Odd Lots - Michael Cembalest on Why AI Is the Stock Market Bet of the Century

Episode Date: May 29, 2025

Michael Cembalest has been an investment analyst for almost 40 years and his research notes have drawn a cult following on Wall Street. He's known for going super deep into a wide range of topics, like energy and healthcare. And lately he's been writing a lot about AI, with a particular interest in figuring out whether all the investment in data centers and compute will translate into actual profits. On this episode, we talk to the chair of market and investment strategy for JPMorgan Asset Management about why AI is the market "bet of the century," why the dominance of US big tech can't be overstated, and why he's pessimistic about the outlook for small modular nuclear reactors.Odd Lots Live is returning to New York City on June 26. Get your tickets here!See omnystudio.com/listener for privacy information.

Thanks for listening to OddLots. Follow the show on Amazon Music for more future episodes or just ask Alexa play the podcast OddLots on Amazon Music. Hello, OddLots listeners. I'm Joe Wisenthal. And I'm Tracy Allaway. Tracy, we're doing another live show and it's right here in New York City. Yeah, this one should be our biggest yet. And we're going to have a bunch of OddLot's favorites and do something maybe a little different to some of our previous live podcast recordings. When the guests are revealed, the show is going to sell out right away. So you should really just go get your ticket right now. It's June 26. It's at Rackett, NYC. And you can find a ticket link at Bloomberg.com slash oddlots or Bloomberg Events.com slash

oddlots live and why. We hope to see you there. Bloomberg Audio Studios, Podcasts, Radio News. Hello and welcome to another episode of the Oddlots podcast. I'm Tracy Alloway. And I'm Joe Wisenthall. Joe, does it feel to you like every day when you wake up in markets, there is a new sort of dominant narrative. And often it is the exact opposite of the narrative that was dominant a day or two ago.

Go on. Say more what you mean. Where are you going with this? I mean, I agree in some sense, but I think you have something specific in mind. Well, no, I actually have a bunch of things in mind. So, for instance, you know, recently there was a lot of talk about the bond vigilantes being back. and people worried about the fiscal deficit. And then this week, that seems to have suddenly gone away, or at least it's not the only thing people are talking about.

Tariffs, you know, obviously we had the big market freak out in April, and then we had subsequent delays and extensions of the deadlines. I don't hear people talking about tariffs as much as they were in April. And then through it all is, you know, on days when we don't have to talk about tariffs or bomb vigilantes, then we immediately turn to AI. Well, this is what I was going to say. So this is the thing. There's like this default thing that's in the background, which is AI.

And then that, of course, leads to things like power and electricity and so forth. So yes, I now see what you're saying, which is that on any given day, it's like you spin the roulette wheel or the wheel of fortune. And then it's like, okay, it's an AI day. Actually, we're recording this May 28th by the time people are listening to this and video earnings are going to be out. Oh, yeah. And so today is an AI day, or tomorrow might be an AI day. What I was going to say is one of our previous guests, Victor Schwetz, over at McCory, he framed it really well where he was basically saying, if, you know, if you can't gauge what the dominant narrative is going to be every day or if everything that's happening is so volatile and so dramatic and we're talking about really big changes, then you kind of have to go back to doing the normal thing. And the normal thing before all of this kind of started in the new year was worrying about AI.

Right. And thinking about AI and whether or not all of that AI spending is going to be backed by revenue. So on that note, that's what we're going to do today. We're going to go back to normality, although we are going to talk about some of the current stuff as well. And I am very pleased to say we have someone who I have wanted to get on the podcast for a really long time. And he's finally sort of been let out into the wild. So we are very excited. We're going to be speaking with Michael Sembalest. He is, of course, the chairman of, of market and investment strategy for J.P. Morgan asset management. He writes a great, great piece called Eye on the Market, which he's been doing for- Which has a cult following on Wall Street. Every time a new Sembolist piece comes out, everyone talks about it and reads it. Yeah, and I think it's been going on for something like 20 years now. And he himself has been at J.P. Morgan since 1987. So, you know, a real witness to financial market history. So I'm very excited. Michael, thank you so much for finally coming on the show. Thank you very much. Good to be here.

I guess my first question is, can you explain what you do over at J.P. Morgan asset management? Eye on the Market is a phenomenal publication. I've been a fan for a long time. But I always wonder, like, is there a difference between writing investment outlooks for asset management clients versus writing them if you're on the sell side? Well, yeah, there's a huge difference. First of all, you know, thank God for the Chinese Wall, ERISA rules and things like that. because I have the independence and what our clients want is for me to share what I'm thinking. The best example of that is in February 2021, the investment bank was making money hand over, all investment banks were making money hand over fists in the SPAC market. And I published, you know, two really scathing pieces on SPACs.

The first one was called hydraulic spacking and the next one because it was during COVID was called speccine hesitancy, where I kind of pointed out that the sponsors were the only ones making money and that their break-even threshold was something like an 80% decline in the merged stock price was their break-even. So anything better than that, they would make money. And everybody else was getting slaughtered. So, you know, being behind the fiduciary wall allows me to kind of call it like I see it. And the eye in the market is the external presentation of our thoughts and views. And, you know, on an internal basis, I'm participating our investment committees dealing with asset allocation and security selection and everything else that goes along with managing money for

endowments, foundations, pension plans, insurance companies, sovereign wealth funds, and individuals. It does feel like on the media side, we are at the whims of narratives and all the things that Tracy talked about at the beginning, whereas if you talk to people who are engaged in the art of security selection, it feels that actually might be liberating because on some level, whatever is going on, you could still look at an instrument and say, is this a well-priced instrument or not? And not having to get caught up so much in just, you know, movement, you know, whatever the headline of the day is. Right. And actually, you know, if you're managing money other than kind of a fast money macro hedge fund, you don't have the luxury of changing your asset allocation every time a new narrative comes along.

Right. So you, to some extent, you have to kind of choose your poison, make your asset allocation decisions. and then be either rewarded or punished based on the outcome. So one of the reasons we wanted to get you on the show was because you've been writing in great detail about AI. So let me just ask the basic question. If we're not going to worry about, you know, Trump and tariffs and bond vigilantes

and all the sort of headliney stuff, if we're going back to considering, I guess, the macro and market importance of AI at the moment, how dominant is AI when it comes to market valuations at the moment? I guess how fundamental is AI when it comes to the wider stock market? I don't think you can overstate the importance of this stuff. Wow. When you decompose the evolution of S&P profit margins and you strip off the MagS7 versus the rest of the market, it's like East and West Berlin.

Right. I'm dating myself, right? But that's how different. That's how divergent those are. If you look at earnings growth, I think earning. growth for the MAG 7 over the last 10 years is close to 20%. It's 6% for the S&P 493. So it's almost like two completely different universes of companies. Wow. And so they are extremely profitable. They are spending somewhere between, depending on a quarter, 25 to 40% of their revenues on capital

spending and R&D. Those numbers are unprecedented. And in the history of the markets, even going back to the 1960s mainframe era and things like that, we've never seen anything quite like this. and what the hyperscalers are doing, I would describe, I know we're only 25 years into the century, but this is the bed of the century, that you can spend this much on AI infrastructure, you know, depress your free cash flow margins for some period of time, and wait for the ultimate payoff. Even prior to the AI, you know, you mentioned this extraordinary gap. I mean, there are two things about hearing you describe the earnings picture within the S&P 500 that strike me. One is the absolute gap.

But then there's also the fact that it's... It's the biggest companies growing really fast. And I, in my mind, when I was younger, assumed that as big companies got bigger, they grew slower, and smaller companies grew faster. When you look at stock market history, how rare is the type of earnings growth, the year-on-year, the huge numbers that they put up earnings growth-wise, how unusual is it for stock market history for the biggest companies to also just be galloping past everyone else like this growth-off? You can't find it.

You can't find it. I mean, stock market history on a security-level basis. goes back to around 1980. And there's a variety of databases through facts that, through the University of Chicago. That's about as far back as you can't go much back. It's hard to go back further than that. There's some Wilshire data, but it's difficult, very spotty. So if you're looking for things like earnings growth, if you just want stock price growth,

you can get it. But if you're looking for earnings and margins and things like that's hard. So we can't find any comparable period where the bigger companies are growing faster. And even within the private equity universe, your theory holds true. The generally smaller and mid-sized funds usually are going to outperform over the long periods of time, the larger funds. There's negative size bias. So this is pretty unique. And, you know, over the years, I've always thought that the biggest risk to all of this was going to be some reinterpretation of the Sherman Act and, you know, antitrust.

And while Epic Games has won a lawsuit here and the Department of Justice is picking away at Amazon, Apple, and Google for different reasons, there really hasn't been a fundamental rethinking. of antitrust law that would get in the way of this. Can you talk a little bit more about that? What are the drivers or the circumstances that are allowing Mag 7 to continue to grow in a phenomenal pace, but also just become these giant cash flow monsters? Yeah, well, it has to do with the digitization of the economy and how much technology is at the core of how companies run. You know, J.P. Morgan right now is a firm. We've got 300 different language model and AI projects going on inside the firm.

And all of that requires, in one way or another, a lot of these Mags haven't come. I think we have to put Tesla aside for a minute for reasons that we can discuss. But technology is really at the foundation. And I think a lot of times when people compare and they say, oh, U.S. stocks are expensive relative to the rest of the world, they're missing a couple of important things. They're missing the fact that the U.S. has a massive weight to technology. And in Europe, you know, basically it's a value market, right? It's a lot of energy, financials, and consumer staples. And so the U.S. is really the dominant player in large-cap tech.

There's an amazing chart. Mario Draghi, you know, who he is, obviously, wrote a piece at the beginning of the year. What can we possibly, I'm paraphrasing, what can we possibly, too, to reinfigurate entrepreneurship in Europe? And he had a chart in there that we re-computed in our own way. And it's a bubble chart that shows the number of new companies created in the U.S. versus Europe since the year 2000. And again, it's like East and West Berlin. They are completely different.

Yeah. One of the things that struck me recently, we did an episode with the CEO of a women's clothing company, M.M. Lafleur. We're mostly talking about tariffs. But the thing that I keep thinking about is she said that when she started the company a little, I think, when was it over 10 years ago, Tracy? Yeah.

Her customer acquisition costs on a platform like Instagram, where she was like $12 a customer. And now it's like over $200. And it just struck me that one of the things that must be going on in the economy is everybody's margin becoming Facebook's profits or everybody's margins eventually becoming alphabet profits. That part of this gap that we're seeing is there is this digital real estate that exists. It's where everyone is moving. And they could collect massive rents from owning that real estate from anyone who wants to do commerce there. Yeah, and I think rent is the right word here.

Yeah. And for reasons both good and bad, you know, Amazon requires, if you want to be, if you want to sell to prime customers, you have to agree to use Amazon shipping. If you make an in purchase app in some kind of a game like Fortnite on an Android phone, you have to make that purchase through the Google Play Store. These are some of the issues that have been litigated in some of the, and so I'm not 100% in favor of some of the tying. that essentially they get. If J.P. Morgan said, if you want a credit card, you've got to do your mortgage with us, we'd get slapped down pretty fast and for good reason. And so there's a lot of product time that takes place in big tech, you know, that a lot of companies are beginning to challenge and the DOJ is looking at as well. On April 4, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm.

Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. So we've been talking about the dominance of U.S. tech. And, you know, there are a lot of numbers to support this and you've been laying some of them out. But before we had all these tariff-related market freakouts, the freak out that, Everyone was focused on was Deepseek, of course. And the idea that, oh, suddenly China has invented a low-cost AI model that is very competitive with what we've seen from the giant companies like the Microsofts and the Googles of the world. Looking back at that, was that freak out overstated?

Yes, it was. Okay. All right. For a couple of reasons. One, I think it's become clear. And I wrote a piece at the time. They were not completely transparent about how they accomplished some of their goals. imitations the Sears

form of flattery. They actually used, I think, open AI models to train their models. And so they did a lot of piggybacking that isn't necessarily a sign of some kind of productivity breakthrough. Also, the price at which they are willing to sell something

is not necessarily reflective of its true cost. And so there is some loss leading that's going on in there that, again, is different. That said, cheaper AI models, if that's where we're going, is probably going to make adoption even easier. So if this can all be done with less energy at lower token cost, that's a bullish story for AI,

not a negative one. It also occurred to me when that happened that if there is this existential battle between the U.S. and China for AI supremacy, we'll probably going to spend more on it. You know, if it becomes a nuclear bomb type race, then the impulse is going to be to spend more on it. But actually, you mentioned things like energy costs per token and so forth. And I'm curious from your perspective, or you know, you've been doing this for a long time. How did you learn about that? How did you educate yourself on understanding the sort of the key economic variables that you have to get wrap your head around in order to say informed things about this new topic?

Well, there's a bunch of things. Because we're trying to figure that ourselves. You know, we probably should be doing more about AI and talking. So we need to learn about how you tell yourself about it. I think you have to budget your time. And my most important thing is reading about 2000 to 2,500 pages of research each week. So that comes first. Other than spending time with my wife and my dog, that's the most important thing I do, because that's where, whether it's on energy topics or politics or healthcare or biotech, like it's an information-laden world. And the great thing, about J.P. Morgan is I can call just about anybody and say hi. I'm the chief strategist on the asset management business of J.P. Morgan. I'd like to talk to you about something. Would you talk to me?

I can count on one hand the number of times that somebody has just flat out said no. And one of those people was the bundler for Bernie Madoff. So, and that's the great thing about the Halo effect that Jamie has brought to J.P. Morgan. And remember, I've, I've, I worked to J. for many years when there was no halo. And, you know, that halo effect is very powerful and it benefits me and a whole bunch of other people to company. And I recognize that every day. So I spend a lot of time working, you know, deeply on some of these topics. And energy as an example, I needed to learn about energy from somebody that understood energy better than anybody else. So 15 years ago, I made a pilgrimage to Manitoba in February, right, which was can you prove yourself because it was cold.

and I established a relationship with Valklov Smill, who for over a decade, was my personal technical Sherpa on understanding and learning about energy. That's someone we haven't had on the podcast. He is a hermit. He's Czech. He's over 80.

Oh, okay. And he's a hermit. That's our excuse. And he hates the political dynamics in the United States. So you will never get him here. We'll go to Manitoba. We'll try. I'm still stuck on the idea of the Madoff guy,

like answering the phone and being like, oh, we would prefer not to talk in depth about why our line just keeps going up in a very smooth. Well, remember, it was an 11% compound return with 2% vol. It was an unprecedented return history like in the history of hedge funds. Well, also, Tracy, I don't expect you to comment on this, Michael, but Tracy, if I recall, one of the people who was initially raised alarms about Madoff was actually a guy, Matt Zames at J.P. Morgan. Oh, yeah. So perhaps there was a reason, I don't want to talk to these J.P. Morgan fellows. Actually, if you go back, if you go back even further, in either 2001 or 2002, there was a guy named Harry Markopoulos in Boston.

Yeah, yeah, yeah, yeah. And he wrote a piece basically saying Bernie Madoff is a fraud. That's right. It was a Ponzi scheme. And he sent it to the SEC. And they kind of poked around and did nothing. Did nothing. Yeah. Okay. So when it comes to, I guess, learning about new sectors like AI, I think one of the things that everyone, us included, is struggling with at the moment, is to understand, like, how much actual corporate spend there is going to be on this. And for obvious reasons, companies, you know, they're not necessarily that vocal about how much they're spending on this technology, although they do like to talk about the importance of AI in a sort of very nebulous theoretical manner. But what are you doing to understand the actual spend on this tech? Right.

So there's this multiple stages to this. The first stage was understanding how much the hyperscalers were spending. And once we looked at those numbers, we were astonished. And for the reasons you're mentioning, concerned, right? I mean, there's not a lot of examples. If you look back at cable and airlines and casinos, when you start seeing R&D and CAPEX as a per cent per cent for a sustained period, you know, usually that's turned out bad. So we said here, okay, what's going on here?

The next thing to look at was signs of corporate adoption. Now, there's nobody that hates McKinsey surveys more than me, but, you know, we had McKinsey surveys, you have Bain surveys, you've got census surveys, a lot of which we're pointing in the same direction, which is that corporate adoption has been accelerating, particularly over the last six months. and CEO surveys. I mean, there's too many of them that are all telling you the same thing for them to be wildly off the mark. But again, that doesn't tell you how much they're willing to spend on it, right?

You know, everybody might enjoy lollipops, but that doesn't mean they're going to spend a lot of their personal income on them. Then the question is, how much are the big hyperscaler starting to earn on AI? And so far, Microsoft is really the only, company that's disclosing it very discreetly. And we had a chart. I published a piece on this on May 13. And Microsoft's the only company so far that's really explicitly telling you how much they're earning.

And those numbers are growing at a pretty rapid pace from a low base, but they're earning a lot. 150% over the past year is what I see in the note. Right. Right. The other companies are still burying it in cloud revenue. Yeah. There's a lot of questions just on this so I get to them. It's one thing for a company to spend on AI, you know, a lot of pilot projects on there. When you ask companies in terms of, yes, but is this actually saving money? Is this actually profitable investment? It starts to get cagier. And sometimes when you hear people talk about the returns that they've got from AI spending,

it actually sounds like they're talking about more traditional machine learning algorithm stuff that has been going on for a while. But for this spend to continue, it's going to have to flow. through to the bottom line in some way. And I'm curious what you're seeing on that in terms of like, here is a company that's not a tech company. Maybe it's an airline company. Maybe it's a chemical's company, whatever, where it's like, you know what, they can point to something that's not just algorithms that is actually generative AI that is making money for them or saving money for them in some way. And what are you seeing on that? Yeah, we're seeing anecdotes.

Antiquets. We're seeing anecdotes. Antic data. Yeah. And usually it has to do with fast. throughput and call centers with less people. Stuff like that. Faster customer acquisition, fraud reduction, right? For big banks like J.P. Morgan, generative AI, one of the huge potential payoffs is identifying serial defaulters and mortgages and credit cards and things like that.

So what is interesting is among the best surveys that are done, the surveys are asking them explicitly for generative AI, how much money are you saving? and most of the answers are still in the lowest category, zero to 10 percent. Okay, by far. And then it's the above those levels have fewer responses. We're still in the exploratory phase here, and I don't think we're out of the woods, and there may be a day of reckoning at some point. I think because of how absurdly profitable these companies are,

the markets are going to give them another 18 months at least for the proof statement to bear out. This was going to be my next question was like how much of a leash are investors allowing some of these companies. As long as the cloud revenue and the AI revenue specifically are growing at 50% plus. And as long as the free cash flow margins don't fall below some critical level, I think the markets will still feel okay about it. I wanted to also ask a question about your overall career. As we said in the intro, you've been at JP Morgan since 1987, which is a phenomenal phenomenal amount of time looking back. What was the hardest or most challenging sector to wrap

your head around over that period of time? Because as we've been discussing, you jump from thing to thing and you basically become an expert on whatever is important to investors or the market at that particular moment in time. Yeah. And I usually get airlifted into things that nobody, you know, that are a pain. Like I became for the asset management business and even for the broader firm, I ended up being our COVID point person. And I, I, I, I, put together a medical advisory committee of biophysicians and mathematical epidemiologists and people like that from La Jolla Institute of Immunology. And so I get airlifted into things. I would say, oddly enough, of all the sectors and sub-ectors I can think of, the one that drives me nuts the most

is health care and managed care. Because, you know how they snuck this provision into the energy bill three years ago where the government can negotiate drug prices finally after never have been able to do that? But on the face of it, that should be negative for the big farmer companies. But because they can play around with all sorts of discounting mechanisms and distribution arrangements, a lot of which aren't super public, the actual impact on their margins so far has been somewhat negligible. So there's an impenetrability to parts of the way the health care system function that drives me a little bananas.

That said, large cap pharma and biotech right now are among the cheapest sectors if you compare price to book to projected ROE or something like that. We have this giant map of all the sectors and subsectors and industries and where they trade relative to price to book and R.A. and R.O.E. And, you know, healthcare is cheaper. But it would normally make me want to dive in and kind of come up with an asset allocation recommendation for our investment. process and it's just very difficult to do. This is how I feel about U.S. health insurance, by the way, as someone who did not grow up that much in the U.S., I still struggle to wrap my head around how all of this works. It is truly impenetrable to me.

I think many people would agree with that assessment. I want to go back to actually AI for a second because you're talking about the hyperscalers and this booming business for them. It's still actually a little bit hard for me to understand because, for some companies, their AI revenue is very easy to understand, right? So Microsoft sells various things that go along with its suite, that whatever, whether it's coding tools or so forth, okay, they're selling an AI service. Maybe Google one day figure out how to do that too.

For a company like Meta, their model is an open source model. They don't sell Enterprise Lama as far as I know, yet they spend like crazy on AI, and they talk about how much they spend. Can you articulate for me what Meta's AI business model is? Because all I hear is about their spending and I don't even know what revenue they're deriving from. Yeah, when you try to understand that, what you learn is that meta basically considers itself a generative AI company.

Yeah. And that everything they do and the success of all of its algorithms and social media tools maintain their market dominance because of how well they function. And the way that they do. function is because they're driven by AI. And I agree. I think they thought that the open source LAM model would become an indirect profit center as people built tools and applications on top of it. And it seems like they're stepping away from that right now. Like they, a year ago, that was the kind of message you were hearing and not so much right now. But they, you know, I think they would say

are in. But somehow all of this spending, which they talk as AI spending specifically, even though they're not selling AI, but somehow it's accruing to their fundamental business of selling ads or selling content or selling time. And here's how I would articulate that. If you go back three, four years, there were these really interesting cell side presentations that talked about, well, Google's going to try to compete with Apple here. Apple's going to try to compete with Nvidia here. Invita is going to try and compete with opening out here. And guess what? A lot of those modes are still in place. And a lot of the encroachment that people were looking for hasn't happened.

So meta's AI spend is primarily focused on maintaining an impenetrable mode for other companies that might otherwise want to compete with their core business. And that's how they see it, which means that you basically will want to look at top line, free cash flow and other profitability metrics for the company as a whole to understand the benefits of all that spent. I think investors are right to be skeptical. It's only three years ago that they took a ton of money, put it in a pile, and set it on fire. Yeah, with the metaverse, right? So they have had a history of misreading where things are going. Tracy, by the way, I actually really admire how quick they were willing to pivot off that. Because you think, you know, you make this big

bad, it's embarrassing. You rename your company meta off of this. And then you're like, we're an AI company. I actually that raised my admiration for Zuckerberg as a CEO that he didn't like succumb to some big sunk cost fallacy there. Anyway. When the facts change, I would say, I would say, I would say, I changed my version of the Keynes quote. He has demonstrated a mental agility to pivot as well. Yeah. Earlier this year. And I'm not going to make a value judgment on this. But I was struck by him saying that he needed to re-inject a dose of masculinity. Oh, yeah, yeah. I think that's a sign of somebody that. Yeah. This is audio, so you can't see my eyes rolling. But that's a sign of somebody that is willing to kind of change horses when.

Yeah. Oh, dear. I have thoughts on AI as a masculine endeavor. But anyway, one thing I wanted to ask is, okay, so at the beginning of this conversation, you described AI as the bet of the century. It is. And when I hear, you know, when we're talking about the dominance of the big U.S. tech players and the moats and the billions, if not trillions of dollars,

that people are spending on this now. It feels like almost certainly it's the big players that are going to emerge as the winners from this. It seems very, very hard for smaller entrants to get into this. Is that right? Is it just, you know, fate accompli, I guess, that the big guys are going to wait? At the hyperscaler level, yes. But in a whole bunch of other industries, not necessarily.

Okay. There's plenty of room. But look, you're seeing these unicorns all the time that are chipping away at different aspects of machine learning and generative AI. One of the most fascinating ones is look at Palantir, right? I mean, Palantir is taking a run at the large defense contractors primarily through its nimbleness

and better application and understanding of the intersection between generative AI and defense weaponry. And for everybody that thinks about data centers, the massive concentration of data centers in Virginia and in the PJM ISO region more broadly, is a very clear indication of just how the national security and defense stuff is tied into generative AI and who their big users are. Right.

If that didn't matter, there would be a lot more data centers built in the Midwest where 20 to 30% of the time, electricity prices are actually negative, right? Think about this. You have certain applications that are so reliant on not there being a millisecond of lag that they'd be willing to pay more for power and proximity. than to have cheaper power that has just a tiny bit of lag built into it. The news doesn't stop on the weekends. Context changes constantly. And now Bloomberg is the place to stay on top of it all. Hi, I'm David Gurra.

Join us every Saturday and Sunday for the new Bloomberg this weekend. I'm Christina Rafini. We'll bring you the latest headlines, in-depth analysis, and big interviews. All the stories that hit home on your days off. And I'm Lisa Mateo. Watch and listen to Bloomberg this weekend for thoughtful, enlightening conversations about business. lifestyle, people, and culture. On Saturday mornings, we put the past week's events into context,

examining what happened in the markets and the world. Then on Sundays, we speak with journalists, columnists, and key political figures to prepare you for the week ahead. Join us as soon as you wake up and bring us with you wherever your weekend plans take you. Watch us on Bloomberg Television. Listen on Bloomberg Radio, stream the show live on the Bloomberg business app, or listen to the podcast. That's Bloomberg this weekend.

Saturdays and Sundays starting at 7 a.m. Eastern. Make us part of your weekend routine on Bloomberg Television, radio, and wherever you get your podcasts. One company where there's some question about the moat is Alphabet. And, you know, people are like, well, could Open AI just become people's default homepage or the first page that people go to when they want to learn something? And you've seen Alphabet shares underperform into some of the rebound. I think from a multiple standpoint, it's trading at some of its cheapest levels in a very long time. There's a lot of questions about the degree to which it can productize its own models. Historically, it's not so good at that.

It makes great tech. The models are very good. I'm still discovering new alphabet models that I never knew existed because they're buried under some URL. Is there a risk to their business model specifically that perhaps their distribution, their ownership of digital real estate, risk to the rent that they can extract from it? Yeah, but this is the kind of thing where I would continue to just look at their market share. You know, for a while, Bing was taking a run at them and then nothing really happened. Okay, so in Europe, when you buy a device, Apple's no longer able to make Google the default engine.

Oh, yeah, yeah. So you have to set it up yourself. People pay Google anyway. Yeah. So you could actually argue that Google is overpaying Apple for the default status on their devices. That's interesting. Because people would be willing to pick it anyway. I don't know.

I think obviously the tools, open AI, I personally love perplexity, but I use them when necessary. And I wouldn't bet against Google's generic market share for search until you actually saw them beginning to lose market share. It reminds me for the last 15 years, people have been talking and screaming about the end of the dollar is the world's reserve currency. And every year, we look at this analogy. We look at the data on the share of foreign exchange reserves invested in dollars, the share of corporate and equity debt issuance in dollars, particularly offshore, the share of swift transaction payments. And the dollar share of all these things, BIS intercompany bank loans across border,

the dollar share is kind of unchanged. So it's, you know, you have to wait until you see stuff. Network effects are definitely a thing. One thing I wanted to ask you, or one of the reasons we wanted to have you on, is, as you say, you read a lot, you talk to a lot of people, a lot of clients. And I'm curious, what's the sort of time split or the question split between, I guess, Trump and policy versus, you know, other market questions like AI? Hmm. It's hard. I mean, it's hard to say, right? I mean, the tariff, the whole tariff drama was, took an enormous amount of time. Because I would say

that a lot of CEOs, and I've described this recently, is I went to college in the Boston area, and every time it snowed, some of the local kids would come out, and when there was a lot of snow and ice on the road, they would hold onto the back of trucks and go skitching. And I had never heard of skitching until I saw people doing it. That sounds funny. I never did it because, like, after you stop skitching, you tend to slam phase first into ice. A lot of CEOs are skitching right now on this administration because they thought that what they were going to get was deregulation, oil and gas pipelines, and lower corporate tax rates. That's what they were supporting this guy for.

And they come out of the gate with deportations and tariffs, and the productivity shock supply side agenda got put in the back seat for a while. They appear to be wanting to get started on some of that stuff. And I think we can talk about it. It'll be interesting to see if they can get the Constitution pipeline reinvigorated, which would bring Marcella shale gas to the Northeast. But in the beginning, because so much of the first stuff was tariffs and deportation, There's a lot of focus on that. And tariffs in particular require a lot of work. There's 12,000 HTS codes across 200-something countries, and they came up with a patchwork of legislation that differentiate. And so there's just a lot of math involved in really understanding what the impacts are.

By the way, I see you into Tufts. And so I think you're the first person to ever say you went to a Boston area college in that not being code word for Harvard. You actually did go to a Boston area college, excellent university. There's lots of them. No, I know there's lots, but it all, you know, you weren't being coy when you said Boston area of college. As you mentioned, you know, another huge theme over the years and you write a lot about it is energy and you've learned a lot from Falkloff's Mill and, you know, we could talk hours just about energy.

I would love to talk hours about energy at some point. But in the interest of time, how much has the AI story over the, the last few years changed how you thought about the future of North American energy needs and, you know, energy production and just the energy story in general? You know, not until very recently because it was a negligible part of overall electricity consumption, you know, sub 1% until a couple of years ago and has been growing slowly. There are three big components to this electricity story. One's AI.

another one's electrification of transportation, you know, EVs. And the third one that people should focus on because it's just as big is the potential for electrification of winter heating in homes, office buildings, and industrial locations. One of the hardest things, when you break down energy consumption, an enormous amount of it, an enormous amount of fossil fuel consumption is used for heat. some of the heat, around a third of all the heat that's used, is used at temperatures below 200 degrees centigrade. And that's where heat pumps are extremely efficient and measured by something called the coefficient of performance, which is the units of heat you get per unit of electricity you put in. And so there's three components roughly equal, right? So data centers, electrification of heating, electrification of transport. The wide-eyed, you know, kind of fairy tale projections of the Rocky Mountain Institute and green tech media people are looking at 20 to 25 percent growth in electricity demand over the next 10 years.

I think that's way too high. Adding up those three components. The lower end is closer to like 5 to 7 percent. BNEF, which I think does very good work on these topics, is somewhere in between. Thank you. Thank you for the plug. Again, I know all of the sources and the rigor of their work. And so I think we will need, you know, seven or eight percent additional electricity capacity for generation over the next decade or so. Electricity consumption of the United States has been roughly flat for 20 years, but before that, it went up a lot. The issue back then was we were meeting it with nuclear and large natural gas gas gas.

It's a lot harder today because most of the country is going to try to do it with renewables, which is just very very. difficult. And I'm also, for reasons I wrote about in our annual energy paper, I'm a skeptic of small modular reactors. Great idea on paper, very complicated in terms of execution. By the way, Tracy, it's interesting the heat as one of the big consumption drivers of energy. You know, there's a lot of northeastern chauvinism where they say people shouldn't be living out in Arizona and all these like air-conditioned places. But it really is heat. Like really, if we want to be more energy efficient, my understanding is we should all leave northeast and go to the desert and live in air conditioning instead of heating our home.

No, for real. As someone with an old house in Connecticut, I will push back on that narrative. You know, I have a personal bias on this, but I was very excited about heat pumps, but then we looked into it for a home that was built in 1850, and it turns out you would have to like re-insulate the entire house for this thing to have any effect. Oh, yeah. No, the biggest problem with heat pumps is the fact that while it's more efficient, you're buying electricity instead of natural gas, and per megajoule of energy, it can be, depending

on the state you live in, three to four times more expensive. So it's great for the climate. It's just economically, you know, challenging. And that's why I think some of those Rocky Mountain Institute and Green Tech Media forecasts are way too aggressive, because I think these transitions are going to take place over a much longer period of time. I'm Francine Lacqua, an award-winning journalist, and I've got a new podcast, leaders with Francine Lacqua from Bloomberg Podcasts. I've interviewed to everyone from heads of state to fashion icons about the news of the moment. But I've always been curious who are these people as leaders. I don't think there's one right way to be a leader.

Make decisions. A poor decision is always better than no decision. Listen to new episodes every other Monday. Follow leaders with Francine Lacroix wherever you get your podcasts. Since we have you, since you've been in the business for almost four decades now, If I was reading one of your notes or one of your research pieces in, let's say, 1987 or 1988, I'm not sure what exactly you were doing back then. I was an analyst. Oh, okay. And I was sitting someplace behind a pile of pizza boxes and nobody was talking to me. I don't know. So how different would your research then be versus your research now? I'm definitely somebody that has taken advantage of massive quantities of information and research, right? It was a lot harder. I used to have to go to the Mid-M Manhattan research life.

library to find things. And I used to go and you pull out microfiche and then you would go and you would try to, you know, you put in a quarter and they would let you print it. I mean, it was hard to get historical data back then. And I'm a data hound and so the way the world works today is much easier for me to draw parallels. I've also learned a lot of lessons in investing for all these years. I was the chief investment officer just in the private bank for a long time. And so I've learned a lot of lessons. And we're in the process of putting together a 20-year-on-the-market retrospective. And it's amazing to go back and publish some of those pieces and see what we learned at the time. Oh, well, okay, give us a preview. What were the big lessons? The number one thing, more than

anything else, and we have 30 topics we're going to show. And some of them are in there kind of for humor or sarcasm. But the number one investment lesson was equity's bottom before everything else. And so does credit and so does high yield and so does real estate. In other words, asset prices bottom so far in advance of the related fundamentals. For instance, during the financial crisis in 2009, bank stocks bottomed when only 10% of the ultimate bank failures had taken place. So if you were listening to the really bearish voices at the time, they would say, too soon to invest, there's a giant tidal wave of bank failures coming. But the KBW or whatever index you want to look at bottomed in advance of all of that.

And the same thing happened during the S&L crisis. Stocks bottomed way before all the devolts eventually took place. And if you then take the seven post-war recessions, in six out of the seven of them, equities bottomed before you even saw an upturn in payrolls, industrial production, housing, credit card delinquencies. And so as an investments person, I have to acclimate our clients and our investment committee to be willing to take risk when you look out the window and everything looks terrible. Yeah. Oh, this is a great lesson because, you know, people say right now, oh, the shortages from the tariffs haven't even hit. Maybe we won't have them because the tariffs

have been dialed back so much. But the idea that maybe the market already bottomed for this cycle on everything related to tariffs. Sometimes I like to say hope is the only strategy because if you wait until all of the data confirms that things are back, it's clearly going to be priced in at that point. But I actually wanted to ask something else. I am an SMR skeptic too,

but I form my opinions based on seeing five tweets that I like, okay, yeah, those sound good to me. I'm an SMR skeptic. I'm, you don't, you have a much more rigorous process. Can you explain to us why you are a skeptic of small modular reactors? Okay. So the easiest one is nobody's built one.

That's a good answer. In the United States. Yeah, yeah. One hasn't been built. Point number two, there are three or four of them, depending on how you want to define what small is, that have been built in China and Russia. And the costs have been multiples of the original projections and the timeframes have been much longer.

So failure, even in places that have success building regular. Here's what I figure if they were so good China would have built 100 by them. Right. Third point, there's no industrial project on Earth that is more capital intensive than nuclear power.

And so when nuclear power first started to become popular in the late 60s and early 70s, the goal was, how big can we make them because we want to spread some of the sunk costs over the most megawatts we can build? The notion that you want to

shrink a capital project to make it more efficient makes no sense to me unless you can completely completely commoditize them. The mistake I think people are making is that there has been an incredible learning curve in solar, polysilicon panels, to lesser extent, in wind turbines, both onshore and offshore, and the best example is lithium ion batteries. Best example of a learning curve you've ever seen. But that's because millions and tens of millions of units are being produced. I don't believe that you

can from going from one reactor to four reactors or six reactors develop the kind of learning curve that would result in exponential cost declines. And so, you know, I'll wait until I see it, and I don't think we will, but I would love to eat my hat if ACHLO or some of the other public companies or private companies that are working on this can deliver a small modular reactor in anywhere of the zip code of the original projections. And just as just one last thing, reminder, the new scale, you know, originally said three million dollars a megawatt. When it hit 20, the plug was pulled. So the most recent example of such efforts in the United States failed when the budget

hit 6x. That was a fantastic answer, very intuitive answer, too. Real quickly, what about a geothermal? You think that could be a meaningful contributor to our portfolio? I do. You know, it takes effort. Chris Wright, who's a secretary of energy, he's a big fan. Yeah.

And, you know, there's different kinds of geothermal. The most interesting kind is where you're going really deep, and you're accessing, like, supercritical fluids whose heat is comparable to nuclear power. But, you know, that instrumentation has to be able to withstand very high heat. You have to use plasma drilling bits. I think there's opportunity for expansion of geothermal. But they're long-term industrial projects. It'll take a long time to do, not a quick fix.

Yeah. You know. And the other thing, too, is, and this is, and this is, amazing. I've been very focused on the rising costs of wind and solar power, and I should have been paying more attention to it. I turned my head the other day. The cost of a natural gas combined cycle turbine has gone from 1,200 per kilowatt to 2,500, like more than double, just in the last two years. Supply chains and mostly a reflection of the fact that GE Vernova and Siemens and the other companies involved don't believe the long-term demand story

enough to expand their production facilities. Interesting. So that's why... Classic energy phenomenon. Right. So that's why you go to GE for Nova today. We're happy to deliver you a plant. It's going to cost two times more than it did a couple of years ago,

and we'll give it to you in five years. That's the AI bottleneck. This is how falling demand can result in higher prices, or expectations of falling demand. Right. Right. Because the two to three entities, Mitsubishi as well, that are involved in this sector,

don't feel comfortable enough to go out and do a mega capital raise to double and triple the size of their production facilities for these kind of turbines. And that, I think, is going to end up being a binding constraint for AI at some point. I just don't know why. So one of the reasons I'm a big fan of your research is because it is so wide-ranging. And again, you have a very diverse diet when it comes to what you read and the information and data that you consume. Last question for me, but give us one under-the-radar thing that you are watching at the moment.

What should, basically, what should we do an episode on? Yeah, what should we do an episode on? Let's see. I mean, and not in any kind of order or importance. I'm just going to give you a stream of consciousness. Sometimes I heard you in the introduction talk about narratives that come and go, and I agree with that totally. And one of the narratives that came and went,

but might come back is there are still $5 to $700 billion of, unrealized interest rate related losses on bank balance sheets. That has not gone away. Now, it only goes away if you go back to, you know, 3% or sub 3% on the tenure. But we still have a banking system where many of the participants went hogwild, extending duration on treasuries and agencies at a time of, you know, historically low interest rates because you had this deposit surge from all the Fed bazooca money. And that hasn't gone away.

Now, the Fed has proven itself willing to provide whatever emergency facilities. I actually was, you know, personally against bailing out the VC depositors in Silicon Valley Bank. You know, the average deposit balance in that bank was $2.5 million. I mean, it wasn't really a bank. It was something else. But they decided for systemic reasons to do it. They'd probably do it again. But that issue hasn't gone away.

Yeah. And the higher 10-year rates go, the more those losses get pronounced. So that's an issue that's kind of still out. I'm glad you brought that up because I remember, Tracy and I did a bunch of episodes in like 2021, 22. It's like, why isn't the rate rise having more of an effect? And everyone's like, oh, you know, they got termed out. Everyone locked in low rates, et cetera.

And then it's like, hey, but now rates are higher. At some point, like they only locked them in for so long. I'm almost tempted to believe that the Fed essentially created this problem through financial repression tempted some bad asset liability managers at the, these regional banks to load up at low rates and feel some sense of actual responsibility for this mess because it was fed monetary and congressional fiscal policy that led to this mess. The deposit explosion. Yeah.

So I think they feel somewhat beholden to try to manage around some of the risks. Another thing I would say is I don't think the tariff story is done. Okay. You know, the average tariff rate at the end of last year was something like two and a half percent. I'm using round numbers. Yeah. If you did all the math and you don't assume any import substitution, at the peak of the kind of, you know, Lutnik cardboard cutout, you know, presentation. We were looking at a

a 25% all-in rate. Again, assuming no import substitution and things like that. Now we're back down at like 13%. So the market's like, wow, we dodged a bullet. It went from 20-something to 14. But 14 is still very high compared to two. Isn't it like the highest since the 1930s or 40s or something? Yeah. Now, he may backtrack again. you know, we'll see. But historically, whenever there's been a disconnect between hard data and soft data, it takes about 90 days for that to resolve itself. So some of the shocks that are seeping through the system are going to take until July or August to play out.

One of the things I wrote earlier this year that resonated with a lot of people is the markets are the one that you can't deport them. You can't intimidate them. You can't arrest them. you know, at the end of the day, the markets are going to be a binding constraint on what this administration can do and can't do. And that's actually a plus for investors because it means at some point they're going to have to adhere to some generally acceptable level of policymaking. I just have one last quick question. Going back to AI, you hear a lot about inference versus training and people wondering, could there be some alternative to Nvidia for some of these services? Because right now on the chip side, all the profits seem to be accruing to Nvidia.

Is that moat, you know, people talk about the CUDA ecosystem. Is that an impenetrable moat or are there opportunities for other chip companies to gain some real profits in the AI world? There are, you know, in the piece that we did last year, we listed all the companies that are trying to pick away at the Nvidia moat. But in the asset management business, or when companies hire underwriters. Yeah. If an underwriting goes bad and you picked J.P. Morgan Goldman and Morgan Stanley, nobody's going to question you. If you pick a ranked 8 or 11 underwriter,

and something goes wrong, people will question you. So to some extent, the network effect and the moats that you guys are referring to, nobody's going to get blamed for spending extra money to buy NVIDIA products and wait for a long time until these other products have proven themselves. And so I think it's going to take a long time for that to play out. But yeah, eventually, if this AI story works, we should be looking at 80% inference and 20% training in two to three years, right? At that point, corporate adoption, which drives the training models, are supposed to be the

overwhelming amount of what's going on. But again, Amazon claims to be working on a foundational model, haven't seen it yet. Microsoft has actually kind of hinted that, oh, we've created our own foundational model. There was this great story recently. If you like AI, the head of the Microsoft AI group was basically told to pound sand by the Open AI people when he was asking them to fully disclose how some of the new reasoning models worked. So Microsoft is working on their own foundational model. Has it been released yet, even internally within Microsoft. So again, some of these motes, I wouldn't underestimate the amount of time that NVIDI is spending, defending that moat and improving its products and its energy efficiency so that,

they don't give a market share prematurely. All right. Well, by the time this episode comes out, NVIDIA earnings, will have been published. So maybe we'll get a little bit more insight. But Michael Sembalist, thank you so much for coming on all thoughts.

Yeah, that was a real pleasure. It's a crime that we hadn't had you before. But I guess the fact that you couldn't, you know. I think I've been put out back in the cage after this point. Joe, that was a real treat to get Michael finally, finally on the podcast.

It was so good. And it's like, I've always liked his notes. I mean, A, I think everyone who hasn't read them will now understand why he has sort of this cult following on Wall Street where everyone devours his note. But then even beyond that, hearing him, like, in my mind, it's like, okay, these notes are great. But there's a difference between comfort at like being able to put together this big research product or this big note product, but also just being able to talk about all these things extemporaneously. Absolutely.

It's really impressive. There's so much interesting stuff to pick out there. But one of the things that struck me was, I guess, the importance of geographical location for data centers. Because I hadn't thought about that. But he's absolutely right. You know, you could locate a bunch of data centers in the middle of nowhere where maybe there's cheaper energy costs. But people so far have not chosen to do that. Yeah, I hadn't really thought about that either because it was like, you know, we all know that Northern Virginia is this huge data center cluster.

But why the data centers really need to be that close to. the government, the defense industry, and so forth. It's like, come on. It's like, isn't really that big of a deal. Right. To like your extra millisecond of late this year or something. But I mean, you know, they're obviously there for a reason. I also just thought like this idea of the moats, you know, obviously that the big tech

companies have built up and the amount of effort they have to sustain those modes and distribute through those modes and so forth. And like how powerful that is and how much that is the story of the fact that, the biggest keep getting bigger at literally the expense of anything else. It's such an important perspective I have on all kinds of dimensions, whether you're thinking U.S. versus international, the price of U.S. equities, and so forth. You just have to remember, like, how extraordinarily unusual these companies are. Absolutely.

And the fact that they continue to just throw off cash even while they're spending billions and billions of dollars on AI tech, which we're not entirely sure, like the degree to which it's actually going to be able to be monetized. but they still managed to do it while generating cash. The clarity of the way he spoke about energy is really impressive. And I really liked what he said about small modular reactors. And no one's done it yet in the U.S.? Yeah, and when we talk to Jigger Shah about it, he used an analogy, which I thought was fantastic.

You know, he's more optimistic about them. He used an analogy that I thought was great, which is that the nuclear industry has to be in the business of building airplanes instead of airports. Because airports are all one-offs, but airplanes are obviously, you know, they're built on a very complex assembly line, but, you know, after you've built several, the cost gets cheaper. And this idea that there's so much capital intensivity in building nuclear and the idea that there is no insight prospect for that learning curve where you do it over and over and over again for the cost to fall dramatically, that they haven't even done that in China. Maybe you're not going to

ever really get the sort of learning curves that you get in solar or lithium ion batteries super, very clear, intuitive explanation of why the opportunity may not be there. Yeah, we have to have them back on. Yeah. Oh, so much more we can talk about it. All right. Shall we leave it there? Let's leave it there.

This has been another episode of the All Thoughts podcast. I'm Tracy Alloway. You can follow me at Tracy Alloway. And I'm Jill Wisenthal. You can follow me at the stalwart. Check out all of Michael Sembalist's notes. They're all online. Search for them.

All of them are a must read. I'm going to go back and read a bunch of them. Follow our producers, Kerman Rodriguez, at Kermann Armin, Dashel Bennett at Dashbot, and Kale Brooks at Kail Brooks. For more Oddlots content, go to Bloomberg.com slash oddlots. We have a daily newsletter and all of our episodes. And you can chat about all of these topics 24-7 in our Discord. Discord.g. And if you enjoy Oddlots, if you like it when we talk to market veterans like Michael Semblis,

then please leave us a positive review on your favorite podcast platform. And remember, if you are a Bloomberg subscriber, you can listen to all of our episodes, absolutely ad-free. All you need to do is find the Bloomberg channel on Apple Podcasts and follow the instructions there. Thanks for listening. You can get the news whenever you want it with Bloomberg News Now. I'm Amy Morris. And I'm Karen Moscow here to tell you about our new on-demand news report delivered right to your podcast feed. Bloomberg News Now is a short five-minute audio report on the day's top stories.

are published throughout the day with the latest information and data to keep you informed. Yes, there are other products like this from a variety of news organizations, but they usually rerun their radio newscasts throughout the day. That's not what we do. We create customized episodes that can only be heard on Bloomberg News Now. And we don't wait an hour to publish breaking news. When news breaks, we'll have an episode up in your podcast feed within minutes, so you're always getting the latest stories and developments.

Get the reporting and the context from Bloomberg's 3,000 journalists and analysts we're all over the world. Listen to the latest from Bloomberg News Now on Apple, Spotify, or anywhere you listen.