Fact And Fiction Surrounding The AI Bubble

0:07Groundbreaking. Transformative. A new way to unlock human creativity and productivity. Tech CEOs promise artificial intelligence will do many things for us. They've used those promises to justify billions of dollars of investment in building the language models and data centers needed to power AI. Four of the world's biggest tech companies, Meta, Amazon, Microsoft, and Google, promise to collectively spend$380 billion this year on the AI build-out. That spending coincides with huge rallies in these companies' stock prices. There are now hundreds of private AI companies with values on paper of over a billion dollars.

0:46And in October, the AI boom created the world's first company worth$5 trillion, NVIDIA. It just continues to go higher. It is opening at a$5 trillion market cap. The share is trading for$207.88, up another 3.4 percent. So is the spending justified? Do these companies' stock values hint at a financial bubble in AI that's like past bubbles, or is AI different? We get into it after the break. I'm Jen White. You're listening to the 1A Podcast. Stay with us. We've got a lot to get to.

1:25Joining us from San Francisco is Shireen Ghaffari. She's an AI reporter for Bloomberg News. Shireen, welcome to the program. Thanks for having me. And joining us from Cambridge, Massachusetts is Jason Furman. He's a professor of economics at Harvard University. He previously served as the chair of the Council of Economic Advisors for President Obama. Jason, welcome back. Thanks for having me. Jason, let's just start with the basic definition. What is a financial bubble? There's actually two parts to it. One is that asset prices go up a lot and then they go down a lot. There's not some formal threshold for either of those.

2:01But sometimes you see them fall 50%, 80%, 90%. percent. Saw things like that in the dot-com bubble. But the second, which happens in the real economy, where you're spending lots and lots of money building something like homes or railroads or data centers, and then you discover it's not profitable to build those, and you dramatically cut back on your plans and your investment and your real activity. So there's both a financial part in pricing and a real part in terms of economic activity. And when we've seen past bubbles, how long does it take for that bubble to build and then burst? Is there a timeline we can follow?

2:45There's no timeline. It's just been very, very different every time. And one has to be really cautious. The stock market in December 1996 looked like it might be a bubble. Alan Greenspan was the chair of the Federal Reserve at the time and said there was irrational exuberance. And over the next several years, after he said that, the market doubled before the dot-com bubble burst in 2000. And so if you had sold all your stocks the day you were warned by Alan Greenspan about irrational exuberance, you actually would have lost money because you would have missed out on the up and then missed out on the down.

3:26So these things are very hard to figure out the timing of them. So Shireen, what evidence do investors and industry watchers point to when they argue that AI shows signs of a bubble? Well, what some investors are concerned about is the amount of spending that's being done, particularly on infrastructure, so meaning the data centers, the vast amounts of, you know, compute power is what it's called, that's needed to sort of fuel AI software. And some of the companies like OpenAI or Anthropic that are spending a lot, you know, OpenAI plans to spend over a trillion, even trillions, are still not profitable.

4:11And they don't have, at least for OpenAI plans to be profitable for several years, not closer to the end of the decade does OpenAI have plans to be cash flow positive, as we've reported. So that's the crux of the concern is sort of these companies, while they're very fast growing, their user growth, their revenue growth is strong, but what if they are sort of not able to pay for all of these commitments that they've made to building out that AI infrastructure in time as their revenue tries to match cost? Now, Bloomberg published a recent article on circular deals in the AI industry, Shireen, and it included a chart that readers and commentators were sharing pretty widely.

4:50What did that chart show? Yeah, so one, you know, to kind of double click on this problem of, well, how are these AI startups going to pay for this and how are the bigger tech companies that are also engaging going to pay for this? One particular aspect of the financing that's concerning to some is this idea of circular financing. And so the chart that we made at Bloomberg shows how many of the players that are investing in AI companies are also then getting paid by those AI companies for services. So for example, NVIDIA is a major chip supplier. NVIDIA is investing in OpenAI up to$100 billion, they've said.

5:32OpenAI is also a major customer of NVIDIA in leasing and buying chips. So that's just one example. There are many. XAI is also a buyer of NVIDIA chips, and NVIDIA is also an investor in XAI. There's even smaller startups that it trickles down to ones that maybe everyday people haven't heard of, but that's basically the idea is the person who's selling the goods also then making investments in a company that's buying those goods from them. Jason, how is that any different to how cross-company investments happen in other industries, like the auto industry or pharmaceuticals? I mean, the scale of this is really extraordinary.

6:14And the degree to which it is speculative gambling on a payoff in the future is extraordinary. That being said, I don't think it's necessarily dysfunctional that one part of the industry, which right now has really high positive cash flow and make something. And then there's another part of the industry which has negative cash flow, but can use that thing to possibly generate returns in the future that one of them would finance the other. And I think of that as like you're a gold miner, you're going off to find the gold and rather than selling the gold miner a shovel, maybe you give them the shovel in exchange for if they strike gold, you get a lot of money back.

6:59And if they don't, you don't get as much money back. So in some ways, it is actually sharing the risk for businesses that are more capable of bearing it. That being said, the scale of it is just vertiginous and dizzying. But I think possibly actually a good and functional thing ultimately. Well, Danny in our tax club writes, I live in Silicon Valley and see all the advertising for companies trying to sell their AI products. It's redundant and ridiculous. It's taking jobs away from humans for sure. Shereen, what arguments are tech CEOs giving when they're asked whether these investments in scaling AI technology represent a bubble that might not make long-term financial sense?

7:46Well, of course it's in their interest to say this, but the AI CEOs and executives all are adamant that we're actually potentially underestimating the long-term value of AI, that the rate of growth has been unlike anything they've ever seen, both technologically and in terms of revenue. It is true that ChatGPT broke records for being one of the fastest growing consumer software apps of all time. I think over close to, I think, 900 million users right now. So their argument is we're only at the beginning of this. And just you wait until these apps become more and more useful for big businesses as they're sort of fine-tuning or basically specializing the applications of the AI for different job-specific tasks.

8:30Now, obviously, that is the bullish argument for it, and there are others who think differently. Well, Sam Altman is the CEO of OpenAI. He said in August that while AI was, quote, the most important thing to happen in a very long time, he also thought investors were, and I'll quote him here again, over-excited about AI. So, Shereen, how does Altman hold those two statements as being true at the same time? Yeah, so I was at a media event with Sam Altman when I, you know, I asked him personally about this. I said, you know, how do you, do you think that we are in an AI bubble? And his answer, and I'm summarizing here, was more or less that, yeah, sure, you know, yes, we're in some, we're in a AI bubble, but it's not going to impact the companies like ours that are truly transformational, where it's really going to hit are these small companies that may have inflated valuations, these startups with just a handful of work of, you know, founders who are getting billions and billions without even a product with so, you know, so much as a pitch deck, something like that.

9:31So that's sort of his argument that if you look in the dot-com boom, that even companies like Amazon were able to weather the boom and bust cycle because they were truly transformational. I'm curious, Jason, how much of this relies on an assumption about human behavior, that there will be a demand for AI, that there won't be a backlash if people feel AI is replacing people in jobs. Is there a bit of a gamble in that space? I think there is a gamble, and none of us know the future. I am skeptical that there will be that much of a backlash. If this is a profitable thing for companies to deploy, they will deploy it.

10:12If consumers like using it, they will use it. And as Shireen said, OpenAI itself, ChatGPT, is now used by more than 10 % of the people on the planet Earth. And generative AI is probably used by the majority of people on the planet Earth, given that it is incorporated into Google searches. So I do expect this to be more and more used in the future. But that isn't enough to justify these valuations. They also have to be able to profit from it. And to profit from it, they have to be able to differentiate themselves. If Anthropic and ChatGPT, Claude and ChatGPT are both the same, you can't charge very much for either of them and you can't make much of a profit.

10:54So they need people to use them but also to have enough barriers and moats that they can lock users into their own product and charge them more. A member of the tax club writes this, I think we need to regulate the use of AI. I'm not sure how to do it, but I fear that'll create job losses and also it sets the potential for dangerous people to use it in dangerous ways. Coming up, why AI might avoid the fate of past financial bubbles and how spending on AI is impacting the U.S. economy. That's just ahead.

11:29We're discussing the boom in AI investments and whether that spending might turn into a financial bubble. Jason, I want you to explain the boom in stock values for us. Why has Wall Street's valuation of tech companies like NVIDIA, Meta, and Alphabet gone up so much recently? Yeah. I mean, it's really extraordinary. The measure that economists like to use for the market as a whole is what invented by an economist named Bob Schiller, who won the Nobel Prize, and it's called the cyclically adjusted price earnings ratio, the CAPE ratio. And it tells you how much you need to pay for each dollar of earnings.

12:05Right now, that CAPE ratio is at about 40. That is the second highest it's ever been in the day to go back to 1880. The first highest was right before the bubble burst in 2000 for the dot-com. So it is just extremely, extremely high. When you then unpeel it and look, why is the stock market as a whole at this near record high relative to earnings, it's almost entirely driven by these tech companies. And much of the valuation of these tech companies is based on an expectation that they're going to make breakthroughs and figure things out in the future, which is why you're paying so much for a relatively small amount of earnings, because you think there's going to be a lot of earnings in the future.

12:53Now, it varies a bit from company to company. You know, Apple isn't quite that way. It really is selling a lot of stuff today. And then the other extreme, OpenAI, which is a private company, is entirely valuation based on speculation about the future. So, Shireen, I mean, tell us more about how these companies say they will make back the money they're spending on these AI investments. Yeah, I think, you know, they've already proven that they can get a lot of consumers to use their apps. Now, it's about how do you monetize that usage? So, you know, one way to do it is to have subscriptions. That's what, you know, ChatGPT and Anthropics Cloud chatbot, that's their main way of making, you know, money right now through everyday users.

13:35There's also, though, this idea that businesses can start to pay more than maybe the average user would just for their recreational chatbot use if they're actually getting real business value out of it. So already, a really high percentage of Fortune 500 companies, I believe in the 90s, already use some form of generative AI product. But the question is how deeply are they really going to use it? How big of contracts are we going to see here? And are they going to double down on it and not just do a one-year sort of pilot, but actually make sure that everyone in their organization is using this tool in their day-to-day?

14:11And so that's sort of the deeper kind of enterprise value that these companies are hoping to increase in the years to come. What are the founders of OpenAI, Anthropic, other companies building these software tools promising AI will be able to do for their companies? So one big area we've already seen a lot of growth in is coding. They're already starting to automate a lot of the software writing process. and you're hearing new terms like vibe coding for when people just sort of, when coders just have the AI sort of as their co-pilot just doing the work for them and they're kind of spot checking it or guiding it.

14:51Two other big areas they're really trying to push on is healthcare and finance. Healthcare, there's obvious benefits there for doctors who are summarizing or even diagnosing, but also in pharmaceuticals, I think a big hope a lot of these companies have is that they can help find the next miracle drug, something like an Ozempic, right? And if AI can really help in that discovery process, that would be huge. And lastly, finance is also a very important sector. A lot of finance is already automated. There's a lot of tools, everything from spreadsheets to, you know, Bloomberg's terminal, right, are used by people in finance.

15:28But are there ways that AI can kind of take that even further. Jason, when you hear the case, the description of how these tools could be used in different sectors, does that financial case make sense to you? Oh, I have no doubt that OpenAI is going to be selling more subscriptions next year and even more subscriptions the year after that. And no doubt that these are going to be used more and more in businesses. To justify the valuations, though. It isn't just that they can double subscriptions every year for the next couple of years and continue on the same$20 a month-ish model. They really need, as Shireen said, something different, either businesses paying some huge amount for almost like virtual workers that don't tire, don't go to the bathroom, don't make mistakes, work 24 hours a day, but you still have to pay a lot for them to one of these companies.

16:29Or breaking into the hardware market. OpenAI has been working on a hardware product. They have Johnny Ives who did a lot of the development and design for Apple. And I think a billion and a half people have smartphones, something. Anyway, lots and lots of people in the world have smartphones. If OpenAI could make a product that was almost as ubiquitous as that and everyone basically needed to have one, that could also justify its valuations. But it does require not just growing along the vector they're on now, but something really new, transformative that they can also monetize and make money from.

17:07And when we talk about the investment side of this, data center construction is the most visible real-world footprint of the AI boom. Jason, how significant has that spending been for the economy? If you look in the first half of this year, 92 % of the increase in demand in our economy came from just two relatively small parts of the economy, information processing systems and software, both of which are a huge part of data centers. So economic growth is really being propped up by these. Just to be clear, if we didn't have them, we probably would have at least something else to partly make up for that.

17:47So my own guess is absent the data center boom, growth would be about half as fast as it is right now. And we'd have a little bit more of some other stuff we don't have, like maybe home building, but we'd also be missing out on quite a lot. But its growth has become just central to the economy. And it's also part of why, even as we've had things like the tariffs, which have subtracted from growth, this has added to the growth and maybe added even more growth than the tariffs have subtracted. We got this message from Mike in Tennessee who writes, The biggest beef I have with AI is the insane amount of energy that AI data centers consume to the point that these demands consume every watt of green energy that is brought online, preventing the retirement of fossil-fueled generation.

18:33Lots of communities around the country are seeing these buildouts of data centers that consume significant amounts of water and energy to run. Shireen, Jason pointed out how the buildout of these data centers are aiding U.S. economic growth. Market analysts like McKinsey estimate the demand for data center processing could triple by 2030. So to what extent do these estimates about the demand for AI undercut the argument that this spending is a bubble? Yeah, I think these companies are not just thinking about their usage now, but they're thinking about their usage two years, three years, five years ahead.

19:08So right now it is true that a lot of these companies say that they have more usage, more demand for usage than they can provide with their existing amount of energy and data centers and all that that they have. What is more uncertain is, is that demand going to keep continuing up in this crazy curve up, or could it ever weaken in the future? And, you know, they're sort of planning on an environment where the demand only goes up. And Jason, how are these companies grappling with concerns about energy consumption, about the environmental impact of these data centers? There have been some improvements in the efficiency of the algorithms.

19:49So you get more per chip, but they still want way, way, way more chips. The second thing is they've been trying to bring more renewable energy into the mix. But as your caller, the thing you read out from your listener said, a bunch of that renewable is renewables that would have been used for something else. And so now we're keeping fossil fuels online for others. Then there's the last hope, which is that one day these lead to some breakthrough, some new type of material, some greater energy efficiency, something that helps us more cost-effectively tackle climate change and makes up for all of it.

20:28But like everything else we've been talking about, that too is speculative. Jason, take a step back for us. How is the U.S. economy faring this year outside of the growth in AI investment? I don't know. And I don't know because we haven't gotten any data for almost a month and a half now. Because of the shutdown. Because of the shutdown. So we're really flying blind. There are private data sources, but they're pretty imperfect compared to the government. Back when we did get data, it also was a little bit confusing because we were getting mixed signals with GDP growth looking like it was very strong, but job growth stalling out and inflation remaining high.

21:11So, you know, maybe the economy was strong, maybe it was weak, but whatever it is, the labor market does seem to be facing a lot of challenges right now. And so now that we're close, at least, to the end of the government shutdown, what are some of the key indicators you'll be watching once we start to get more data from the federal government? I mean, the unemployment rate to me is the most important benchmark for the economy. That's what you really want an economy to deliver on for people is jobs and a low unemployment rate. And it's also the best measure of how sustainable things are. But I'll also be looking at the pace of job growth, be very interested in what happened to GDP.

21:52Was it as strong growth in the third quarter as we thought? But all of these measures, it's not just that we're not getting the data. They're also going to be affected by the shutdown itself affected the economy, certainly during the month of October, November. I hope we're going to bounce back from that and it won't be a lasting consequence. But it's just been a lot of choppiness this year. Well, a major player in the AI boom is Jensen Wong. He's the CEO of NVIDIA. That's an American company. It makes many of the chips that power AI computing. Here's what he told Bloomberg in October. I don't believe we're in an AI bubble.

22:25And the reason for that is we're going through a natural transition from an old computing model based on general purpose computing to accelerated computing. We also know that AI has now become good enough because of reasoning capability, research capabilities, its ability to think. It's now generating tokens and now generating intelligence that's worth paying for. Okay, Shafari, first just decode what Wong said there for us a little bit. Yeah, a lot of jargon. But what he's essentially saying is that there was a breakthrough in AI technological progress about a year ago, a little more than that maybe, when we started to see the advent of reasoning models.

23:09And what that means is that if you give the AI more time to quote unquote think, that then the AI can actually produce better answers. Now, that requires more actual compute, more energy, more usage of these chips that Jensen Wong's company sells. So what he's saying, I think, is that actually as the AI can give you more valuable answers with more computing power, then the demand for services my company provides is only going to go up. That's sort of how I interpret it. Kitty Mills, all I see of AI is Google and ChatGPT searches. and those are sometimes easily certifiable as wrong. The waste of resources, including electricity, is incredible.

23:52And another member of the tax club says, they're building a data center near me. People in my area are not happy about it. But those who are in favor are talking about the jobs. My fear is that their bubble will burst, people will be without jobs, and we will be left with an environmental nightmare. So to come, what the next few years could look like in the AI industry, and who's most at risk if the AI boom goes bust. That's just ahead.

24:21We got this message from another member of the Tech's Club. It sure feels like the roaring 20s. I'm Gen X and worried about retirement. I'm scared to invest everything because time is not on my side if there's a crash. What am I to do? Other generations don't have the same urgency because they could still recover before they're ready to tap a 401k, but not our generation. And Gary in North Carolina emails, I've sold all my AI stocks and EFTs that hold them. It's a Ponzi scheme. Shireen, I can't help but feel sometimes that AI feels unescapable. Even if you don't use ChatGPT or Claude, Google, for instance, still generates AI results.

25:01Even when you didn't really ask for it, it's there. And so when we're talking about the case for AI, how much are these tech companies building the demand case on what they're offering and then saying like, oh, yeah, people are using AI on Google because they're just giving it to us, not because we necessarily asked for it. Does that make sense? Yeah, it's a very good question, especially for a company like Google that already has such a big sort of capture of our attention, right, through Google search. So they can very easily kind of funnel users into using AI. I think that's a harder argument to make when it comes to something like ChatGPT, where you have to actually go download a new app and use it, you know, in parallel or instead of Google.

25:50So Google was actually kind of following ChatGPT's lead in starting to integrate more generative AI into their responses as they saw that companies, you know, that smaller startup really take off. I do think, though, that there is this, you know, we do see this kind of backlash among many people to certain elements of AI, especially to this idea. I think one of the users who called and mentioned it, AI Slop, you know, that this material that AI may be outputting is of worse quality. And, you know, especially when it comes to something like images, right? You should take some considerable effort to make a, you know, appealing video or take a photo or whatever.

26:32Now with AI, you can just sort of do that in an instant. But is the quality then reduced? Is it as artistic or as meaningful as something, a video that someone filmed themselves or a photo that someone edited themselves or whatever? So I think there is an interesting debate about that and sort of the value of AI-produced content over human-produced content and when we feel that human-made materials are superior. Jason, when we think about the most recent recessions in the U.S., there's the brief one in 2020 during the COVID pandemic. There's the 2008 financial crisis that followed the collapse of the housing bubble.

27:06You talked about that. If there's a possibility that an AI spending bubble could lead to a recession, what kinds of scenarios about the scale of that recession seem most likely, given what we know right now? Yeah, so two of the last three recessions were caused by bubbles bursting. When the dot-com burst in 2000, we went into recession, and that was a relatively shallow recession. When the housing bubble burst, we went into a very, very deep recession, often called the Great Recession. This, to me, has all the hallmarks, if it bursts, of being more like the dot-com shallow recession than the housing bubble deep one.