Mother of all Capex Cycles

Welcome back to the second part of the Capital Cycle Podcast, this is Edward Chancellor, and I've got with me Chris Wood, who's the global equity strategist for Jeffries. So welcome back Chris, and I hardly need to say that the world is AI obsessed. You pay close attention to AI in part because you're focused on Asia and Taiwan, and Korea have been the prime beneficiaries of the AI CapEx boom. You've written a lot about AI investment over the past year, and we've covered similar ground in many of our podcasts, largely from a capital cycle perspective. So when you wrote earlier in April this year that we were witnessing the mother of all CapEx cycles, this was music to my ears. In case any of our listeners have been living on another planet, Mars, perhaps, can you explain what you mean by the mother of all CapEx cycles? So back in about the end of 22, you know, the tech set, the desperately needed a new story, the tech propagandist, the best they could come up with is we've got another PC, a place in the cycle, another, so they were desperate for a new story. A fair amount at the time, the magnificent seven worked down more than twice. They worked down, and I was saying fangs of peeters were sending to S&P, and I was looking right. I have to be clear, I've got absolutely no interest in using AI models, zero. And I remember actually I was in Miami, and I suddenly heard from these tech guys about this, and I realized this is going to be the next big thing. They became clear to me very quickly, if you're macro, you needed to understand what was going on. I missed him for three weeks, it cost me some opportunity to cross, but a former colleague who used to cover samples, and he explained to me in simple English why NVIDIA was in this unique position, because NVIDIA's got both hardware and software, so I put seven percent of them in my global equity portfolio, but then I was always already hearing from so-called tech specialists that NVIDIA had already gone up too much, and you have to buy AMD. People can't stand buying a rising chart. And this guy said to me, this is going to be the biggest memory cycle ever, and you've got to go in two stocks, NVIDIA and Hynix. And the reason why this cycle took off is because these big tech companies who were looking, as you said, have a slightly precarious position, 22. Suddenly this new thing came along, and you knew they faced a prisoner's dilemma, because this was the next big thing, they had to chase it. And that was a bigger motivation, the fear of being disrupted for their legacy businesses by this new technology. I should say that prisoner's dilemma framework is key to capital cycle analysis, that you have two companies deciding whether to invest and not invest. It would be optimal if neither of them invested, but it's sub-optimal for one to pull back from investing to find the other ones taking their place. And fun enough, that prisoner's dilemma analysis was actually put out by one of the Sequoia capital, people in a note, a couple of years ago. And then earlier this year, Sequoia invested in open AI at a very late stage investment. Interesting. I think the prisoner's dilemma, exactly the right way to look at it. But the other thing is, the great thing about these big tech companies from 2008 to 2022 when it began to go wrong was they were each operating in their own, what was fashionably called "mote" and they generated this amazing free cash flow. We have to admit, certainly, it's probably the best business model ever. But what they've done is, in the last few years, they've been exiting their modes, they've all converged on the same space with the exception of Apple. They're draining their modes because they're spending their cash. Google, I shouldn't say Google, it's not called alphabet, just had its first negative free cash flow event. And it's just one big colossal punt. That's the way to look at it. I think in your presentation you've been showing to clients this week, I think you're pointing to forecast trillion dollars of CAPEX and hyperscalers next day. That's what they're guiding for. And actually, I've seen some phones come at the even higher. So the numbers are absolutely enormous, there's no other way. As you say, the CAPEX is exceeding operating cash flow, which is another key for the other measure. The key point is why the stock market has been so resilient to the higher bond yields, the geopolitics, et cetera, is that the EPS market wide looks very strong. That's because this CAPEX cycle is incredibly earnings accretive, probably the most earnings accretive in EPS term CAPEX cycle we've ever seen. The semi-conductor companies will be a micron in the year, some of the video, be it Korean and Taiwan guys, they're booking their profits up front. Whereas the hyperscalers are in no hurry to fully pay for or count for their CAPEX, so the depreciation bills will be delayed as far as possible. Meanwhile, the details of these data center financings, they don't own these data centers. There's got these complicated SPVs. Yes, which is always a dodgy sign, and they're all of balance sheet, apparently. And then you've got the circular arrangements where they are lending money to these two big large language models and Thropic and OpenAI to buy their compute. Envid in the middle doing circular financing to the front end models, but also doing it with the so-called neo-clouds. neo-clouds. Some of which are pretty dodgy former crypto miners. Yes, so neo-cloud is basically borrow money to lend out chips. I think so, there's sort of the digital equivalent of we work. But the bottom line is that I think the prisoner's dilemma is the way to look at this. My base case until proven wrong is that this will prove to be the opposite of the internet economy, because the internet economy, think about it, was winner takes all, incredible network effects, on which the two best examples in the Western world were Google and Facebook. Had this incredible share of advertising, whereas I think this is going to be much more equivalent to commoditization, more equivalent to the airline industry. I see that point. The other big picture point is let's say they build all these data centers. What are they doing that for? To turn themselves into data factories basically. But what's so great about being a data factory, I'm not a tech expert, it's more equivalent to running a toll road. Do I want to bet my whole balance sheet on becoming a toll booth authorator? But also, it's a toll booth authorator where I've got incredible concentration in two tenants, open AI and anthropic. Another point is will people want to stay on the cloud with AI? That's one they want to get. You're pointing to the fact that again, I'm not pretending to be an expert. I mean, other people have been saying this too, that it's not clear that the large language model is way for the small language model that can be run on desktop computers may be the way forward. We simply don't know because the technology is evolving so fast and you say there's a security advantage. First of all, if you're a corporate, you want your own proprietary database. So first of all, banks aren't on the cloud because obviously data security is all important for a bank. So they've never been on the cloud. But if data security breaches are much more likely or much more at risk in AI world, which I'm assuming they are because it's cyber security stocks, you just read the newspapers every day. Obviously, it would seem to me obvious that every company in an ideal world would want to be on what they call on-prem, not on cloud. And the end reason they're not on the cloud is because it's too expensive, but this technology works like every other technology will become cheaper and cheaper and cheaper. So therefore, if we're going to get a move away from the cloud in the big picture. And on the small language models, there's a very interesting Stanford study, which said that you can do a lot with small language models, use much less energy, and you can do it on a PC, a Genetic AI you can't for now. But so let's say that's true, it's going to be all going to PCs, then actually that would mean you're better off buying Dell or Lenovo in Asia. You have also been saying quite consistently at the end of the day, you think the Chinese are going to win the AI race. Well on the mass market, definitely, if they're allowed to, because the AI evangelists in the US, which are most important, brother guys in who left open AI and went to Anthropic, they have had this missionary faith, religious faith in reaching what they call AGI. Yes. So I came to the inclusion about a year ago, talking to people and then much more about this, but it's also common sense, actually, that you'll never going to reach AGI by scaling up large language models, because these things are basically glorified pattern matches. I'm not saying there are no value. Stochastic parrots. Yeah. They can sum up a legal contract very well, allow you to ask more intelligent questions of the lawyer, or if you want to find out the average wage costs or do preaching costs of every listed Japanese car maker, that will help you, right? You don't have to pull out every, I'm not saying it's not, but it's not the holy grail of human intelligence. I think you're right. That is the sort of core delusion at the heart of Silicon Valley. So go back to the Chinese, you're saying the Chinese. To the Chinese, yes. The Chinese, they don't care about AGI. No, they don't kill themselves for AGI. Well, I put it very well to Beijing about 18 months ago, he said, listen, if it's eight out of ten, it's good enough for us. We're all about coming up with commercially viable, practical applications for this technology, you need to choose. things and you need compute and energy. I will admit the Americans ahead of the Chinese in the compute. - So the Chinese are catching up. - They're catching up very fast. I will back the Chinese to catch up quicker than most people say. The Biden administration made a fundamental error in stopping the Chinese buying the advanced US chips because the Chinese were not expecting that. They were genuinely shocked. So what was already a priority in China became a national emergency. So that was a bad decision from the US standpoint. Plus, it deprived US tech companies of which the video is the most important from the biggest customer. And now when the Chinese finally used the leverage on rare us, they could have done that any time they did it last year, then the Americans belatedly realized that the Chinese had the leverage in this area. Then the Americans on the Trump administration tried to do the deal saying, will relax the restrictions on the semiconductors if you relax the restriction on the rare us. But it's clear that Z decided that his hell bent on being independent in the technology. So the Chinese said, no, we don't want your chips. Obviously, the Chinese private sector were loved at the US chips, but the former line out of Beijing is we're developing our own chips. - Let's talk about the semiconductors. You pointed out that you don't think the economics of the data centers are robust. Do you question whether these frontier models developing in the mad pursuit of artificial general intelligence will get there? - But you're still, as the Americans say, constructive about the Asian picks and shalls. - Well, I think the easy money's been made on this. That's for clear in the first three years. So there's no doubt about that. But if you're a portfolio manager, these stocks are generating huge profits. They're also, if you're a global emerging market manager, they're 30% of your index. - And this is SK high nix, Samsung and TSMC. - So to own none of these things would be a colossal bet. And they are making huge amounts of money. So the way to look at it, it's not a valuation bubble. It could be construed as an earnings bubble. - What I call a fundamental bubble, when the bubble is in your earnings, and adding to your book, then it's in the fundaments rather than in the valuation. - Right, but you see, they had a big sell off in July. So they've had a kind of healthy correction. So it wouldn't, if there's AI, - But that was just the South Koreans ants running for the. - Yeah, but it's filled the gap in the chart. Right now, there's no sign of AI-capic slowing. But I think a key thing in the next three months is does this anthropic IPO happen? In a world where it happens, this thing will continue. It wouldn't surprise me if these Korean semiconductors retest their highs on the summer. - Are they still in your model portfolio? - Well, yeah, but I'm not the benchmark waiting. So I'm underweight. What's changed this year is a no longer spending cash to borrowing money. So I think the most likely end game for this AI-capics is when the market takes away the credit card. Either market says I no longer want to fund these enormous amounts of spending. And then all these stocks will go down sharply. - And you pointed out in your pieces what you want to look for is the underperformance of the large hyperscalers. - And the credit spreads, I think. - And the credit spreads. - No. You see, they're now, we're now got the bond yields rising. The cost of capital is rising. Partly because the hyperscalers are borrowing money. We're really in a government bond market sell off. So I think the real sense of indicator is going to be the credit spreads. - And so I think that the investment grade spreads that the hyperscalers now have paid much more than the best banking system. - Yeah, but they got tons of money so they can continue to do that. It's going to be the guys at the margin. But also I think we need to keep an eye on data center securitization. - In that sense, it reminds you slightly, if you remember, of the sort of rollout of the alternative telecoms companies in 2000. They were really like Ponzi finance schemes that as long as the valuations went up and capital flowed in, they could carry on playing the game. - Yeah. - And then when the valuation stopped going up, debt spread blew out, then the game was up. - But the problem is this cycle, but there's one big one, fatical problem. If you look at the high yield bond market spreads, we're going to rise. But actually, this cycle, the high yield bond market, in my view, has got much healthier bonds, dodgy debt in this cycle is not in the high yield bond market. The dodgy debt in this cycle is not going to do with AI, but private credit. - Yeah. - That's where the dodgy loans are. No one in private credit is incentivized to own up to any problem. The AI connection with private credit, private equity, is that in the last five years when all this money poured into private equity, what was the sector they did the leverage buyouts in software? - Software. - Now, from a very simple standpoint, when LBO started in the 80s, you did leverage buyouts on companies with physical assets. The idea of doing a leverage buyout for software company would have been viewed as insane. And it is insane. These people were making assumptions of stable revenue retention. But obviously, software is the area of most vulnerable to AI. So, very clever gentleman from Silicon Valley is in about 2011 that software will eat the world. And so, the question right now is, will AI eat software? - The sea. - Now, the best reason to think that Microsoft's not going to be blown up by AI is not because Microsoft was a fantastic program. It's just because of the sheer inertia and incompetency, right? Of course, it's having to move. But there's plenty of software companies who are clearly vulnerable and to do a leverage buyouts, it's just obviously a point of vulnerability. - I see that the private credit problems and the massive over expansion of private equity are during the ultra low rates. It's something lingering in the background why this AI-- - No, no, no, no, no, this is the real bubble in the cycle. And AI is a capex thing. It will become a bubble the longer it goes. We've only this year gone into being debt financed. - And your notes say, you're only going to call it a bubble when it's debt financed, which is what you think. - Which was only a private equity, private credit is clearly the bubble in the cycle. And it will turn out to be the biggest mess allocation of capital literally in the history of a US fund management. You've got endowment, very distinguished endowment. Not in the UK, in the US, and more than 50% of their assets. - It's what's called everyone wants to look like Yale. They all follow David Swenson at Yale. - But it is private equity and alternative assets. - It is frankly insane. - And actually, we have competent capital. - So you have what you know all about. But the by-point is the LBOs on the software is where the two are connected. And you're getting increasingly get private credit funding these data center securitizations. - So last word, you believe that the US stock market as a share of the global market peaked in December 2025. - Christmas Eve 24. - Christmas Eve to be precise. - Christmas Eve, like the Japanese market peaking. - Yeah, bit over 67%. But we have to-- - And now down to what 63 and a half a second. - But it doesn't mean just say it has to collapse. But I mean, it's 67%. It's an enormous-- - That's where Japan was the share of the MSCI. - It was never got that big. - Okay. - This is enormous. And China and India, when I last checked a 3% each. So this obviously makes no sense. - You think we are, even though the bubble in inverted commas hasn't burst. You think we're now on a long, almost secular move of the revaluation, particularly of emerging markets relative to the US. - Yeah, but right now, the most dramatic evidence revaluation is not in the equities, it's in fixed income. We're emerging market local currency government bonds dramatically outperformed G7 government bonds since March 2020, which is where Burma 40-year bull market and bonds ended because the G7 central banks led by the Fed went bonkers when they printed all that money to deal with the pandemic. Since then, there's been massive outperformance of local currency emerging market government bonds in US dollar terms. - Very last word, you're not even a closet gold bug. You're pretty open to gold bug. I think that after the financial crisis when the quantitative easing and so forth was starting, I think you were saying, one day we'll be at $10,000 nice. - No, I'll be able to go since 2002, 'cause my view has always been that the US dollar paper standard is on borrowed time. I mean, what's amazing, frankly, is how long it's lasted because Richard Nixon wrote the last discipline, right? - Dollars of gold in '72. - Yeah. - And the dollar standard has remained so long. But now we're now moving, 'cause we're already in actually a de facto gold standard because central banks globally at current prices own more gold than the only US treasury bonds. - Do you say it's only 25% of their FX reserves are in? - Yeah, but they own more than treasury bonds. Central banks outside G7 have just been consistently buying more gold. That really kicked off an early '22. At a time when gold should have gone down because the Fed was tightening. So the only plausible explanation for this was that it coincided with the freezing of Russia's foreign exchange reserves. So clearly central banks, governments globally were shocked by that. - So Chris, what is your, I know your medium term forecast. Well, give me a 10-year forecast for gold. - Well, unless Kevin Walsh gets religious stores and takes it to the US dollar, otherwise goes inevitably going to 10,000 higher. But I think the real trigger will be if they fix the yields in the US. - Which you think they will do at some point? - My guess has been that they'll do it between five and six and the 10-year, but that's just a guess. - So we're not very far away. - No, but we've already had some evidence last quarter of a desire to manipulate yields, right? - It's got better. - Yeah. - So is he drew a line of the sand at 4.75? I think right now, there were 60 basis points higher. So right now, he's looking a bit like King Canute. - Yeah, well, if you remember, King Canute. - Of course I remember. - King, anyhow, so thank you. That's been, I've enjoyed our conversation and thank you. - Thank you. (upbeat music) - Thank you for your time today. I hope you will listen to the next edition of the Capital Cycle. This communication is provided for information purposes only. Please refer to Marathon's website and the global investment reviews for further information, including important disclosures.