215. How Near Is An AI Crash?

0:04This episode is brought to you by Google Chrome. You think you know a browser, but Gemini and Chrome? That's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration block, or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses set up required compatibility and availability varies 18+. When you need to build up your team to handle the growing chaos at work, use Indeed Sponsored Jobs. It gives your job post the boost it needs to be seen and helps reach people with the right skills, certifications and more.

0:41Spend less time searching and more time actually interviewing candidates who check all your boxes. Listeners of this show will get a$75 sponsored job credit at Indeed.com slash podcast. That's Indeed.com slash podcast. Terms and conditions apply. Need a hiring hero? This is a job for Indeed Sponsored Jobs.

1:07Hello and welcome to The Rest is Money with me, Robert Peston. Steph McGovern still on a short break, which is very annoying. But I am nonetheless delighted to be joined by my friend Azima Zar, founder of Exponential View, and one of the more thoughtful commentators, thinkers about all things technological. And I think we're going to concentrate today on where we are in respect of a couple of the great debates around artificial intelligence, the industrial and social revolution that we're living through. One of them is whether or not the amount of money that is flowing into to just whether it's AI companies or businesses associated with AI, supportive of AI, whether that's too much, whether we're seeing a bubble, whether people are going to lose a ton of money, whether stock market valuations around this place are too high.

2:04So that's one big and very topical issue. And then the second thing we're going to look at is where we are in terms of the latest evidence about the impact of artificial intelligence on the jobs market. There has been some evidence or some data in the US, which seems to suggest that we're already seeing a particular cohort of workers struggling to get employment as a result of AI. But it is ambiguous data. So again, we want to get into what we think is the most intelligent interpretation of the statistical evidence. So Azeem, I want to dig into what are the signs we should be looking for. I thought I might just, though, begin with a sort of personal story, which shows you quite how hard it is to know when a bubble is going to burst, even when you have clarity that it is a bubble.

3:07When I joined the BBC, literally within a few months, because I was sort of aware that there was some insane lending and investing going on in the city, particularly into these newfangled instruments, the collateral debt obligations and all sorts of debt products linked to the very toppy, very overvalued US housing market. in fact, housing markets all across the world. I went to see the editor of the BBC's News at 10, and I said, we've really got to do a big piece about this bubble and when it's going to burst. This would have been, I think, probably the summer of 2006, maybe early autumn 2006.

3:59And this guy said to me, when is this great disaster going to happen? And I said, well, look, the truth is, It could happen in a matter of weeks or it could happen in a matter of months. And his response was, it's a jolly good story. And it did the equivalent of sort of taking the notes, putting it in his top drawer and parking it. And of course, it didn't actually go pop until the summer of 2007, even though so many of the warning signs were already there, you know, months and months earlier. And so I think one of the things we've just got to be aware of is it is just when you get what you might call irrational exuberance, it's incredibly difficult to know when that exuberance turns to despair and capitulation.

4:45And certainly one of the things that does strike me is if I talk to investors, I mean, I'm sure it's sensible and intelligent for the Bank of England and the IMF to make a warning of this sort, but I don't, you know, I'm also being aware of some really quite heavyweight investment banks saying, don't be underweight. Well, you always want to be careful with market timing.

5:05Azeem Azhar:I too have my own housing bubble story, if I can share it. You were early, Robert. I mean, summer 2006, that's when Scion Capital and Paulson started to understand this was going to blow up and they went off and made money like Croesus by shorting those instruments. And I was lucky as well because I was able to participate in a bit of that. But the scale of the housing bubble was so much larger than what we're seeing today. We had$1.4 trillion of subprime U.S. mortgages, mortgages given to people who perhaps were earning$15 ,000 a year, mortgages of$700 ,000, subprime supporting about$7 to$8 trillion of derivatives over the top.

5:56Azeem Azhar:And the only real cash flows coming into the system were from those subprime homeowners. At that scale and that scale of leverage is when that bubble finally popped. And I think one question we can ask, and when I put together my methodical, historical, empirical framework on bubbles, was to look exactly at questions like this. So where are we? Where are the stresses? Where are the pressure points? And how close are we to each of them? So talk me through, because I know you have done some fundamental research on this, just talk me through your numbers. Yeah, absolutely. Well, we looked at every investment boom from the tulips through to today.

6:40Azeem Azhar:It's about 35 of the ones we looked at, which turned into productive booms, which turned into bubbles and how and why. Can I just ask one question? If you look at tulips, right, there was nothing productive about tulips, right? Well, some people say the same about AI. Well, let's come back to that, right? But, you know, if we look at the dot-com boom, even though it was a bubble and a lot of equity investors lost a lot of money when it was popped, it was undoubtedly, in my view, you know, it was productive, right? I mean, you know, an enormous amount of really useful investment took place, even if some of the evaluations were way over the top.

7:24So just because something is a bubble doesn't mean that ultimately it doesn't turn out to be good for humanity. Well, that's absolutely true.

7:31Azeem Azhar:And the railroads were the same and the telecom dark fibre from the 2000 was the same. And in fact, what we did was we narrowed it down to 22 productive technologies, starting with the canals in the 1790s and moving from there. And what we would want to understand is what are the best stressors that you could find? And we found five. The first was the economic strain. How much of GDP is going into this industry, this technology? And is it putting pressure on the economy? The second is industry strain. Can the industry afford to make these investments or is it borrowing on the never never? The third is how fast are sales growing?

8:17Azeem Azhar:Because if sales are growing quickly, the industry will be able to pay for this. And if they're paying for it with genuine customer money, that's a good thing. The fourth thing is what do valuations look like? So has the emotional exuberance of the market, of retail investors set expectations that can only be disappointed? And finally, and this is where the housing bubble scores extremely badly, what's the quality of the funding? Are we seeing honest capital going in or are we seeing financial shenanigans, hiding all the bad things, getting people to come in and invest here? So those were my five gauges.

8:55Azeem Azhar:You can think of them as instruments on a pilot's dashboard. Yeah. And they plainly are the right questions to ask. Let's just start with one that will immediately, I think, come to most people's mind. Are we seeing many AI businesses that basically have been given huge valuations, but have almost no customer revenue? Therefore, it's all based on hope. On the public markets, the companies that are on stock exchanges, they all have longstanding revenue. There are one or two exceptions that are new companies, but even their revenues run into the billions. And if you look at a company like Microsoft, which was around during the dot-com, at the peak of the dot-com bubble, Microsoft was valued at 60 times its earnings, a P of 60.

9:47Azeem Azhar:Today, it's a much bigger company. It's got much deeper relationships with its customers. It's valued at 38 times its earnings. Now, that's expensive. It's a bit choppy. You might decide that's not for you, but it's not 60 times. So we're still quite far below the absolute crazy heyday of the dotcom on that valuation metric. Now, on private companies, you're absolutely right that there is these companies being invested in by angel investors and venture capital investors. They're getting more and more expensive. And there are some astonishing deals. I believe that the press said that one of the OpenAI co-founders, Mira Murati, raised money at a$32 billion valuation for a company without a business plan that didn't yet have a product.

10:35Azeem Azhar:So there are stories like this, but these are in the context of hundreds of billions of dollars being invested more broadly and in the context of real companies like NVIDIA and Microsoft and Google that have great customer bases. I mean, if we look at what you might call, I don't know whether anybody says poster boy anymore, but anyway, what you might call the poster boy of the industry, OpenAI, which is still a private company, do we have enough transparency to see whether the kind of valuations that are being put on it in its various different fundraisings look reasonable compared with its revenues?

11:16Azeem Azhar:Well, I think it depends what reasonable means. At the height of the dot-com, Amazon, pardon me, looked incredibly unreasonable at the price you might pay for it. Its shares have gone up 40 or 50 times since then. So it's a perspective. But let's dig into OpenAI. So it recently closed some funding at a valuation of$500 billion. Its revenues at the end of this year will be$12,$13,$14 billion, which is not bad for a small company. Of course, it's still losing money because it's still investing in the future. So you then have to take a view. Do you think that paying 35 times its revenues is a deal or not?

11:57Azeem Azhar:And some people did. What can we argue for? Well, they're executing pretty well. The growth rates are phenomenal. We'll end the year with a billion people using chat GPT every week. And yet they need over$100 billion of funding to get to profitability by 2030. Where's that going to come from? They are coming to agreements with data center providers where they don't have the revenue to meet those commitments just yet. So, of course, it's a high octane situation. It's Icarus in a way. But we don't know whether they'll fly too close to the sun or not. So history has said sometimes people do, sometimes they don't.

12:39Azeem Azhar:But they do seem to have a real business that customers like. Yeah. And certainly, I mean, you know, this is not supposed to be an advert for chat GPT and open AI. But, you know, I have to say my customer experience of their products has always been, you know, I have been I have been burned by hallucinations by, you know, the chat bot lying to me. But mostly I'm wise enough to work out when that's when that's happening. I mean, I'm also slightly amused. You described it as a small company, a 500 billion valuation. But maybe that's the kind of extraordinary world we're in. There is the ugly side of this, which is the circularity that we're starting to see in some of these deals.

13:19Azeem Azhar:I mean, perhaps that is a problem. Are you thinking of the OpenAI deal with AMD then, for example? Well, there's the OpenAI deal with AMD. There's the OpenAI deal with NVIDIA. There's the OpenAI deal with Oracle, which has a deal with NVIDIA, which has a deal with AMD. The OpenAI deal with Microsoft, which has a deal with CoreWeave. and CoreWeave has a deal with NVIDIA. And just to be clear, so people understand what we're talking about, each time you do a deal of this sort, you are putting a sort of mutual valuation on each other and you're sort of propping each other up in that sense. Well, what's happening really is that NVIDIA makes the chips that all these systems need to run on.

13:58Azeem Azhar:And NVIDIA has tons of money and a really strong balance sheet. And their customers don't necessarily, and OpenAI certainly doesn't. So NVIDIA is leasing, in a sense, chips to these various companies, and it is sometimes taking an equity stake or a right to an equity stake. And so you're getting these interlinkages. And people like you and I who remember the telecoms bubble remember that the Nortels and the Lucents of this world engaged in something similar. Actually, it was much worse for technical accounting reasons than this. And it was one of the reasons the bubble burst so badly. So this circularity is something that is worth keeping an eye on and seeing whether it's really going to create strain.

14:43You know, you made your name with understanding exponential progress. Sometimes it's growth, but sometimes it's technological change. And this is sort of supranormal change or growth. Now, one of the reasons why the Bank of England, for example, fears this may turn out to be a bubble is because it says there is a risk that, for example, in the energy consumption of these services, that we may suddenly see essentially exponential savings. and the amount of energy that these services consume may suddenly start to shrink really very rapidly. And its concern is not necessarily that the providers of the services will turn out to be overvalued, but that quite a lot of the businesses that are getting high valuations as a result of, in a sense, building out the infrastructure for these services may turn out to be producing, you know, redundant infrastructure.

15:54So if it turned out that a data center could be run on, you know, essentially a half or a third or a tenth of the energy that we currently think they need, then if you're building nuclear power stations, or you're supplying all sorts of cables and all the rest of it, you'll suddenly find that the demand for your products and your services is going to fall off a cliff. Have you looked at the sort of energy side of this? And is the Bank of England right to, again, be cautious on that front?

16:27Azeem Azhar:I have no idea how we ended up in that position. It really astonished me. Energy is a lifeblood of our economies. It's a lifeblood of our health and well-being and we've known that for 10 000 years some point 30 years ago in the uk and western europe in the us we decided to use less and less energy and reduce our our consumption per capita and not all of that came from efficiency we actually in order to move into the 21st century we need a demand stimulus to our energy system we need more energy we need energy for carbon capture. We need energy to electrify transport. We need energy to electrify industry and to move it away from fossil fuels, which we will over time.

17:21Azeem Azhar:So this idea that we might produce too much energy at a moment where we're straining to serve this wide range of needs for a 21st century economy, I find absolutely farcical. I turn it around the other way, which is that it's fantastic that there are rich customers, the tech companies, willing to force custom to the energy system that is underinvested in it, that will drive innovation. And you know what? If they make their systems really efficient, all that's going to do is reduce the cost of electrifying trucks, buses and everything else. OK, well, I hope the bank is listening because what you just said seems highly rational to me.

17:58Now, look, before we move subject, could we just go back to your markers or your tests of whether there's a bubble? And could you just, I suppose, pick out one marker that you think should make us a bit more anxious, if there is one, that there's a bubble here, and one that would say, actually, no, there's some way to go to run on this investment cycle?

18:24Azeem Azhar:Well, it's actually they both point to the same marker, which is this is really all about customer revenue growth. And in the last couple of years, revenues have grown by more than 100 percent a year. If that slows down, then it's going to be like Wile E. Coyote running off that cliff. You know, they're going to wake up and realize no one's going to pay for this. It'll be the railroads in America in 1873 or the telecoms companies in 2000. that marker is the marker that will really be the first one to go i suspect and it's also the one that is healthiest right now because people like you people like me many of your listeners many of the companies they work for are really getting results from using chat gpt and they plan to invest more next year and i was in las vegas a few weeks ago and i asked 500 it directors by a show of hands?

19:17Azeem Azhar:How many people plan to spend more on generative AI next year? About a third of the hands went up. So that's the marker that I'm most closely trying to keep an eye on, figure out. But of course, Robert, as you know, as a seasoned journalist, bad behaviour is never discovered until it's discovered, whether it's Enron, whether it's WorldCom, whether it's any of these scandals. So, you know, I think the small option might be that there might be some bad behavior just hiding in that fifth gauge, which is about funding quality. Yeah. OK, that's incredibly interesting. I mean, I was talking to a number of very seasoned investors and they take the view pretty much in every industrial revolution.

20:01Some people have lost an awful lot of money, along with those who've made a colossal sum of money. There's always overvaluation. Sometimes it's at the macro level, sometimes it's more narrowly focused. But we will have disasters of various sorts. I suppose the final question I wanted to ask you is if you go back to the dot-com bursting, because most of the funding was equity funding and because it was outside of the lending system, outside of the banking system, what happened when the bubble was burst is that a lot of investors, some pension funds were hurt, some individual investors were hurt.

20:53But actually, it's remarkable how little the overall economy was knocked off track. I mean, typically, when you get that kind of a bubble burst, you get what economists call a bit of a wealth effect. So people are a bit poorer, and some of those people will consume a bit less. But as I say, it was not a massive, it really wasn't a massive macroeconomic shock. We didn't get a recession off the back of it. A complete contrast with 2007-8, where we started the bursting of the financial bubble, the global financial crisis, where the entire global banking system was quite close to collapse and banks struggled to lend.

21:39Now, that caused the deepest, widest recession since the 1930s, Great Depression. If we look at the funding here. There's not, as far as I can see, a lot of bank lending. So I don't think we have to worry about banks collapsing if we see some of these businesses going bust. But there is quite a lot of so-called private credit. There is quite a lot of lending going in. What would be your hunch if we did see some big businesses going bust, if we did indeed see a collapse in valuations? do you think we're guaranteed to be tipped into recession or do you think a bit like the dot-com bubble we'll get through it i mean it'll be painful but we'll get through it without in a sense a generalized decline in people's prosperity there is a bit of debt and jp morgan recently

22:36Azeem Azhar:pointed out that about 14 percent of investment great debt uh is now linked somehow to the ai build out, most of that is not AI companies. It's power companies and data center companies and people who build transformers. So these are really high grade industrials. And I think it's very far away from the contagion levels we saw during the global financial crisis. But I do think that there is a contagion risk because so much of the global stock market return has been dependent on the US stock market. And that's been dependent on this bet on AI, which is in some sense a bet on open AI. And that wealth effect is quite real.

23:18Azeem Azhar:And there are recessionary forces kicking around in the US. And so you might see a risk where if this did burst and equity valuations collapse really badly, but because AI isn't delivering, because something goes wrong, one of those big companies execution. I just wonder whether that might be significant enough, given that there are so many bearish forces surrounding us to have an effect in the real economy. For that, I would turn to an economist. That seems to me to be an appropriate moment to take a quick break. I want to, after the break, come and talk to you a bit about the mounting evidence about the impact of artificial intelligence on jobs.

24:05See you in a minute.

24:12This episode is brought to you by Google. Now, if you ran a small business, what would you do with an extra three weeks a year? That's how much time the average small business could save with AI, according to Google's latest AI Works report. Three weeks isn't spare time. It's the ability to get the day-to-day done faster, giving you time to plan for the future. So why aren't more people using AI? Google's research found it's mainly a lack of training and confidence. That's why they launched AI Works, a people-first training pilot that helps small business workers understand how they can use AI tools like Google Gemini to do more.

24:53In a matter of hours, people were using AI as a super assistant, instantly unlocking productivity gains. They were using it to write documents, crunch numbers, brainstorm and even role play new business ideas. If every SME in the UK gained those three weeks, that's a serious lift to job satisfaction, innovation and national productivity. Just one example of how Google's AI is fueling growth and transformation. Find out more at goob.go.com slash AIworks. So welcome back to The Rest is Money with me, Robert Perston. And I'm delighted Azeem Azhar is still with me. And Azeem, the other question that's been on my mind a lot recently is this slightly ambiguous data, particularly coming out of America, about what kind of impact artificial intelligence is having on jobs and wages.

25:46And in particular, there has been this debate around some data that shows that young entrants to the job market have seen a fall off in their employment. There's actually been a decline in American businesses employing younger people. But weirdly, those sort of 30, 40, 50-ish, we've seen a rise in that cohort being employed. Now, some, like an economist who's made a bit of his name in this area, Eric Bernolfsen, have said that this does show that generative AI is already beginning to replace people in the jobs market. Now, there are others, and I was struck by a very interesting bit of analysis by a man called Joshua Gans, and he's looked at the data, and he thinks that what we're actually seeing is artificial intelligence enhancing the productivity of those with more skills.

27:01and that is the explanation for why we're seeing more older people being employed, but some pressure when it comes to or declining demand for younger people. What's your take on all of this? Because it is an absolutely, I mean, I personally think it's an absolutely fascinating question.

27:22Azeem Azhar:It's a fascinating question. I'm going to give you a third point of view to Eric and Joshua. were. We should also acknowledge that outside of graduate hiring, young people hiring, non-graduate hiring is worse. So if you have a degree, you're finding it tougher. If you don't have a degree, you're finding it even tougher. So this is something wider than white-collar AI disruption. Here's what I think is happening on the white-collar side. And I'm being very cynical here. Companies overhired and the labour markets were really tight post-COVID and we've now moved into this rather difficult economy.

28:04Azeem Azhar:Recession fears, inflation fears, political uncertainty and companies need to manage costs in that situation. It's much easier for a boss to do a non-hire of a graduate than to fire loyal employees. And if you're doing that, you want a good news story to tell your board and your shareholders. And that good news story can be, we've used a magic technology, we call it AI, to improve our efficiencies and our productivity. We're innovative. And that's why we're hiring less graduates, when in reality, we're just doing basic cost management. And my sense is it's still a bit early for this technology to really have changed the way firms hire in any way that shows up in the numbers.