Regulatory Risk is Coming For AI | David Woo on AI Data Center CapEx and Iran War

Today's episode is brought to you by the Tukurium corn fund ticker, C-O-R-N. Let's get into it. Very pleased today to be joined by David Wu, independent economist known for his game theory based forecast on global markets and geopolitics. David, welcome to monetary matters. Thanks for having me. David, what do you think about AI, the huge capex in data centers that's now global and just the AI data center trade? For me, the big takeaway from the first quarter was the fact, by the way, the combined capex, okay, of the five hybrid scalers that is Microsoft, Google, Amazon, you know, whatever, Oracle and Facebook were actually down for the first time in three years, actually on a quarter on quarter comparison. In fact, even on a year on your comparison, growth rate dropped relative to the fourth quarter last year. And then not to mention the fact that obviously the first quarter of these hyper-skillers were paying more for whatever they were buying from Micron, Samsung and practically everything else. Okay, so as a result, if they were paying more for the same stuff that they were buying and they're spending actually less than they did in Q4, what this means is that actually capex in real terms was probably down quite a bit. Okay, now it's interesting that why people got so excited because of all say, the earnings growth is going through the roof up 25%. But let me tell you why, imagine Microsoft, all Microsoft did was to pay a hundred dollars more per basically membership to Micron. Now Microsoft only expense, let's just say 10% of that because they only expense the capex, you know, essentially based on depreciation. So I just say they have to expense 10 dollars of the hundred dollar increase in Micron chip. Okay, Micron because this is just literally like free money because they just literally they don't even have to produce more, they're just charging more. That is 100% basically margin business. Okay, so they can basically book an increase in profit by hundred dollars. So 100 minus 10, that is combined earnings go up by 90 dollars just because Microsoft is paying a hundred dollars more to Micron. Okay, for the same chip, what I'm saying to you is that ironically the inflation of the components that go into data centers have actually given ironically a boost to basically nominal earnings growth because equity investors are dumb. They only look at nominal and rather than real for them. Wow, nominal looks great. But I think so from now putting me this is actually very important because so I'm telling you ironically I don't disagree, the notion that more capex to the extent that hyper-skillers think that there is more upside for them in terms of productivity growth in terms of whatever it is that's bullish for AI. But I would argue by the way, not only there was a slowdown but actually the slowdown in real terms poverty was quite considerable. On top of that in the first quarter, I think the numbers were in general. Okay, since should the window address by token maxing as you know, right? We know all these software engineers were racking up massive consumption of AI because like for example, Amazon, Amazon at a certain point were ranking their AI whatever the software engineers based on who's using more AI. Okay, so everybody was trying to max out, you know, saying should the AI usage? As we saw you saw the search in AI usage. This is why you know what you probably heard this, you know the CEO of Uber coming out a few weeks ago saying that they ran through their entire token budget for the year in just four months and have very little to show for and now they're running back. So what I'm telling you is that very likely we're probably going to see a slowdown in Q2. So Q1 numbers were very impressive. No doubt about that. Okay, and I think it's for the wrong reason. I think Q2, you're probably going to see a slowdown. So you said that there was a slowdown in actual cat X and Q1? Yes. Quarter of a quarter. Wow. No one talked about that. In US dollar terms, there was a quarter and quarter drop. And in year on year, basically growth rate decline in Q1. Wow. This is even before we're taking account of the inflation, probably we saw in Q1, which it would driven down essentially the real increase, the volume increase in the actual cat X. So what explains that quarter over quarter decrease in cat X from the fourth quarter of last year to the first quarter of this year and how come no one else is talking about this? I think people just don't want to get a number of. I mean, the main reason is Microsoft actually Google actually increased their their cat X a little bit. Microsoft and Facebook basically pull back a bit, which actually makes a lot of sense to me. Actually, because Microsoft, we know that they have some issues, you know, because in the past, they thought that they had this exclusive deal with charge GPT with open AI, right? And then they lost that. And then they've been trying to they are trying to and then they saw also the adoption, corporate adoption of essentially of Microsoft co-pilot offering has been much basically worse than expected. This is the reason. So I think from that point of you, I mean, I think Microsoft, probably has always been a bit more conservative. I think to the extent that they're not like for Google, I think for Facebook, they've always felt that God, if I don't invest, I'm going to be left behind. Therefore, I'm just going to keep chasing this, whatever it takes, and so on and so forth. I don't think Microsoft basically Microsoft historic has always been more conservative. But I think this is the reason why we saw a bit of pullback. And I think from that point of view, like again, in the case of Google and Amazon, the funny thing was that yeah, so Amazon and Google, they're like the capex basically picked up again in Q1. But then again, both companies and basically book a massive evaluation of their holdings of Enthropy. To offset that. Okay. And that of course, we know it's quarterly like artificial kind of stuff, right? I mean, so I think from that point of view, I mean, the bottom line here is that if you actually look at capex relative to operating earnings, for example, income, I mean, there's no doubt. I mean, if you actually leave for the top for the five hyperskillers, the capex to operating income is now 135%. So, and I think this is the reason why suddenly Google is talking about like, you know, wants to raise $80 billion, $85 billion, equity going to the finance of the capex, Facebook is not far behind and so on so forth. I think, you know, we're now getting to that point that these companies, you know, I think, you know, in the past, when we say all while these companies have the pockets was because they were sitting on a lot of, they were very, very cash rich companies that were earning a lot of money, which they could basically immediately put to use by increasing their capex. Now, they're practically using all of the operating earnings towards capex, having to now raise money, okay, from additional share issuance or debt issuance in order to finance the capex, that I think it's going to get that is starting to obviously worry the market. My focus is on the intersections between economics politics, geopolitics and technology and I will say right now, I think, you know, the last six months, my biggest focus has been the war that is oil and AI, of course. So, like, you know, just like, you know, I'm bearish, you know, I'm, I've been bullish oil and bearish equities and bearish gold. So that's been sort of how I've been rowing for the last few months. And bearish gold as well. Bearish gold and so far this short term, there's no doubt. What do you think capex is going to be for this entire year and what did you think of the capex guidance that the companies gave in terms of how much money they expect to be spending over the course of this fiscal year? I think many companies did increase their their capex guidance. Again, there's a very big difference between increasing capex because you think that you're going to spend more to get more and having to spend more because everything you are looking to buy now costs more. Yes. Because most of these companies have actually, in fact, if anything, delay the time table in terms of when the completion of these data center construction is going to be actually done. So I think that makes a very big difference. And I think from the point of view, I think, you know, we, we, otherwise, it's just basically income transfer, is it income transfer for Microsoft to basically micro. I see we say, if we're a home builder and we say we're going to build 10 million homes and we're guiding that the capex to build those, sorry, we're going to build 10 million, 10 homes and the cap, we're guiding that that's going to cost us $5 million in capex to build those 10 homes. And then the next quarter we say we're raising our capex from $5 million to $7 million dollars, but we're still building 10 homes. Your point is we're not building more homes. It's just it's become more expensive to build homes. Exactly. Essentially, it becomes an income transfer from the hyperskillers to let you say, I don't know, like, you know, the memory producers, except in this case, accounting wise, for the hyperskillers, they only have to book a very small percentage of what they spend, okay, as actually cost, okay, in terms of their bottom line, we're asked for the memory makers, they get to basically book 100%. Okay. So from now point of view, again, it creates this optical illusion about earnings acceleration when essentially all that's happening. It's an income transfer, in fact, of the worst kind. Tell us, is this thing that you're describing where a company spends money to build out infrastructure on capex? And then it realizes that cost, not when it spends the money, but over the lifetime of the acidity. it takes a depreciation expense every year. The same is true for a railroad, building out a new track on the railroad, building out a new factory, for example, for another firm. What about this dynamic in the AI data center cat-backs right now is such that it is perhaps obscuring the true economic value. I don't wanna put words in your mouth or making it look better than it seemed. Or is this just, I mean, the same accounting rules apply with Elkhae-Vex? - No, because in, sure, it applies to everything, except in this case, the recipient companies have, I just told you, like if it just basically a pure inflation story, okay. Then essentially the recipient of this increase in price, essentially, you know, has a 100% profit margin. That's what we're talking about here. It's all questionable degree here. And I'm just saying in this particular case, the degree is such that it's pretty, you know, like it just exaggerates the actually earnings growth momentum in the understand. At an aggregate level for the entire index, that's what I'm saying to you. - Yes, I think you're absolutely right that the companies whose earnings are going up the most are not the companies spending on AI. They are the companies who are selling to the companies that are spending on AI. - Yeah, and by the way, I just wanna say, what we also need to understand is for Samsung, Samsung, high-nase, and microt. These are three companies that make, I've been with memory chips. I mean, let's be honest here. I mean, I've been with memory chips are not nearly as sophisticated as, let's just say, I don't know, an Nvidia Blackwell Rubin chip, right? So from now on, technically it's less complex. And too, there's a lot more competition. You know, like there's only one company who's basically making Rubin chip. So you can argue that Nvidia commends, you know, whatever massive margin because like they're the only ones who can actually design these chips and so on and so forth. You have three companies, microt and Samsung, and based on kinetics or competing with each other. And they're massively ramping up their cat-backs, okay? So you're talking about, you're not talking about, you're talking, I don't wanna say it's homogenous, a total commodity market. But let me just tell you this, memory market. These have always been more commoditized than any other market. Now, I can tell you, even the Chinese are gonna be ramping up. The Chinese are not moving into DRAM, you know, NAND Flash. They're very good at it. Even though they cannot, the Chinese are not, at any point that they're gonna start to be able to compete with Nvidia on Rubin chips and Blackwell chips. I guarantee you within two or three years, they're gonna be right there, I think, in terms of the Samsung. So therefore, I can't even, I struggle to see how far you can see the kind of the pricing power that these companies have right now. It's gonna be very short-lived. I mean, let's be honest. The only reason why prices going through the roof is because doing the pandemic, there was a surge in PC demand, buying phones, buying whatever to work from home. And then afterwards, after the pandemic, since should the demand for these things collapse. And this is the reason why these companies, and in the prices of memory chips collapse, this is why these companies stop basically investing capacity, because they were literally down on the hill. And this is the reason why when this AI boom suddenly came along, like we're now moving from learning chips to basically inference chips, all of a sudden demand for essentially memory has been driven through the roof. Again, memory makers are subject to this kind of boom bus cycles for a reason. And this has been almost typical. In fact, if you wanna talk about boom bus, it is probably more true about memory makers than any other industry I can think of. Okay, in the whole entire world. And that's a lot. So what I'm saying to you, and this is why we need to understand what is really going on here, because I mean, that's what it is. - You're absolutely right that chips have been a commodity business, and memory chips have been the absolutely most commodity-tized, most cyclical business of all time. What do you say to the bulls-on memory who say that this is a shortage, and all the chips that micron are making right now, as we speak, are sold out into 28, into 28 for forward sales, and that the experts say that the new capacity isn't gonna come online until the second half of 2027 or 2028. - Yeah, I think 2027, that's what it is. I mean, Mark is supposed to be forward looking, but once the capacity comes online, prices are gonna go down a lot, very quickly. But you gotta basically like the stock market in the day, you're not just thinking about the next month or the next quarter or whatever, you're trading multiples of basically the earnings of these companies. So you gotta basically think about how much these companies are gonna be earning two or three years down the road, and not just what it's basically doing right now, and not just basically what the growth rate is right now. And I think that's what it is. So from that point of view, like this is, no, I think that's what this is about, okay. So like, but I think it goes way beyond that, because I mean, obviously the issue with the memory, I don't even have a major issue with that. - Okay. - Let's just say that this is part of the boom buzz to market overreaction and so on and so forth. Mine basically bigger issue. So I'm just telling you, short term, I think you're gonna see, I think you're gonna see earnings deceleration, okay. In the second quarter, compared to the first quarter, because number one, cap X is actually slowing, okay. And two, that essentially that token maxing, I think probably would have led to a decline. I think in terms of the growth rate of revenue, especially for the likes of Anthropic, or even open AI, the big LLMs, okay. And therefore, the people with which they sell, they share revenue with. But longer term, my big issue with the whole entire AI thing is that AI has gotten too good for its own good. Now, what do I mean by that? Until three months ago, I think the big problem for the whole AI trade was the fact that the capabilities of the frontier models seem to have plateaued, okay. Then came clock mythos, right. I mean, which was a breakthrough model, okay. It's very good at doing lots of different things. Now, it is so good at doing so many different things that it also happens to be extremely good at identifying vulnerabilities in network. In other words, anybody getting called of clock mythos can literally like to break into a system at will. And then let's just understand the cyber attacks when it's with a roof starting last December. - Absolutely. - This is even before clock mythos, anybody been basically new to name clock mythos. This is a reason why it's so far only been distributed to 150 users. This is why all of a sudden it's inviting regulatory distributed and so on and so forth. Mike, and it was actually interesting because political.com, right now the article last week, suggesting that within the White House, the one person really wants to rain back AI, especially the clock mythos, those smallest with clock mythos like qualities, is actually Picasso. Because Picasso is worried that if this thing basically falls into China's hand, China's gonna use it immediately with the US. So my point is this, you're telling me, so clock mythos was released on, what, April 7th. Since then, NASDAQ has added $5 trillion to this market capitalization because people say clock mythos is amazing. It's a halfway there to AGR. But if only 150 users have access to clock mythos, how can anthropic even monetize clock mythos? Do you ask me, right? You mean, how are you gonna basically make money? Because the point here is, this is what I'm saying, the right now, the issue is no longer about the capabilities of the frontier models having plateau. It is now about the accessible capabilities of the frontier models having plateau. Because at this point, government, I have no doubt heading into the midterm election, AI's gonna become a huge political issue. Okay? I think Trump's decision, reluctant decision last week to sign off on this new regulatory approval requirement for AI, it's just basically a very tiny baby step in the direction that we're gonna go towards. Okay? Writing in general, that's my issue. And then you probably heard about this even from an anthropic last week, which is that the fact that it has recursive learning ability, that's probably going to make it even more difficult for proponents of AI to argue for why dissemination of the technology. So this is a big problem. - What did you make of the release of mythos? - Well, they're talking about many mythos, whatever it is with some capabilities, but not quite mythos and so on and so forth. That's just basically get you a little bit. I have no idea if it's a 10% better, 15% better, than opus, but whatever, 90% what I'm saying to you is that anthropic obviously going IPO, they're gonna want to basically get everybody to oversubscribe and that kind of thing. They're gonna tell whatever story they want right now. - Yeah. - And if I were in that, they're sure so probably do the same thing, but what I'm telling you, there's no doubt. What I'm telling you right now, I'm maybe the first one talking about the downside risk to AI now is regulatory risk, but there is no doubt in my mind that this, I think more and more people are gonna talk like me and think like me in the coming months because it's inevitable. - So the bear case for AI is not that it's not powerful, but that it's too powerful and it's going to be reigned in by government's worldwide, including the US government. - Absolutely. - Hope you're enjoying today's interview. This episode of "Modulatory Matters" is brought to you by the Tuchium Corn Fond. on ticker-c-o-r-n. If you follow the show, you know we spend a lot of time on macro themes like energy transitions, geopolitical risk, and global food security. Corn sits at the intersection of all three. Most people watching the straight of her moves are focused on oil. They should be looking at nitrogen. A third of the world's fertilizer trade passes through that choke point. When that corridor tightens, fertilizer prices react, and corn farmers feel it first. Corn is the heaviest nitrogen user in US agriculture, so rising input costs hit their margins quickly. The longer-term story could cut the other way, too. If farmers pull back on fertilizer application, yields may come down. If the economics get difficult enough, some may choose to abandon corn and plant something else entirely. Either scenario potentially tightens corn supply, which could be price-supportive. Bottom line, the pinch on the producer side could become the price story on the commodity side. Two-cremes corn ETF gives you exposure to corn prices. Ticker-C-O-R-N. Traded on NYSE ARCA. Access it through your brokerage account, no futures account required. TOOKREM also has a family of agricultural ETFs, including the two-cremes wheat fund, Ticker-W-E-A-T, the two-cremes soybean fund, Ticker-SO-Y-B, and the two-cremes sugar fund, Ticker-C-A-N-E. Head to to tocremem.com to learn more that's t-e-u-c-r-i-u-m.com. This material must be preceded or accompanied by your perspective. The perspective is available at tocremem.com/corn. Investors should carefully consider the investment objectives, risks, charges, and expenses of the fund before investing. The perspective contains this and other important information. Investing involves risk, including the possible loss of principle. Commodities and futures generally are volatile, and instruments whose underlying investments include commodities and futures are not suitable for all investors. Pass performance is not guarantee future results. Thanks for listening. Let's get back to today's interview. Tell me about what the Trump administration signed, and you said that this is only kind of a harbinger to come that what's coming is way bigger. Tell us what you mean. So basically, the previous version that Trump decided not to sign will require a 90 day cool off period. I mean, the sense that new large language models had to essentially go through government approval. For 90 days, now they brought it down to 30 days. [CRYING] OK. So I don't-- honestly, I don't know. It's going to be a big difference. I mean, I think if there's a problem, I don't think it's going to matter if there's going to be 30 days and 90 days. I'm just thinking about for a second. Last year, Germany alone said that cyber attacks cost the German economy 300 billion euros. Now, imagine what we're talking about here. What we're talking about basically, mythos. Mythos is very good at performing multi-stage task on an autonomous basis, which makes it a perfect cyber hacker machine, basically. And so from now point of view, this is what it is. Now, imagine if you're a Trump, you're so OK, fine, but who cares? Let's just give it to everybody. Let's just give it to the S&P 500 companies. But nobody else. Think about this. Trump's rise is all about-- well, Trump's for the little guys, for the small, medium-sized businesses. So what? You're going to give the best models, the most productive model, to only the big guys, but not the little guys. You can screw everybody else so that you can basically do favor for your friends. You think that is politically basically sellable to his base? Obviously not. That's what I'm saying to you that this is a problem. Because the first 50 users they gave out to, you can make a case for it. I mean, they're like the big banks and the big hyper-skillers, because they control the digital network. So you got to basically make sure that they're safe. Because if something happens to them, everybody else is going to be in trouble. But now imagine the next year, there, where are you going to give it to? Whoever you're going to give it to, you're going to basically smack off favoritism. This will be a political problem. Already, the big ones are getting bigger, the small ones getting smaller. And nothing exists. You want to sell the right-- whatever saying that there is no doubt my mind that this AI thing is quickly becoming a political habitat. And I think politically this is going to be a big problem. This is what I'm saying to you that this whole issue might come up as a major political issue ahead of the midterm election. I totally believe that it is. I mean, AI is already a political hop button issue. And I yield to you that it certainly would be probably a politically good thing to do in terms of popularity for President Trump to regulate AI. But does that make you think that that is what the administration is going to do? Because I might say, I don't think the Iran war is terribly popular here in the US. I think actually looking at the polling data, it's one of the least popular wars just looking at how many days we are into the conflict. But obviously, that's what President Trump did. So President Trump isn't afraid to kind of buck the trend. And he's not going to do something just because it's popular. Listen, there is nothing to do with it. It has to do with the fire. Again, as I said before, there are two issues here. One issue is that if you release-- let's put it this way-- if you release clogged mythist to 50 users, 100 users, you can still more or less sort of essentially control how is used, who's using it, and so on and so forth. If you give out to 500 or 1,000 people, you can no longer be sure who actually has it, and who's going to have access to it, and so on and so forth. So if you do this, and there is a major hacking event, what people are going to say that you lost it, like you basically allowed the genial of bottom. If you basically give out to 500 users and ends up, basically, the Chinese now have it. Yeah. Then what are you going to do? So what I'm saying is that there is no-- I mean, especially when you just think about this for a second, they're talking about now that mythist has recursive learning capability. In other words, mythist version 1 is going to be able to create mythist version 2 without any human intervention. It's just going to keep improving. So imagine anybody having access to it, what that can do, my friend. I think it's pretty obvious. So this is not about from being popular in that part. Listen, obviously, he's only interested in money. And there's our own money. Like the amount of money that his family, him, have already made over the last whatever year and it's something of his presidency, just amazing, actually. I've no doubt that he and his family, his friends, their cronies, all have vested interests in essentially all these space acts, and throughout big, child G, B, T, and so on and so forth. So obviously from Trump's standpoint, he wants to see his company-- in fact, this is the reason why I think he hasn't pulled a trigger on Iran because these companies they go public first. Why not get all these companies do IPO when the market is maniac and basically bubble and so on and so forth? But I do think, again, but if you are a medium term guy, you understand that there are some things that Trump cannot, basically, he cannot stop. And I think this is one of these issues. I think that he's going to be a problem, especially given the people who have major issues with AI right now, happen to be from his base. OK. And I think that's very important. So it's a matter of national security. There's going to be a crackdown on AI and the frontier models. What do you think that looks like? It's already happening. What I'm saying is, this is the reason why it's called mythos. Only has only been given to 150 users. And this thing has been out since what a program. So two months later, you know, listen, you don't just think about how ridiculous that sounds, right? I mean, like, like, Chajjee beat he 4.0, like three months later, there was Gemini 3.0, whatever. These models, every three months, there's like a new model coming up. Just imagine for two months, basically, and mythos has gone nowhere. He said, basically, now being used by 150 people. Just think about that for a second. So that's already, you know, the factor in effect, basically government control that essentially, that the market has not recognized. So what did you make of the supposed rise in annual recurring revenue of Anthropic from, you know, something like $9 billion in December to stated, like, 42 or $44 billion. Listen, as I said before, I think, obviously, you would expect you to grow, no doubt about that. But there are two things going on. First, I just told you about token maxing, right? Token maxing would have definitely, OK, boosted these numbers. And then you would have seen a slow down in a second quarter. I mean, again, I'm not talking about absolutely decline. I would think that the growth rate would have slowed in the first quarter because companies clearly were spending too much on these tokens, which is what they were paying to Anthropic. OK, that's the first thing. Secondly, the reason why Anthropic did especially well was because Cloud Code. OK, Cloud Code was released. And everybody decided, well, it's the best thing since whatever slide spread. I guess what, three, four months later, now Cloud Code has got serious competition. Microsoft just canceled practically most of their Cloud Code licenses, OK? Because they're trying to get their people to use co-pilot on GitHub. OK, now you've got Codex, which is very good. You've got the Chinese, basically. The coin is pretty good. There's a bunch of other ones. So there are at least five competitors. Cloud Code now. They're as good, maybe not good in the same thing, but they're good in different things and that kind of thing. So I think from now point of view, that's very important because this is about the fact that, again, we've seen this movie again and again, which is like, oh, wow, like a one company. It's like, well, Chagy PD first came out. They was off the game. [BLANK_AUDIO] "Well, this is gonna be a winner takes all." Because again, think about this. The stock market is thinking, "Well, who's gonna be the next Google, who's gonna be the next Microsoft?" Because these are the natural monopolies. These companies made a lot of money because they're the only one. They account for 90% of the market. Microsoft 90% of the, I mean essentially desktop, essentially operating system and Google basically the search thing, right? So people are pricing these companies as though they're gonna be all the next Microsoft and the next Google. And assuming that there is a winner take all outcome. And I'm just telling you so far what I've seen is that, oh wow, so Chachi BD was good for a while and then DeepSeek called out. And after that basically, overnight basically made a search and then now basically, Claw basically called out. Actually, Larry Ellison put it very well. Whatever a week ago when he said that large language models, they're all gonna get commoditized simply because they're all learning from the same public data. And that's it. So from now on people get excited about one model. And we've seen this movie again a few times now, actually just in the last three years, which is a model moves ahead. And next thing you know, the next three months or the others basically play catch off. So that's what it is. But while that one model is out that there's a big breakthrough, it gets obviously the revenue acceleration and everything else that goes with it. But the question now is, and I think from now on, if you I just told you already, Claw code, which is the main driver of the growth rate of the anthropic, I think that already, I don't want to say it's been commoditized, but there's come some serious competition. And Claw myth is, let's just say Claw myth is still very good. Right now only 150 people can use it. So that's the bottom line. - David, you've laid out at least three arguments for a bare case on AI and AI stocks that are very, all three are very, very interesting. Let's see, first that there could be a regulatory crackdown. Now, as you say, it's already here for mythos 'cause it hasn't been distributed. Second, as you just said, that AI is going to be a commodity. And the third, what you began with by saying that the inputs are being inflated in terms of price. And that is where the real earnings growth is. What's interesting to me, David, is that you so far haven't said, and a core AI bare argument that so many AI bears said, at least in the beginning, which is that AI is not going to transform society and the economy. It seems actually like you might be more of a believer in the power of AI than some people who are long stomach inductor stocks or stuff. Could you just comment your view on just how powerful AI is to be used in the workforce and in terms of increasing productivity and the like? - I think AI is very productive. I mean, listen, I use AI, as bearish as I am about AI, I cannot live without AI. I mean, in fact, I spend probably at least two hours a day. Okay, basically I'm various AI platformers, and so forth. I think let's put it this way. I think AI is going to struggle to replace, I would say, the top 10% of the people. Okay, because if you look at, I mean, you know, listen, I mean, you take me as an example. So I became, you know, before I retire from Wall Street, I was the global head of great currencies, Merchant Market Fixing, Cum Economics Research. I managed to team a 50 analysts. We were the number one ranked team on Wall Street for macro strategy, right? - Bank of America, yep. - A bank of America. And I can tell you, like, I think it would be difficult for AI to do that. (laughs) Okay, I mean, to the extent that if you want to be very, very good, there's no, I don't think AI can be ever as good as I am, because by definition, it's because by definition, because AI for what I do, what it does for the most part, it simply just distilled consensus. Now it goes out there. So you say, well, AI do this for me. It just basically goes out there and the look for information. I mean, the information out there is going to be written by somebody or this Roy Turz, Bloomberg, whatever it is, and comes back to you. It doesn't, it's not going to come up with the original solution. It's going to come back with you. It would come back with a suggestion that somebody else has already done. So from that point of view, if you're talking about, because every profession, I don't care what profession you're in, right? The most successful people, okay. In any profession, tend to be also the most creative. I mean, creative, in a sense, that being able to think originally, come up with something that hasn't been done before, okay? Whether it's a thought, or a product, and so on and so forth, AI's not very good at doing that. But AI can probably outperform 90% of the workers, because 90% of the workers, they're just basically copy and paste type of guys. Okay, there is not that much creative dimension in what they do. So I think AI essentially, of course, can potentially take these people's job. Now, of course, we also know that, like blue collar workers would be more difficult, right? I mean, you saw like Starbucks, just canceled their AI program for inventory management, because apparently AI was making two-minute sticks. In the problem, not even so much AI, the problem is the machines, the camera, and so on and so forth. I mean, one of the things that, you know, but I wanted to basically save this. I mean, this is actually very important. As good as AI is, it is also true if you look at Wable. I still remember, you look like a young guy. I remember, like, this was like, already in 2012, they were like predicting, Elon Musk was predicting that within three years, like we're gonna have like, robot taxis, autonomous driving, all that, went to 2026 already. I, OIC is that when Tesla's latest robot taxi went into whatever trial in Texas, you know, it was literally like creating traffic incidents, almost running to people, like even WAMO. (crying) I mean, you probably saw some of the footage lately. I mean, it's crazy. After all these years, you're still making this very basic mistake. So what I'm saying is for software engineers, AI is amazing because any job that's by definition digital that has, that can be very, very well defined in terms of the objective, AI can do extremely well because they copy and paste codes and then basically do this and that's it, okay. That, so there are certain things AI can do very well. So for software engineers, it's basically, now, for what I do already, like macronalysis, whatever, I think is good in terms of like assembling information. - Yeah. - It's very good at simply collecting information. But it's not very good at coming up with new ideas. It's certainly not very good at coming up with new investment ideas. It can only chase after whatever one of the people are doing, it's also for, and then for things that's complex as driving, it's actually shit at it, at least, and this is after already 15 years of trying, okay. So, I mean, and this is basically Google, which is the greatest search engine company and they still cannot master basically basically based autonomous driving. So what I'm saying to you, everything has to be taken with the green and salt. I do think however, that for many white collar jobs, the pencil pushers in an office, okay. Especially younger people who don't know anything anyway, okay. They're probably gonna be at risk, okay. But there's no doubt that I think people tend to exaggerate the devastation that AI is gonna bring about in terms of jobs, except for software engineers. I think 50% of software engineers gonna be gone within probably three or four years. That's probably not even exaggeration. But to what extent this is gonna be true for every other industry, it's not clear to me. Roughly how many jobs do you think are at risk from AI? And I guess the way, 'cause I know you're a economist, is like by how many basis points or percentage points, do you think the unemployment rate is going to go up because of AI? - I think listen, first of all, we have to realize, I mean, I think it's gonna depend country by country. The US right now, right. Labor force is no longer right. You no longer have, you have no more net immigration, birth rates collapsing. So it's literally like this is the situation. So again, this is the reason why there is right now, it's in the foreseeable future, there's probably gonna be a deficit in terms of labor force. And so for that point of view, if AI just takes some jobs from software engineers or else being equal, I think that's gonna be manageable. But I do think that what is interesting to me though, is that I think AI, what it really does, what I'm hoping what it does is this, right. It's like one of my clients was telling me this, right. I mean, this is like, so I run a global macro advisory business work, advise some of the biggest investors in the world on macro investing. Now, obviously there are some very big guys, right, in the industry, like the city, the world, like the Renaissance technology of the world and so on and so forth. If you look at these places, they have invested very aggressively over the last whatever, 20, 30, 40 years on infrastructure, software infrastructure, okay. Renaissance technology for example, they supposedly have thousands of software engineers working with PhDs and building models of that kind of thing. And so the advantage, the fact that they were able to make a lot of money has lots of to do with the fact that they have a lot of software engineers, they've got a lot of computers, they will crack, essentially, cranking numbers and able to trick very quickly. Now, as one of my clients basically said today, Okay. And this guy is also very successful. He's not a city.org Renaissance. He said, now, you know what? I just have to hire three people. I'm gonna be able to actually recreate everything in the science and city.org. Okay. So which is as long as we can now use AI to do this. So say you should be building infrastructure. It's actually gonna be very interesting because AI in that sense, I would argue, that it actually makes, it actually empowers small teams. And this is where it's gonna get very, very interesting. Because actually, this is where, like, I'm not even that negative in that respect. Because like my son-in-law is a software engineer. He was until recently working in AI for one of the biggest whatever tech companies in the world. Okay. Here in Israel. And he just decided to leave, okay, resigned to basically set up his own shop. Because he said that no time in his career, he can remember another time, that if you just have an idea, you can literally create it yourself with the help of AI, with claw code. You just have to have an idea. You don't need to, like, use to be a case if you wanna have a startup, if you wanna set up a startup, you need a lot of money because you need initial, you need at least 10 people to basically start writing codes. Listen to that. Now, you can literally just give the command to a few agents and they'll just basically crank it up for you. And that is actually very, very interesting. So I would say in that sense, like, even for me, I would say like right now, like, I've got a couple of guys working with me, but now I feel like, you know what? With AI, we could possibly take on some of the very big banks in terms of the quality of our research. Because I don't need that many people if you do it efficiently. So that's the beauty of this, which is that it empowers people who know how to use it. And actually, you know how to use it, that just basically means that, so does that mean that it's gonna cost job losses? Not necessarily, actually. It just means that there's gonna be a problem, it's gonna be injection into the entrepreneurial spirit, actually. - So who are the winners and who are the losers in this new world empowered by AI? Sounds like some entrepreneurs who can manage small teams, they are really empowered. But where is all this surplus that sounds like you believe that AI is gonna create? Where does it go and to whom is the surplus negative or is hurt by it? - Listen, at the end of the day, we all know. I mean, I think, you know, Elon Musk has made this point, I've made this point, a lot of people have made this point which is, I just think about Star Trek, right? If you think about, you know, if you think about Star Trek, what do people do in Star Trek? They're like exploring the universe. Like with what money? (laughing) Right? Because like, you know, if they want food, you say, wow, I want some food and then you basically create, what I'm saying is that what, what, I, you know, in the ultimate sense, like AI, if it's as good as it can be, it should create a world of surplus. So that we all be putting our thumb, watching that flex TV every single day and doing something that we enjoy, actually. And I happen to enjoy working, but it's a different story. But the point here is that that's what it is. So I think from that point of view, like, yeah, if you have a lot of robots, okay? You know, I could easily investigate your world and what you're gonna be a robot's working and human beings are just basically exploring the world. You know, we'll think about big thoughts. That would be the idea. I mean, I think human beings, I mean, I don't know, that's the world in which I want to live. But I can see that as just sort of the ultimate endgame. But, you know, but however, this is where it gets interesting because I get, at the same time, there's another scenario, which is that, and actually, what's his name? The Jack Clark, the co-founder of Anthropic, gave a speech at Oxford University last week. And he was talking about the fact that there is definitely scenarios out there in which the AI can bring about human extinction. And not, we're not even talking about 30 years down the world. And I can tell you more and more people are talking about, I mean, this is the reason why a bunch of AI companies last week, you know, essentially wrote a joint letter to, I don't know, the government saying that we need to control the supply of some kind of bio-weapon, whatever, you know, essentially the ingredients. Because with AI, somebody could literally unleash a max external extinction event for human beings. So, I mean, and then these models, you probably heard, we even in Anthropic, they admitted that mythos is less predictable than some of their less sophisticated models. So, who knows? You know, so I don't know which, you know, so if AI is just gonna be listening to the human command, I can even imagine one day, none of us will be working, and that would not be so bad. But there's also a scenario in which we're all gonna be working for AI. - Do you think that the CAPEX being spent on AI and data centers right now is a bubble? - I don't wanna say it's a bubble or not, because, you know, again, you know, you're gonna have to base, you have to, you have to imagine that for these companies, right? I mean, like, I can understand the technology at the rate at which it's been progressing, you know, you sort of don't wanna be left behind if you're one of these hyperscalers, right? I mean, because, I mean, it's one of these things that if you make a wrong move, you could become totally irrelevant. I'm the role, okay? Now, like Microsoft is finding out, right? I mean, so for example, like Gemini already, I mean, it's a Google having invested in Gemini where it's Microsoft now have to start from scratch to create their own LLN because they can no longer rely on chat, GPD, and so on and so forth. But my point here until now, but this CAPEX spending is predicating on the assumption that the capabilities of these frontier models will continue to improve at a very rapid rate. And my point to you is until three months ago, that capabilities, okay, the improvement, the incremental improvement of capabilities of frontier models will getting smaller and smaller. And then with math, math, obviously that change, but now the accessible capabilities are getting smaller and smaller. And that's the problem, okay? I think so from that point of view, given my view that actually the that the accessible capabilities will get smaller and smaller, I would argue that we will quickly get to a point that suddenly that the CAPEX growth rate is gonna be slowing and slowing quite a bit. - And do you think that not just the growth rate of CAPEX is going to go down, but the actual level of CAPEX is gonna go down as it did, talk about the dot-com boom and then the dot-com telecom bust in 2001, 2002, like spending actually went down and it went precipitously. So when we talk about a bubble popping, that's what it looks like. Do you see that or just the growth rate slowing? - I absolutely see that. I mean, that obviously would require something. I mean, just wanna say, imagine what happens in the recession because we know like Google, for example, 60% of Google's earnings come from advertising and avenue, right? Google and Facebook, most of their actually earnings come from, I'm saying, which is highly cyclical. And by the way, at this point, did you know what avenue is a revenue, I mean, add revenue is like 80% of the whole thing, 85% of the whole thing. So these companies have done very well by taking market share from TV, magazines and print media and so on and so forth. But now they've got 85% now. So I mean, there's gonna be a limited, a limited room in terms of market expansion. And at the same time, so if you go into a suddenly a recession and all of a sudden companies are not gonna be advertising as much, this is gonna be a big problem. I mean, Amazon, it's an assembler situation these days. So essentially, like, that's what this is about, actually, which is the fact that right now, we all know it's like, oh, right now, it's like everybody that, you know, that, because it comes down to just two companies, right? Open AI and Anthropic. They are the ones who are making the money, so to speak, right? Because I mean, when you're using AI, you're paying them. And they're paying the hyperskillers, as a share of their revenue for the tokens or whatever and whatnot, right? And the hyperskillers then buy chips from NVIDIA and so on and so forth. So right now, what it's going on is that everybody's giving money to the Anthropic and the charge the Open AI of the world to keep these companies afloat, okay? So that they will keep this whole thing going. I mean, this is why NVIDIA was only too happy to invest in Anthropic. And what I'm saying to you is that that's why this whole thing is sort of like, you know, it comes down to Anthropic and in Open AI. What kind of valuation they're going to be trading at? And I think this is where it gets very interesting. Because at a certain point, I understand, in a way, Silicon Valley, I think these people are pretty smart, right? So it's the Washington, I mean, they all want to, they realize, I mean, to them understand, AI is like the only game that the US has now relative to China, right? And therefore, he's trying to basically pump this up. He's trying to get all these companies to help each other, right? And so on and so forth. Trying to basically build this momentum so that the US has a total edge when it comes to this whole thing. But I think this is where it gets very interesting. Because again, that's what I've telling you before. I disnote now that because of the sanctions, okay? That has been plebbing against China in terms of like, the ability of China to secure the most advanced AI chips that China is now being led behind. There's no doubt about that. I mean, they only have the equivalent of the H200 chip. Now you got black, well, you got rule, but in the so-China, it's two generations behind. There's gonna be a problem. But if, as I said before, Karl Mithis turns out to be too good, and therefore there is a regulatory basically blowback, then this whole thing. That's why I'm telling you, to me, the biggest risk on the horizon for AI. I'm not saying it's gonna be tomorrow, but let's just say on a six to 12 month basis, there's no doubt my mind is regulatory risk. - And that you think is what stops this AI cycle? - All right, it could be very well-being. I mean, I think it could be very well-being. But again, it also has a lot to do with the fact that, again, the industry lends itself to commonitization. And it's not just the large language models. I mean, look at, even look at Nvidia, right? Nvidia is very good at learning chips, right? But inference chips, okay, are much less complex. This is why even Intel can compete in inference chips, AMD and Broadcom. This is why Broadcom, it has not yet revised up their guidance, because again, it has to do with competition. I mean, this is a wonderful thing about this, which is again, if you look at what happened, if you look at Microsoft, Google and Facebook, these are not the norm. I mean, technology has a tendency to actually promote competition and therefore, commonitization. And the market right now is pricing, as though every company is gonna be a winner, take off. It's gonna be a winner in this AI race. And that's just by definition impossible. - And so you think there'll be a time when compute prices and compute supply drastically outstrips demand. So the compute shortage will turn into a compute glut. - I think it's very, it suddenly is very, very likely. Absolutely. Because especially given the way the technology is changing, right, as we said before, because inference is much easier to do. I mean, again, you have to understand that this whole monopoly business, right? That yeah, if Nvidia is the only designer then Nvidia is gonna be the winner takes all. But already, this is what's going on is that, like once models have learned everything, (cow mooing) you don't need more learning chips. You need instead inference chips, like all these agent models, basically run in inference chips. And as a result, this is why you need more Intel CPUs, rather than GPUs, okay? And then you need more memory chips, right? Because it's like my RAM chips. So everything's changing as a result. But what I'm saying to you is that once you move away from GPUs, the other stuff, there's a lot of producers. Nanships, I mean, like the Chinese already are like, I mean, the Chinese is gonna be very big in NM flesh memories. For example, so you gotta take all that into account. And that's what it is. - What do you think about the NASDAQ index or the semiconductor index in NM? - I'm short. I mean, I've been short for a while, so it's been very painful. But thank God I made some money, basically from, from being a low oil, so I haven't done so badly this year. But that's what it is. I mean, listen, we had, you gotta respect momentum. And there's no doubt, like momentum was what carried the day for NansDAQ. I think, you know, but we had a bearish key reversal week last week. And I think, you know, I think most people who are, who've been buying into this whole AI of mainly return investors, you know, who are mainly technical traders, that's what they look at these days. And I think that's gonna be a very difficult, I think that that very strong and compelling reversal sign will make it very difficult for people to really trying to take it back to all time high. So we'll see how it goes. So I, but I do think that right now, like I think especially given, I think given the way the war's going, my view, I mean, I feel even more strongly about the way the war's going, which is that Iran is gonna make it very difficult for Trump to tackle. I mean, Trump wants to tackle, obviously, but Iran's gonna make it very difficult for Trump to tackle, which means that from the situation, that means that that's the reason why I'm in long oil. Okay. And I think if this war is not resolved soon, which I do not believe will be the case, and this could all end very badly. - So it sounds like you made money being long oil, early in the war, you know, price of oil shot up. Congratulations on that. I happen to be long oil. I'm actually down like the price of oil has gone down. Why do you think the price of oil is going down? As the war has continued, and the straight up permuse, officially, at least optically remains closed. - I count some of the biggest hedge funds in the world as my clients, and I can tell you, I cannot find one person who's in long oil right now. Because I think the day is hard, psychological. You just think about this for a second, right? You know it's gonna come down to Trump, right? I mean, and you think, I think the day it's about information to symmetry, right? Because Trump obviously knows much better about what he's gonna do than you. (crow cawing) Right? So if he tells you, oh, there's gonna be a deal in two, three days. Who are you to basically say that he's wrong, right? You have to catch him out. It's a very dangerous thing because he knows more than you do about what he's gonna do. And this is the reason why people find it difficult to actually question Trump, and that's what it is. And therefore, especially more recently, right? I mean, like it's like, well, everybody knows that inventories, oil inventories are crashing, right? And that unless there's gonna be a deal soon, like oil prices will go through the roof. But presumably Trump knows that too. So therefore, Trump has a very strong incentive to actually get a deal before oil price should up to 100 feet. Yeah, I mean, 'cause I'm sure all that oil executives are calling him like, well, president, like if the straight up who is still closed by the end of July, like oil price gonna be $150. So from now on, you can have to assume that Trump is under pressure. So, and this is the reason why it's a sort of like, it's one of these things that you've got a president who's pretty market savvy, he has tried to get Secretary of State markets savvy. And Trump is very good at, he understands the market very well. He knows the market struggles to fight him. And despite the fact that he's been telling us for two months every week that deals within reach in days and next week is also for the market still has not been a position to catch him. To basically, it's column bluff, basically. But I do think however, it's because market doesn't understand, because my view ultimately has to do with the fact that in fact, Trump by trying to talk up how close the deal and this and that, yeah, it might have pleased the market, but it also convinced the Iranians that Trump, okay, is desperate, that Trump needs to deal, which means that the Iranians, this is all his, all Trump's big talk about the deal is close, would only enter the harden the result of Iranians, which will probably, you know, basically get them to demand even tougher terms. And then I think there's a, there's a price at which it'll be even too high for Trump. And that's when it gets very interesting. Okay. Tell me what the Iranians are saying and tell me what, other than observing that, Trump might be desperate as you say that, that what they are thinking, what their calculus is like. Why? And what do you, how do you assess the odds that a deal that everyone accepts is going to be signed in the near term? Let's say the next month. - No, I'm saying that just imagine what this, right? So we don't know, you and I don't know how close or how far the two sides are. But presumably the Iranians know, but they should know, right? Now imagine, so at the start of, like we've been in the ceasefire mode for the last two months, and Trump has been telling us for two months now that the deal's next week, next day, whatever it is. Now, you could argue initially, I thought that this was a negotiating tactic, right? The Trump was trying to essentially use this strategy to put pressure on the Iranians. Okay. Now, obviously it hasn't been working. (laughs) So the Iranians now, let's just imagine they're very far apart. The Iranians are looking at this every day, say, "Well, listen, like we're still very far apart. Why is Trump going to keep saying that there's a deal going to be next week or tomorrow and three days?" Equal to me one thing, that Trump really means a deal. If Trump is signaling to us that he'll willing to make more concessions in order just so that we don't walk away from the negotiation, that's what, if you're a Iranian, that's exactly what you think. Okay, number one. Number two, let's look at the action that speaks louder than words. What happened on Sunday, I said in Israel, on Sunday morning, okay. Hinsbullah, fire, essentially a barrage of rockets into Israel. In clear violation of the ceasefire deal that had been over here last week. Israel didn't fire back, as it should, okay. And then the next day, as you probably know, on Monday, Iran basically fired missiles at Israel. Let me tell you this, this has never, ever happened. It's never, Hinsbullah in Israel had been fighting over the years for a long time. (laughs) It's never happened that Israel as they should be Israel and Fisbullah are fighting it out and the Iran basically fired missiles at Israel. What is that telling you? That would suggest a certain boldness on the part of Iran that has never been there before. And they've only gotten this bold because guess what? They think Trump is weak. They think that they're trying to actually exploit this situation to actually drive a wedge between the US and Israel. So what I'm telling you is that you think that Iran, you know, some Trump say, oh, well, they took down our helicopter and our however our our our our pilots are safe. Just think about this. When Iran shut down the helicopter, do you think the Iranians knew that the pilots are going to die? What do you think? They couldn't know. Yeah, yeah, yeah. Of course, when they fire at the helicopter, they couldn't know that they were not going to kill them. Absolutely. You're totally right. As a result, it tells you that when they fire at his helicopter, they were prepared. Okay, that they were going to die and that this was going to actually terminate negotiation between the US and Iran because had these pilots died, Trump would have had to basically walk away from negotiation. So when I'm just telling you again, you got to basically look at talk is cheap. Trump is talking the rain is like if you look at what they're doing as opposed to what they're saying, it was a just that actually they're pretty dug in. They're pretty confident that they're in a strong position. And that's what they're doing. So from now point of view, I think I think this is the reason why I think they have no incentive to give Trump an easy way out. In fact, they're going to make him pay through the nose, actually. If from a game theory perspective, that would make all the sense of the world. And then the Iranians invented chess. Let's not forget. So they yes, that's a great point. What what do you think the end game to all of this is? I don't I listen, I think the great thing about the great thing about markets is that. And I don't like this, I call myself a strategist, I suppose an economist, whatever, because these days, I don't make GDP. For cats, I can't care less. Okay. Like it's about whether you're long or short. I think to me, whatever the end game is, I mean, there could be different end games. It could be basically different paths, but what I carry is that most of these scenarios we're going to end up. Probably you're going to see high oil price and that's going enough for me. Okay. So I don't really need to basically get into this. Oh, well, what's going to be that the exact outcome because that's not how investing works. It's that's completely right. And I'm glad you said that so you're you're the end game you see is a scenario where it's likely that. The price of oil is going to be higher. Tell me. What is the percentage roughly that you think of the oil that's getting through the straight up for moves because it's officially blocked not just by a ran, but by the US, we have a blockade. So it's officially double blocked. And yet there are these claims that the ships are turning off their signals. So it's not being picked up. And so the official numbers of ships of tankers that are bringing oil through the straight up for moves is a very, very low number. And so it's a tiny percentage. And yet, yet some percent like how much oil do you think is getting through. I think it depends on week by week. I think two weeks ago, quite a few got through. I think last week very few. Okay. And I think you know, like in the last few days in general, very, very few. I mean, you know, there are people who do this for a living. They didn't care about the GPS and it's not. They're sitting there in a golf or hormones and just basically counting every ship that's going through. I mean, there are people who make a living just doing that. So from that point of view, like Kepler, for example, if there are a lot of ships going through, well, eat me. This will be the front page of the Wall Street Journal, right? And then hearing it from the Secretary of Energy, whatever it is. So what should I mean? Like, so from that point of view, like, you know, I, you know, I tracked the Bloomberg actually has a very good tracker. And I'm a blue subscriber and I get the data. And I think it's pretty decent. And so for what I know is that I think, you know, there was the week before last when the US managed to get a few tanker owners to turn off their GPS and trying to basically get through by selling through, you know, the north, the southern part of the straight, which is closer to a man under US whatever. I think that probably was the reason why we saw an increase in a number of tankers, but then Iran put his foot down. They started to deploy mines in that part of the sea. I mean, I think in general, like, I think at this point, I think that number is pretty low. David, we'll leave it there. Thank you so much for coming on monetary matters. Tell us about your research service, David Wu, unbound and where people can find you. Yeah, so if you don't want to pay, you just basically visit my YouTube channel, David Wu, unbound. If you're interested, we offer a retail subscription service for people who want to know more about my investment strategy for, especially when it comes to picking stocks. If you're an institution investors, just contact me directly because we have it. That's the bulk of what we do, which is providing some of the most sophisticated institutions in the world, on navigating geopolitics, politics, economics and technology. Thank you, David. Thank you, everyone, for watching. Please leave a rating and review for monetary matters on Apple podcasts and Spotify and subscribe to the monetary matters YouTube channel. Hope you enjoyed today's episode. Those interested in learning more about the Tukum corn fund ticker C-O-R-N can find more information in the link in the description. Until next time.