This is an audio transcript of the Unhedged podcast episode: ‘AI peak is peak AI’

Robert Armstrong The vocabulary word of the day today is ouroboros. For those of you who don’t know that word, it is a snake that eats its own tail, and it is a powerful symbol both of life and of death.

[MUSIC PLAYING]

Today on the show, we will be discussing whether AI is the synergistic, parabolic, high-growth, high-return investment of a lifetime, or a monster in the process of consuming itself and taking all of us with it.

This is Unhedged, the markets and finance podcast from the Financial Times and Pushkin. I am Rob Armstrong, coming to you from a refreshingly cool New York City, where I am joined by the editor of the Lex column, John Foley. John, welcome back to the show.

**John Foley **Thanks, Rob. It’s a delight to come here and discuss ourobori with you. (Laughter)

**Robert Armstrong **(Laughter) In the last couple of weeks, the kind of diagram that is dominating Wall Street research reports is what you might call a spaghetti diagram of the relationships between the companies that make up the AI industry, so OpenAI, Google, Microsoft, Nvidia, AMD, CoreWeave, Oracle, Intel, the US government, there’s like different kinds of lines connecting all these things that represent kind of customer relationships and investments and take or pay contracts, etc. Can you just give us a little flavour for the ouroborish . . .

**John Foley **Ouroboreality.

**Robert Armstrong **The ouroboreality of this? (Laughter)

**John Foley **Yeah. So all of these companies that you’ve mentioned all do different things that are necessary for the arrival of AI and the mass adoption of AI, from make . . . Nvidia makes chips. AMD also makes chips. CoreWeave puts the chips into data centres. Microsoft also is using data centres to create cloud products for its customers, etc.

But what’s happening at the moment is that all of them are kind of finding different ways to work together and invest in each other to bring this new AI reality to fruition. But it’s very confusing and it ranges from direct investments. So Nvidia, for example, investing in OpenAI and saying it would invest more, OpenAI investing in chipmaker AMD. Some of these are just kind of commercial collaborations. So for example, like you can now use OpenAI’s models on Google’s cloud even though Google also has its own models and is a competitor to OpenAI.

So everyone’s just kind of trying to work out like who can do something for me right now? It doesn’t matter whether they’re a competitor or whether they’re literally doing the same thing or doing something different. If they’re a supplier, they’re a customer, it’s a free-for-all.

**Robert Armstrong **And strategically, what do we make of all this? Let’s look at it from the point of view of Sam Altman, the head of OpenAI. Strategically, what is the point of all of this basically handing back and forth of quite large cheques?

**John Foley **So Sam Altman has a problem and an opportunity. The problem that he has is that he wants to build an enormous number of data centres that process an incredibly large amount of data. He’s looking at spending about $1tn, we’ve worked out at the FT, on data centres. So that’s the problem, is he’s gotta work out how to do that and how to fund it. He needs all the ingredients. He needs the chips. He needs the bricks and mortar. He needs the power, he needs the cooling systems.

The opportunity is that OpenAI is red hot. Everyone wants a piece of it. No one is not taking Sam Altman’s call at the moment. It looks like he can dictate terms. So he can call all these companies that are part of this ecosystem and say, I need you to commit to do this. I need you to commit to giving me chips. I need you to commit to build me data centres. Here is what I want in return. And they say yes.

**Robert Armstrong **Yeah. Buy some of our equity, or we’re gonna buy some of your equity, or we’re gonna have a contract where I commit to spend $50bn with you over X, Y, or Z period.

**John Foley **Absolutely. And he really is mixing and matching. So some of these deals are similar but different. So for example, Nvidia has promised to buy a stake in OpenAI, a large, you know, multibillion-dollar stake in OpenAI. AMD, which is a rival to Nvidia, is instead giving OpenAI a massive slab of stock — 10 per cent of the company, pretty much. And it’s not quite giving, but it’s basically giving it away for a pittance.

So in each case OpenAI gets a promise that it will receive loads of chips, and it also gets something else. Nvidia’s putting in cash, AMD’s putting in stock. They’re sort of paying tribute to Altman and saying thanks for being a customer who’s gonna buy a shed load of our chips and in the process is gonna keep our own stock market valuations high because we share in that glow that surrounds OpenAI right now.

**Robert Armstrong **Let me articulate an anxiety to you. Picture the spaghetti diagram in your mind, with the names of the companies we’ve mentioned here are all circles and there’s lots of different kinds of lines connecting them. The anxiety is that there isn’t enough money coming from outside of the diagram, from outside of the diagram’s perimeter to keep the whole show going.

Like obviously a lot of investment capital has gone into these companies and their stocks are worth a lot of money, but with any industry, there has to be cash flow coming from the outside world out of customers’ pockets rather than investors’ pockets. And the worry is there’s not enough of that in this ecosystem and that the kind of passing back and forth of big cheques is what you do in the absence of outside cash flow from customers.

**John Foley **That is a very real concern. And I would split what’s going on in AI at the moment into two camps, ’cause you have companies that have a business model of sorts already built around AI, or they are starting to. Google is a great example, right? Google is building data centres but it also effectively rents out its data centre space, its cloud computing platforms to companies and people who pay for it.

Facebook Meta doesn’t do that, but it is using its AI to make its Facebook advertising engines and Instagram and so on better. And it’s actually getting real money for that. We can see that that’s creating more revenue because we’re spending, sadly, we’re spending measurably more time on Instagram and clicking on ads more often.

But ultimately, as you say, like someone has to pay for all of this. OpenAI in particular needs to find customers who will pay for its products because all of those data centres, that trillion dollars of data centres is all supposed to be dedicated to OpenAI’s products, which currently make some pitiful, I mean, in real human terms, it’s a large number, but in relative terms, like $13bn of revenue maybe, which is really not very much.

**Robert Armstrong **It is slightly reminiscent on the non-revenue side of the early days of the internet where all these internet publishers were making money amorphously from eyeballs. And it took some time for those eyeballs to appear, and appear as money, I should say, as revenue. But for some it did and some it did not, you know? So you’ve reassured me on some of my anxiety, but I have more.

**John Foley **I didn’t mean to.

**Robert Armstrong **(Laughter) Well, you did. You are not sufficiently cynical, John. Bear that in mind. I want to raise another anxiety, which was brought up to me by Todd Castagno, and I’m sorry if I’m mispronouncing his name, but he is a valuation accounting and tax strategist at Morgan Stanley. And in a note called “Mapping Circularity”, he talks about how there is a worry that for investors in these complicated interrelationships between companies, transparency is lost. It’s hard to assess the quality of the revenues, or what is coming from where. Do you think that’s an issue here?

**John Foley **It not really is an issue. I think at the moment it’s a relatively small issue, but it’s certainly the case that if you are a portfolio manager and you bought stock in Nvidia, well, like after the deal that AMD announced with OpenAI, you will own Nvidia, which owns a stake in OpenAI, which owns a stake in AMD. So you suddenly, like . . . (Robert laughs)

Also, like Nvidia bought a stake in Intel. You maybe didn’t want to own Intel, but now if you’re a shareholder in Nvidia, you kind of do. So there is a bit of murkiness around what do you actually own. And is it helpful? How do you diversify when everyone is buying shares in each other?

It’s also just not really the done thing, right? We like to see companies invest their capital in things that further their business, not just in stakes in each other. It’s not necessarily a great use of shareholders’ money.

**Robert Armstrong **Well, you bring us kind of neatly to the next topic that I wanted to discuss, which is sort of historical analogies for what we are seeing now, and the first one that jumps to mind is corporate Japan. Or corporate Japan as it was classically, kind of in the ’80s, ’90s, which was Japan Inc was one big company. The banks owned bits of the operating companies and operating companies owned bits of each other, and carmakers would own carmakers, and it was all very insulated. Not particularly shareholder-friendly in a lot of ways, not particularly accountable to the outside world in general. And that’s changing now, and people are very excited about that. Do we have, just as we had Japan Inc, do we have AI Inc?

**John Foley **There are a lot of parallels between those two, right? And it’s not just Japan. Like Germany did this too. China, when I was based in China, we saw this sort of interlocking shareholdings. There are definitely things about that that are not so great. And it is true that we do seem to be moving in that direction. I guess what I would say, I’m trying to be a bit open-minded about it ’cause like it’s very easy to see a pattern and kind of take that too far and assume it will end in the same way.

I think that one of the things that was seen to be good about the Japanese keiretsu model and actually was copied by Chrysler, was an American company that deliberately tried to copy this aspect of it, was that it got suppliers and customers to work together in a more collaborative way instead of just engaging in a sort of destructive tug of war where my gain is your loss and vice versa. And if you’re trying to create complicated products with complicated parts in them, it can be helpful to see each other as partners to some extent. I’m not saying you should buy shares in each other.

**Robert Armstrong **Yeah. And Chrysler of course learned the hard way, I think, or at least this is what they teach you in business school, is that Chrysler learned the hard way from watching other Detroit carmakers who were too nasty to their suppliers. And then when a down cycle came, the suppliers didn’t survive. And so suddenly you’re the carmaker and you can’t get a steering wheel. And it’s a little bit of a problem.

**John Foley **And by the mid-90s, Chrysler was, I mean, then things went a bit wrong after the Daimler merger, but it was at one point the most profitable US carmaker by quite a long way. And that may be one reason why.

So I don’t . . . It’s clearly a concern when companies start teaming up in this way ’cause we can’t, as you say, we can’t exactly see why they’re doing it. Who’s calling the shots? A particularly worrying thing is like how is the influence being exerted? If we can’t see that from the outside, is it now the case that you can call up your customer and just tell them what to do and they’ll do it because, you know, you’re a shareholder as well. That lack of visibility is concerning, but some kind of collaboration may not be the worst thing.

**Robert Armstrong **Yes. One idea that I’ve heard kicked around is that how appropriate a kind of corporate cross-holding model is depends what stage of life your economy and your capital markets are in. So, I mean, this is often brought up about Korean companies. There is a term for these . . .

**John Foley **Chaebol.

**Robert Armstrong **The chaebol, exactly. Is that these grew up at a time in postwar Korea when the economy was still developing quickly but had a lot of kind of weak spots in it and the capital markets were shallow and unsteady. And having a chaebol model where the different parts support each other, that’s what you need in the absence of a fuller, richer, more mature economic and market ecosystem.

**John Foley **I think these arguments are all good. But I think the thing that we have to come back to is that we’re also in the middle of a massive bubble here, which is the difference, right?

**Robert Armstrong **Yes. Yeah. That was my point, right, is that we seem to be in the opposite position than like 1960s and ’70s Korea was in. There’s gobs of financial capital around just there for the taking. We have a very robust legal and economic infrastructure in the United States. It doesn’t seem like the kind of time that we would need the mutual support networks that you’re discussing, unless, of course the project of building all these data centres is so expensive, such a truly massive project of historically unprecedented scale.

**John Foley **If the people who are doing this believe that the financial markets, red-blooded and sharp-elbowed, would not understand the long-term returns on this and therefore would not put money in, so you have to just short-circuit them and strike these kind of backroom deals.

**Robert Armstrong **And you do hear quotes from big shots at these companies that are like this, that are kind of like financial returns be damned. There was one from one of the Google guys I think. I can’t remember.

**John Foley **Yeah, it’s not just Google, but like Google is one that said like, and Zuckerberg said it at Facebook too like, we may, like, effectively we may torch tens of billions, but you know, it’s worth giving it a try. It’s better than not competing and being left behind.

**Robert Armstrong **It’s not like you’re developing a new model of car. It’s like the space race. Or it’s like the race to build a transcontinental railway. Like there’s the people with the vision. And as you put it very nicely, now they just . . . you wanna short-circuit the bean counters in the process of doing that.

Should we talk a little bit about whether we believe the end customer is gonna show up for that? That is the bet the vision people are making, that if you build, call it what you wanna call it, super intelligence or AGI, artificial general intelligence, or whatever, the revenue will come.

**John Foley **Maybe. So the classic example of revenue not coming is Microsoft letting people pirate its software in China on the understanding that once they were using it, they’d be like, I love this. I’ll pay for it. And they just never did. The other example of how people do sometimes pay for stuff and pay more for it is something like Netflix, which has managed to push through price increases against all odds, and people pay it. And subscribers don’t unhook.

So OpenAI and its peers are clearly hoping that they’re more Netflix than Microsoft in China, and that people will . . . You know, ChatGPT is basically . . . I mean, it’s not really free. You can pay for it, but the people will increasingly see financial value in its products and pay high prices for them. Maybe they will. I guess we’ll have to see what those products are.

Already though, companies are paying, a lot of companies and businesses are paying, enterprises are paying for AI services, paying for like AI-enabled software. That’s real money. But then their customers need to pay them to make that worthwhile.

**Robert Armstrong **I mean, a great free model, right, is Google. Google figured out that if you make the internet accessible to people and you become the little boat people float around on the, you know, on the search engine going to the bits of the internet they like, the customer won’t pay for that. The advertiser will pay, right? And I wonder if in the back of their minds, AI people are thinking what Google showed is if you make a computing product that customers love, you can always just sell ads against it.

**John Foley **I think that’s at the front of their minds actually. I think that is literally what is gonna happen. Everything will have ads.

**Robert Armstrong **Yeah, everything will have ads. And so even if the average Joe doesn’t pay for his chat bot, the ads will do that for him.

OK. You said what is gonna happen. Let’s talk, as a way to close this discussion, about what is gonna happen. Where are we five years from now, 10 years from now? Where does this land? What has the spider in the middle of the web Sam Altman woven?

**John Foley **There are actually . . . There are two spiders in the middle of two overlapping webs here. One, Jensen Huang at Nvidia and the other is Sam Altman at OpenAI. And I think they’re both kind of doing a similar thing at the same time, which is to make themselves completely indispensable, to make their companies too big to fail. And so what I think is quite clever about this interlocking relationship with all of these suppliers and customers and partners is that everyone increasingly has an interest in seeing this project succeed.

So a few years from now, if OpenAI is struggling to raise the trillion dollars that it needs to spend on data centres, there are now a lot of people who will want to help it because either those data centres are gonna contain their chips, so their revenue is at stake, or they need the products that OpenAI build, or they’re just mutual shareholders.

I think that the risk for everyone and the opportunity, again, for Altman is that he’s creating something that will . . . a tide that will like raise all boats because everyone is interlinked. But the problem is that means just as everyone rises together, everyone potentially falls together, and if one link in the chain starts to get weak, everyone has to put in more money to keep it going.

[MUSIC PLAYING]

**Robert Armstrong **Listeners, we’ll be right back with Long and Short.

[MUSIC PLAYING]

Listeners, welcome back. This is Long and Short, that portion of the show where we go short things we don’t like and long things we like. John, what do you have for us today?

**John Foley **I’m gonna go long steak, Rob.

**Robert Armstrong **Steak.

**John Foley **Steak.

**Robert Armstrong **E-A-K.

**John Foley **The meat.

**Robert Armstrong **OK, got it.

**John Foley **Because cattle prices have been rising dramatically and we had this summer the lowest slaughtering of the herd in the US in about a decade.

**Robert Armstrong **So steak is a momentum trade.

**John Foley **Yeah. And it’s a commodity that responds very slowly to price signals because cow gestation periods are quite similar to human gestation periods so it takes a while. Feed is very . . . there’s been droughts, grass hard to come by, so stockpile steak.

**Robert Armstrong **Well, this is, yeah, this is terrible news. I get my steak at Costco and I buy ’em in like five big steaks. I got ’em. I’m gonna go buy a bunch.

**John Foley **I go to Whole Foods.

**Robert Armstrong **Yeah. I buy a big package. You’re a better class of person, basically. You go to Whole Foods . . .

**John Foley **Grass-fed.

**Robert Armstrong **I’m in Costco. They’re not far from each other in Brooklyn.

OK. I will take a long, and while we’re on the commodities theme I guess I’ll be long cocoa prices. We had a huge cocoa price bubble and it has popped, and I don’t know anything about speculating commodities, but I believe in chocolate and I believe in the human desire for chocolate. And in these trying times, I am betting that cocoa is a commodity you can’t keep down because we all need a little treat sometimes.

[MUSIC PLAYING]

Listeners, write us at unhedged@ft.com and tell us which commodities you are speculating in, and we will be back in your feed next week. Until then, stay sharp out there.

Unhedged is produced by Jake Harper and edited by Bryant Urstadt. Our executive producer is Jacob Goldstein. Topher Forhecz is the FT’s acting co-head of audio. Special thanks to Laura Clarke, Alastair Mackie, Gretta Cohn and Natalie Sadler.

FT premium subscribers can get the Unhedged newsletter for free. A 30-day free trial is available to everyone else. Just go to ft.com/unhedgedoffer.

I’m Rob Armstrong. Thanks for listening.

[MUSIC PLAYING]