, which really sparks innovation, increased demand for it, and you mentioned this in your piece.

AI and Increased Demand for Nuclear Energy

Politano: Yes, I think it's very interesting because AI, to be clear, is not a big source of electricity consumption on a national level. We expect things like electric vehicles or heat pumps, electrification of other parts of the consumer and industrial economy to be a much bigger deal than all of these data centers. They're industrial facilities, but there's lots of industrial facilities in the US, and I think that's something important to keep in mind. The difference is that these facilities are very concentrated power sinks.

Politano: So, we talked about Loudoun County. They have a ton of data centers in one area because they get all of these agglomeration benefits from being so close to each other, and because if you're a company like Google, it's really beneficial for you to be able to run training through one of these absolute massive data centers. So, it's a big competition to get as big as possible in one location. Then, you run into the problem of, well, we have an electricity grid that's built for diffuse consumption throughout the United States, and you're trying to introduce this massive consumer in one specific area that might not be ready for it.

Politano: And in a lot of cases now, they're using very weird— I'll give you an example, the Tesla AI stuff. They were using portable gas power generators. It's these truck-sized things that they'd park right next to the facility, and they'd just run gas through them all the time and then people would deliver more. One, it's hilariously expensive to basically try to reconstruct a power plant in the aggregate. Two, it's very much something, environmentally, that we want to avoid, and a lot of these companies in particular, like Google, have commitments to 24-7 clean power, yadda yadda, hence this big push for nuclear.

Politano: I'm hopeful, but I'm skeptically hopeful. Let's say that. There's always been talk of this, "Okay, well, now is the moment for nuclear," and every time, in the US, we kind of bungle it. I think it's telling that it's become more bipartisan. J.D. Vance, who's the new vice president, is very pro-nuclear. People in the Harris campaign had written stuff that was very pro-nuclear. And I think it's also telling that people view this [through] a bipartisan lens of, "This is an area where we are losing it to China." If you think the number one limiting factor on AI training is going to be electricity, I don't think it's going to be.

Politano: But if you think that, then the fact that China can build large nuclear power plants at some regularity that the US and most of the rest of the world can't is a very threatening thing, because it means, naturally, that those data centers would try to move to China where this electricity is more plentiful. So, that's where I stand on that. It's interesting. I think it's somewhat weird how much story the power part of it gets relative to all other parts. Also, water consumption, where if you look at water consumption for data centers, it's not that big, but it's something that gets a lot of headlines, I think, partly because they're controversial.

Politano: And I think, maybe to circle back a little bit more, I've seen a lot of takes— this might be my media environment, but I've seen a lot of takes that are like, "Oh my god, I can't believe we're getting nuclear and it's to make stupid AI slop," or, "Why should I turn off my air conditioner in a heat wave when there's this massive data center in Texas that's chewing up all my power?" That's the zero-sum mindset, again, that I sometimes worry about. And so, I don't think power is the number one constraint. I don't think it's going to be for the long term.

Politano: I expect the efficiency of chips, for example, to get better over time, but I think it is a very important part of the story, and it's telling. They turned on the Three Mile Island reactor again. The reactor had closed in 2019. They brought it back. Companies like Google and Amazon are investing in these small modular reactor startups that they hope will be able to deliver power in 10 years or so. They wouldn't be doing that if they weren't at least partly serious about this.

Beckworth: Again, it may not be the main economic benefit from AI, but man, to have these alternative energy sources— You’ve got to go have actual demand for them in addition to maybe easing the regulations so you can actually find it practical to use nuclear technology. So, that was the first thing. One last point, and then we need to move on, and that is the use of AI not just in driving cars, factories, but there was a recent article on AI in science, in research itself. That seems profound. It will help create new ideas, new discoveries, and efficiencies. That there, by itself, could really make a marked difference in the trend path of growth. Any thoughts on that?

The Importance of AI in Driving Scientific Research and Growth

Politano: Yes, I think that it's going to be a big continuation of the growth story if you think about how difficult it is. We know from experience that science is getting "harder." You can look at things like how many people are on Nobel Prize-winning papers in chemistry, and it used to be that you could get a Nobel Prize as one person. Now you need 300 people in a lab to be making these breakthroughs. That's very tough, and I think there is a lot of general benefit that can be gained by applying, say, an extra half-scientist for every scientist. You can shelve off some of the tougher parts of the work. People joke about ChatGPT being as smart as a dumb grad student. 

Politano: But dumb grad students do a lot of really hard work in the economy. We love our grad students, and that's actually profoundly important. I was at this conference that NVIDIA did here in DC recently, and I think that the most profound, important parts to me were when they were talking about the partial breakthroughs they had in, let's call it automating science stuff. A lot of it was very technical in ways that were maybe above my pay grade, but the ability to automate these parts of material science, where they were looking at very complex combinations of products, trying to find things that match precise strength and durability needs, seemed like actually very useful research that is perhaps not the most glamorous thing, but can have very large downstream implications.

Politano: I don't expect this kind of— I think that sometimes if you listen to people who are maybe too booster about it, they're like, "Okay, well, we're going to hit some inflection point, and then we're going to hit 8% GDP growth because the AI is going to have automated all of science." I don't think we're there yet, and I think it is also a bit telling that we have perhaps hit some local performance maximums with some of the large language models. I fully expect people to get past them eventually, but I think it's telling a story of like, okay, we had this S-shaped progress curve where it was doing bad Dungeons and Dragons. Then, it was rapidly doing high school level math. Then, it got to [being] helpful in developing new parts of science. Then, it petered off, because it's hard to get more information from that.

Politano: Maybe you need better strategies. Maybe you need more compute power thrown at these things. People are going to be trying all of that. But I'm remembering when one of these things happened very early, people were like, "Okay, well, if you really think that AI is going to 10x GDP because it's going to automate all of science, real interest rates should be so high, it's not even funny.” And real interest rates have gone up since that paper was published.

Beckworth: A little bit, yes.

Politano: A decent amount. An amount where you could plausibly say, maybe this is a better-

Beckworth: I told you so.

Politano: -but not an amount where you're like, “Wow, we're going to have colonies on Mars in a year or so.”

Beckworth: Yes, I recently had an individual named Zach Mazlish on the show, and he recently had a Substack [article], I don't know if you saw it, where he extensively went back and showed that people actually did experience lower real earnings once you do the proper measurements. And there's a popular measure that shows, for example, that the lowest quintile did better. But when you go from hourly to weekly to annual, across the board, that's not actually the case. But he also had a piece we talked about where he talked about this very thing. If AI, the AGI, the artificial general intelligence, kicks in, this threshold where AI is actually doing science for you, creating new ideas for you— If you had 10% real GDP growth, basic models tell you you're going to have real interest rates at 10%. What does that mean for macroeconomic policy?

Beckworth: It's a pretty profound, different world we would be in. Now, maybe with such rapid growth, we wouldn't care about the implications of 10% real interest rates, because it'd affect the discounting of stocks, could affect mortgage rates, all kinds of stuff. We're in that world, that's a small fish to fry. We’ve got bigger, more wonderful things happening. But let me circle around and talk about where we are today, since you touched on this, in the time we have left. Since we're on this topic of growth, AI, and where we are today, as you alluded to, help us understand, where are we today? Because you had a piece that mentioned, despite all these good things, manufacturing productivity isn't so fantastic. On the other hand, you had a recent piece that says that overall productivity is doing great, phenomenal, maybe even sustainable.

The Current State of Economic Growth

Politano: Yes, I think this is a really interesting tale of two cities. We've been doing a lot of very broad industrial policy in the US, I would say, going back six or seven years. You can trace that to Trump's first trade war and then to the CHIPS Act and the Inflation Reduction Act in the Biden administration. Very little of that has moved the needle on US manufacturing productivity. If you think about it, manufacturing productivity should actually be growing faster than the rest of the economy. You'd expect it to be easier to automate the production of cars than it is to automate doctors or food service or things like that.

Politano: In the US, it's almost the total opposite. Manufacturing productivity has been stagnant for, at this point, approaching 20 years. In the 2010s, you could point to that and say, "Well, all productivity is stagnant." We talked about the secular stagnation and stuff. And so, it makes sense that manufacturing is dismal. Why would you expect them to be doing well and the rest of the economy doing poorly? Everyone's doing badly. If you were to point to manufacturing and say that they're the odd one out, true, but they're not doing tremendously worse.

Politano: [During] the post-COVID era, or really, I would say, the last eight years, [it has been a] completely different story, where you had pretty robust service sector productivity growth in the four years leading up to the pandemic. That was when the labor market started strengthening, when we were approaching record employment levels in the US, record employment rates. Then, you had COVID happen, and there was a lot of fear of, "Okay, well, all this progress is dashed." And it was actually the flip side, where the productivity increased at a faster rate in the years post-COVID than it has in the years pre-COVID.

Politano: That is actually probably going to get revised up to even faster, because we're likely going to get negative job revisions in the coming months. So, you're looking at a very historic productivity surge in the US, a US that was already doing faster than most pure countries, like Canada, or the European Union, or the UK, or Australia, pre-COVID, now dwarfing, absolute— like no contest. I think that's a crazy story. And I think it's actually interesting in that how much of the story you can tell without talking about AI at all. I think this is partly a story of [how] we had— we talked about this before— the very aggressive macroeconomic response.

Politano: Of all the countries with large fiscal responses, we had the largest. I think that, secondarily, it's very interesting that, accidentally, we shuffled a bunch of workers around. In the UK, and in Canada, and a lot of other places, they had very strong furlough schemes. People were kept in a job where they were basically paid to do nothing for a bit during early COVID under the presumption, not totally unreasonable, that they'll just go back to doing their old job once COVID is over. We didn't have the infrastructure for that in America. It was just not possible. The Paycheck Protection Program, which was the closest thing that we had to that, was a mess and was set up retroactively because that was all we could do.

Politano: And so, you had an absolute ton of job churn in the US. One of my favorite charts from that piece— If you look at employment in the top 25% of wage industries, it's very detailed employment by, do you work in food service? Do you work in computer manufacturing? Do you work in information technology? If you look at the top 25% versus the bottom 25%, it's like they move up in tandem for years. Before COVID, industries like food service were gaining jobs at about the same rate as industries like tech or doctors’ offices.

Politano: Then, post-COVID, it's just total divergence for several years, where you have an evaporation of a lot of these jobs in low-wage industries and [then] people who are moving up in the world to jobs in higher-wage industries. And so, it's the story of someone who worked at Chipotle, lost their job, and then got a job that was higher paying at staffing a financial services firm. That was several years of that. When people were talking about the Great Resignation, that was the same thing. Then, we look back now, several years on, [and] those people— they've settled into their new roles, they've up-skilled in a really strong way, and now the whole economy is much more productive than it was a few years ago, and that's virtually all a service sector effect. The same thing does not happen in the manufacturing sector. If anything, you would say— I can somewhat confidently say that a decent chunk of the productivity increase is people being poached out of manufacturing into more productive service industries.

Beckworth: That's so fascinating.

Politano: And it was telling as a comparison to other countries. Like I said, telling as a way of saying that this probably didn't actually have that much to do with AI yet, big yet there, because they have ChatGPT in Canada. This was an interaction of both the macroeconomic forces and technology. It was like the two paired together and you can't separate one from the other. And now, we're in a position— collectively, the United States— we have this technology that, basically, could be a massive tool for service sector productivity, and we're in a much better place to deploy it than a lot of countries, because we've had this productivity growth over the last five, six years.

Beckworth: That is so fascinating, because typically we talk about the service sector as being the drag. We talk about, for example, Baumol's cost disease, that it's the tradable sector that usually has the efficiency gains. Then, you’ve got to pay your barber more, because you're getting paid more and you want to keep the barber in the town. He's got to be able to afford to live. And so, the higher real gains to the productivity-enhancing trade sector or manufacturing typically flows over. But this has completely reversed that story, which is shocking, surprising, and a pleasant surprise.

Beckworth: We want to see service sector jobs have massive productivity gains. So, [it is] an unintended effect of our dysfunctional unemployment system. Some might call it— In fact, many in 2020 were calling us dysfunctional. We didn't have state capacity. The curse of our system became a major blessing in terms of really supporting robust productivity growth. Another point, though, and as you were saying this, I was thinking through, and you have wonderful charts, you've made this point many times, on social media and in your post, is that voters saw inflation. They didn't see these counterfactuals that the US could have been.

Beckworth: And unfortunately, you and I can talk counterfactuals, economists can talk counterfactuals, but the voters, they don't see, "Wow, we could be in Europe right now." What they see is, "Well, I didn't have this inflation before 2021." So, I think, if nothing else, we've learned some important lessons about how much the public dislikes inflation, and if you want to be politically smart, you’ve got to keep that in mind. Finally, the bottom line is this, you're telling me an amazing story that suggests potential real GDP is lifting, and a sustainable growth rate may be above the 2% we've seen in the past, maybe two and a half, three, [and] maybe even more, I don't know. But we are, at some point, seeing a permanent increase in the growth capacity of our economy. Is that fair?

Politano: I'm hopeful that that's true, and I think it's also— not talking too much about counterfactuals again, but I think it's also telling that we, as the United States, have gone through this period where, if you look back at pre-COVID forecasts, they look dismal, in many ways, compared to where we are right now, on growth and productivity, [but] obviously not on inflation. And if you look at other countries, like the European Union, [they are] below forecasts. Countries like Japan [are] below forecasts. But even places like China, who we think of as having very robust long-term growth [are] below the forecasts that they had five years ago.

Politano: Then, conversely, we've been on top of the IMF's forecast list for two years running for short-run GDP growth, and if you look at where people think long-run GDP growth is going in the US, it's been upward revisions. If you look [at] where it's going in Europe, it's all downward revisions. And so, I think that's a big reason to be optimistic in the short to medium-term about the American economy.

Beckworth: Which is something that President Trump, I'm sure, will claim as his own handiwork, even if it was laid [out] many years before with other policies and such. Now, we are out of time, and, Joey, we did not get to your fantastic work on tariffs, speaking of Trump and [its] implications.

Politano: Yes, we have a lot more to talk about.

Beckworth: Yes, so we have a lot more. We could do a whole separate show. And so, despite all the good news we've talked about today, one concern might be those tariffs, because in those pieces, you note that an important part of the AI boom are these imported goods that make AI booms possible. So, we'll provide links to that piece in the show notes, but thank you once again, Joey, for coming on the show. Our guest has been Joey Politano.

Politano: Thank you so much for having me.