Full Transcript: Aswath Damodaran on Valuing SpaceX and AI

The Dean of Valuation on Investing in a Changing World

Kai: Our guest today is the legendary Aswath Damodaran, who has been teaching corporate finance at NYU for over four decades, earning him the moniker “The Dean of Valuation.” He has published a dozen books on finance and investing and posts regularly at his blog, Musings on Markets. I am particularly excited to have Professor Damodaran on as a guest on The Intangible Economy, as he has been a long-standing advocate for the importance of intangible assets in company valuation.

In my opinion, this lens has helped him apply classical valuation techniques to a much broader set of companies, ranging from asset-light tech companies to consumer brands to younger firms earlier in their corporate life cycle. Professor Aswath Damodaran, thank you for taking the time, and welcome to The Intangible Economy.

Aswath: Thank you for having me.

Kai: So let’s start with the news of the day: SpaceX. SpaceX, of course, went public last week, and the company brings together almost every hard problem in valuation today: high and uncertain growth, negative profits, limited financial history, few company peers, huge addressable market, massive capital requirements, disruptive narratives, and key person risk.

You first valued the company in April and then updated your valuation once the prospectus dropped a couple weeks ago. We now know what happened. SpaceX IPO’d at a valuation of $1.8 trillion and has been met with unbridled investor enthusiasm. As of this recording, it now trades at around $2.7 trillion, making it the world’s fifth-largest company with a market cap greater than that of Amazon and slightly below Microsoft.

So obviously, the narratives around SpaceX are super upbeat. However, you’ve argued that the narratives must be supported by the numbers, in this case, the valuations. Can you walk me through your SpaceX valuation? What was your target valuation, and how did you arrive at this figure?

Aswath: I mean, if you look at SpaceX, it started as a space launch business. In fact, people don’t seem to quite know this. It’s older than Tesla. It was founded by Elon Musk with the money that he cashed out with on PayPal about a few months before Tesla was founded. So it’s been around. In terms of chronological age, the company’s been around a long time: twenty-five years. And when it started, I think the Musk vision was people need to look to space, go to Mars, and that for many people when you think about SpaceX, they think about like Blue Origin. It’s another company trying to get people out to Mars. But over time, that’s not how that part of the business has evolved.

What SpaceX essentially has done is it’s reinvented the space launch business. Let me step back. Pre-SpaceX, if you wanted to launch a satellite into space, you were dependent on either a government agency or an established defense company, and they did it on the side. This was not their primary business. It was a one and done. You launched something into space on a rocket, the rocket blew up, and you were done, but it was an expensive process. And when SpaceX came along, the response from many experts in the field — and that’s something to remember, Elon Musk has consistently proven these experts wrong — said, “You can’t do this. This is a very different business. You don’t have the expertise to do it.” And he changed their minds. He changed the business by creating relaunchable rockets. That’s a revolution he brought in, and it gives that part of the business an incredible cost advantage over competition.

Recently, there are some young companies that have tried to do what SpaceX has done, but essentially they’ve lowered the cost of launching things into space, and they’ve been lucky in a sense that we need more and more stuff in space to support what we do need on Earth, from GPS to our phones. Satellites have become an indispensable part of what we do. So at its core, that is SpaceX’s original business, and for about fifteen years it built that business. It was slow to take off because it’s a very technological business. I mean, it’s an engineering marvel. This isn’t an app company or a software company. This is a physical infrastructure company, and it took them a while to get their feet on the ground.

But even after they got their feet on the ground, it’s not a big business. It’s not like you got thousands of people lined up to launch stuff into space. It’s a small niche market, and they have a huge market share of that market because of the cost advantage. But about ten years ago, they realized that they could launch satellites into space cheaper than everybody else, they could create a second business, and that’s of course the connectivity business, Starlink. Broadband internet using satellites. Now, the advantage there that SpaceX has over the competition, again, is since they can get satellites into space far cheaper than everybody else, they have far more satellites in space, ten thousand, which allows Starlink to have coverage that no other satellite-based internet service company can offer.

It allows Starlink to offer service in parts of the world where traditional internet has broken down. That business is now at the core of the revenue for SpaceX. If you look at what’s delivering the money for SpaceX right now, it’s the connectivity business, the Starlink business. But if you stop right there, this would be a very good business, but it would be a niche business. Basically, neither of those markets are big enough to sustain a trillion-dollar company. What changed SpaceX was, of course, the acquisition of xAI, parent company for Grok, and also started by Musk in February of 2026, where the companies came together. And what that opens up is a third business, which is the AI business.

How big is the AI business? According to the prospectus, it’s $26 trillion. It’s the largest ever total addressable market I’ve ever seen. We can debate that. Is that really $26 trillion? But I think after the debate, even if you are not as optimistic, you’re gonna come up with a really big business. What that opens up is a huge business where they’re competing against, at this point, two other private companies, OpenAI and Anthropic in the main. There are other side players, but a bunch of established tech companies, including Google, which has Gemini, and these are all LLMs trying to go after that really big market. That market is huge promise, huge potential. It’s a big market. That’s where growth is going to come from. That’s what’s driving the trillion, two trillion, two and a half trillion, two point seven trillion dollar pricing.

But the business is really not a business yet. People are struggling on what business models work. If you look at Anthropic, which is furthest along in trying to make this a business, they’re still struggling with this subscription versus usage cost. And I think increasingly they discovered that if you give subscriptions to people to use AI products and you don’t constrain them, you’re very quickly going to lose money. So they’ve increasingly shifted away to usage-based models. But even with those usage-based models, you run into a second challenge, which is: How do you make enough money? The unit economics right now are not great. Let’s take Claude Fable, for instance, which is the AI version that was pulled off the market last week because of the US government.

Now, I’ve read, and I don’t use Claude Fable, but I’ve read that it costs $6,000 an hour to use Claude Fable. Whether it’s $6,000 or — it’s a really expensive product to use. And if your reaction is Anthropic must be making a lot of money on Claude Fable, you wouldn’t be right because it actually costs to deliver the high-end AI product service.

Where do the costs come from? They come from the data centers, the power, the water, all of this stuff, and those don’t easily lend themselves to economies of scale. So there’s a business problem at the heart of the LLM business that nobody solved, but somebody might solve it, and that player is going to have insanely high value because it’s a big market, and you’ve solved the problem. xAI, as part of SpaceX, is part of that market and a large amount of the numbers that you see — and I think Elon put out a number of $1 trillion in revenues in 2030.

The only way you can get there is if AI takes off and if Grok happens to be one of the lead winners of that AI market. So this stage of three businesses loosely connected. Why loosely connected? Because the connectivity business is built off the space launch business. And even the AI business, there’s an argument that being the best player in the space launch business might give you an advantage in the AI market. It’s right now almost science fiction of putting data centers in space. But, you know, let’s face it, putting data centers in traditional — putting it on the ground somewhere is becoming more and more difficult because the... it changes the landscape wherever you put it.

So there’s naturally political pushback from people living in the area saying, “We don’t want a huge data center sucking up power and water here.” So we’ve got to put them somewhere that people don’t see them. I mean, I’ve read stories about data centers in the middle of the ocean, data centers in space. Right now it’s more science fiction, but you could argue. So there’s a loose connection across the three businesses. Now let me get to the brass tacks of how I valued each. I took each one, and let’s face it, there’s not much there yet in any of the business. The connectivity business has about $15 billion in revenues.

It accounts for sixty, seventy percent of the total revenues of the company. But collectively, SpaceX is losing money. But to cut it some slack, much of that loss, in fact, all of that loss comes from the R&D expense they had in the most recent year. And this cuts to the heart of this intangible question, is as long as accountants mistreat R&D as an operating expense, any young high-growth tech company, even after it starts to make money, will look like it’s losing money. If you correct for that R&D, I mean, you might turn that negative into a positive number, but it’s still a company with small revenues and at the margin, very small profits. So the entire value of the company comes from what’s gonna happen in the future. So I told three separate stories. The keyword here is stories.

You think, “Why are you using stories when you’re doing valuation?” ‘Cause it’s all about the future. I can’t just extrapolate the past. And with each one, I thought I told a story that for me was not just reasonable, but perhaps upbeat. So I assumed that the space launch business would continue to grow as companies increasingly want to put things into space. And I assumed that SpaceX would continue to dominate that business because of their cost advantages. Sixty percent market share ten years from now of a business that’s going to be eight, ten times larger than what you see today. On the connectivity business, the business itself is big if you define it as internet services, right? Because all of us need the internet for our life, for our work, for even for our homes. And right now, most of us get our internet either through fiber optic or phone lines. And at the moment, for most of us, Starlink is a suboptimal option because it cannot deliver the speeds that fiber optic and cable can do.

So in my story, I assume that the technology for satellite-based broadband would improve, but it would still remain only a portion of the total market. If you live in a city or in a well-serviced urban area, you’re not gonna switch to Starlink. But their reach will expand into parts of the country where traditional internet is not delivering good service, as well as from companies that need to provide internet service. An airline’s a classic example. If you’ve used traditional internet on an airline, it’s so insanely slow that at the end of it, you get one email coming through every hour. It’s pointless. So United has talked about converting eighty percent of their fleet to Starlink, and that will be a dramatic improvement. So I did assume that the market would grow and that because of, again, the fact that it has the most satellites in space, Starlink would win that battle as well.

So big stories. On AI, here’s the challenge. The market is big, but the unit economics don’t look great yet. And as far as I can see, I’ll accept the big market. I’m willing to go five trillion, six trillion, not to the twenty-six trillion. But I don’t see an easy way for this to become a market where... By unit economics, I’m talking about how much you make on the next unit you sell. It’s going to stay, in my view, a low gross margin business because of the nature of the business. In fact, there’s an internal tension here, which is at the high end of the spectrum, AI can replace a person, but it’s so expensive to deliver that AI that you’ve got to be a McKinsey consultant that I’m paying $1 million for AI to actually pay off for me.

So if you want to build the market being big story, it has to go with bringing down the cost of high-power AI, and that’s going to be tough to do. This business is still very much in formation. We’re gonna run into this again when OpenAI goes public and Anthropic goes public. The total addressable market is what bankers and the company are going to use to dazzle us. “Look at how big the market is.” I’m looking to see what they tell me about the unit economics, and my guess is they’ll underplay it because it doesn’t look good now and there’s no — this is not like unit economics in a software business where, “Hey, you know what? As I scale up, my margins are gonna go up because it costs me almost nothing to produce that extra unit.”

I’d be interested to see what the stories are for these AI companies as they go public. In the case of SpaceX, I did assume that Grok would actually become a legitimate contender for the enterprise AI market, which right now it’s actually third in line. Or even fourth in line, because it hasn’t really focused that much on that market. I assumed that they would find a way to focus. They recently — I mean, they just announced Cursor’s acquisition, and that’s really an enterprise AI solution. You’re providing coding solutions to companies that need coding. So I took them seriously. I took them at their word. But because my margins are lower in this business than in the other two, it turns out that the net effect of going after the bigger enterprise AI market, which is what I allowed after reading the prospectus, kinda canceled out.

We make this mistake of assuming growth is always good. But growth, when it’s accompanied by huge amounts of reinvestment and substandard gross margins, which unfortunately the state of the AI market now is insane amounts of CapEx, might not just be neutral to value, but actually be value destructive. I did assume there’d be value creation, but it’d be far less than people will get at first impression by looking at the total addressable market growth.

Now, after I did the valuation, people pointed out that — you know, SpaceX is actually generating, or will be generating, almost $2 billion in revenues from Google and Anthropic by renting out data centers. They said, “How come you’re not counting that?” And I am, but it’s actually going head to head against your big AI story, right?

In the big AI story, what’s SpaceX saying? “We’re going to be players in this space. We’re going to compete. We’re gonna win a significant market share of the AI market.” But in the same breath, they’re also saying, “We’re renting out space in our — from our data centers to our biggest competitors.”

That’s like a manufacturing company claiming that they’re gonna get a big market share, but they built a big factory, and they rent out two-thirds of their factory to their two biggest competitors. Something in this story will have to gel. Either SpaceX decides that it’s not going to go after the AI market and says, “Our competitive advantage is building data centers. We’re really good at that.” Maybe that’s where the space launch data centers will come in. “We’re gonna make our money charging OpenAI and Anthropic and Google for using our data centers.” It’s a viable model, but the trade-off is it’s a much smaller total addressable market. You might have higher margins.

It’ll be interesting to see how SpaceX plays it because right now you’re looking at unformed companies still struggling, and I don’t think any of them know how this is going to play out. They’re trying. You can see them almost throwing stuff at the wall and hoping something sticks. But my end result is I value the three companies, and I think ultimately the companies will figure out something to happen. So if your reaction as you look at my story and valuation is that’s a lot of assumptions you’re making, yeah, absolutely.

What choice do you have? You know, if you’re number bound, I’ll tell you up front, SpaceX looks awful as an investment, right? If all you can focus on is what they have on their books. And I think the title of your session is intangibles. When people think about intangibles, they think about brand name and stuff that you don’t...

Do you know what the biggest intangible is? Future growth. Even for a company like SpaceX, which is not a traditional intangible company, right? It’s not a software company. It’s not a brand name company. If everything for your new value is coming from what they will do in the future, you can’t see it yet. And if —

Kai: My definition of intangibles, you can’t see it. That is the ultimate intangible. And from that perspective —

Aswath: the bulk of my $1.3 trillion value. Now, at the same time, this is my story. It’s not the story for SpaceX. And one of the things I did in my earlier valuation that I will probably follow up and do on the post-prospectus valuation, is I introduced the uncertainty I feel about the estimates. That the total addressable market for AI could be much bigger, that they will find a way for gross margins to improve. The way this will play out, instead of having a point estimate valuation, which is what we’re taught to do in valuation classes, this makes your big inputs into distributions, and what it’ll create is a distribution of value.

And distributions of values are good to counter hubris, because what that’ll show you is this is my estimate of value. This is how wrong I can be. That’s one piece of output you should get from this, is you have an estimate, but you could be wrong. And the other is when somebody else comes with a different number. If somebody says, “Is it possible that SpaceX is worth $2 trillion?” Of course it is, given the uncertainty about the future. And if someone else insisted that they think SpaceX is worth $2 trillion, who am I to contest them? They have a story. They want to put their money behind it. So I value for an audience of one, which is me, and for me, this is a company that has great promise.

I love — I mean, I think it’s an engineering marvel, an amazing company, a company that perhaps only Elon Musk could have created, by thinking out of the box and hiring incredible engineers. But at two trillion, two point two trillion, two point seven trillion, that’s a bridge too far for me to cross. And for me, it’s a great company, but at the wrong price.

Kai: Right. I mean, I think what’s so interesting about your framework is you’re trying to bridge two competing camps, right? There’s your kind of numerically driven value investors who say, “Price to sales ratio one hundred, negative income, no thanks, pass. No way.” And then you have your kind of more story-driven narrative investors who hear the story you told and perhaps inflate it even further, and they’re like, “Yeah, $2.7 trillion.” I mean, that’s where the market’s trading, so obviously someone believes that that’s the correct valuation.

And so what you’re trying to do, what’s so interesting with your framework is, through the decomposition into sum of parts and then kind of interrogating the assumptions around, say, TAM, unit economics, and then reinvestment requirements, putting together a framework by which now anyone who has your spreadsheet can download it and ask the question of, if I’m at $2.7 trillion, where am I differing from you?

Aswath: Or what does the gross margin have to be? Or you can — remember, an Excel spreadsheet, you can solve for any input. How big does the market have to be? What does the gross margin have to be for this to play out? And then decide for yourself, is that something? Now, when I teach my valuation classes, one of the first questions I ask my students is, “What are you more comfortable with? Working with numbers or telling stories?”

And because I have MBAs who come from very different backgrounds, of course, there are quant-oriented people who say, “I like working with numbers,” and storytelling people who essentially... And in the investing world, this plays out in the form of if you go to bankers, especially old-time bankers or value investors, they’re very focused on EBITDA and revenues and multiples and trying to make sense of that.

And if you go to VCs and founders, they’re big storytellers. They say — My problem with both sides, if you stay in your camp, is you’re missing big parts of what you do. The number crunchers are so focused on what’s there now that they forget that this is not a company. The analogy I would offer is this is like having a kindergartner’s report card and extrapolating from that what they will be doing in college, which is essentially what you’re getting with the SpaceX financial statement.

There’s not much there, and there’s not much there because financial statements reflect your history, and you don’t have much in your history. My problem with the storytelling that I hear is it stops with the total addressable market. That seems — I mean, I’ve watched CNBC as to why SpaceX should be worth $2 trillion. And your story does it. It’s a big market. And then I wait for the rest of the story, and it doesn’t come. And here’s my problem. This is a business. I don’t get value from having a big market. I’ve got to take that big market, I’ve got to monetize it as revenues for me. So I need to get your story as to why I am going to be a beneficiary of that big market, which requires that you talk about the relative advantages of Grok versus OpenAI versus Claude versus Gemini.

You need to get at least a working acquaintance of those. And then you got to talk about unit economics, which is how much are you gonna charge for your product. These stories are incomplete. Incomplete stories are basically a recipe for essentially doing what everybody else is doing. In other words, these are not stories that tell you whether you should be buying the stock. They’re stories you use because you’ve already decided you want to own SpaceX, and you’re looking for a justification for why that was okay. We all do it. Numbers people do it, story people do it. We make decisions first, and then we look for rationalizations later.

It’s human nature. What I’m trying to do is slow the process down because like everybody else, I have instincts that lead me to want to do something. I want to slow the process down and give my rational side a chance to at least mount an argument.

Kai: Right. So what stands out to me about the point you made over the years is that intangible assets, narratives, they’re real, but ultimately, you need them to flow into the financials. Like at some point, they need to show up in the higher margins, higher growth, or some sort of lower cost of capital, something like that. And so that’s one of the interesting things too, which is, for example, we know that SpaceX has built a lot of organizational know-how and IP around reusable rockets and satellite launches. That has to show up somewhere in the financials. You can’t just say, “Well, this is great,” and then you double count the value, right? So I think that in addition to the point you made about capitalizing R&D, which makes a small difference in this case, in terms of the future value of the company, you really have to believe, for instance, that they will continue to have a commanding cost advantage in launch, to believe that that value materializes and then, of course, to the AI point as well.

Aswath: Yeah, I mean, I think that’s the — it’s a reality we tend to forget. Ultimately, you’re valuing businesses. You’re not valuing promises. Converting a promise into potential into a business takes work. The problem though is we have selective memories. We look at success, we use hindsight bias, and we act like we would have picked the successful company at the start.

Amazon, classic example, right? People assume, they look back at Amazon, “How could you have missed it?” Especially younger people, “How could you have missed it? Why didn’t you buy Amazon in 1999? It’s obvious internet was going to take off. It’s obvious that Amazon was —” Was it? I was there in 1999. Neither of those things were obvious. So I think there’s a lesson from previous big buzzword disruptions that you can bring to the AI space.

Like the internet, it is going to be huge. It’s probably gonna change — it’s already changing the way we live and we work. This is not the metaverse, right? Which I was never excited about. What the heck is the metaverse? Why should I care? Remember that brief period where Facebook was throwing in billions of dollars in the metaverse. This is real. I see it in my wife, who teaches fifth grade, dealing with ChatGPT-generated solutions. I see it with my son, who works at Disney, who’s had interactions with OpenAI because Disney and OpenAI created a partnership, but they’ve since dissolved, to bring AI into Disney. So this is showing up in our daily lives.

It’s showing up when I open up my iPhone in the form of AI-generated responses that I’m getting to things that I do on a daily basis. This is real. I think we need to accept that. So the people who hide behind “this is all entirely fluff and buzz,” they’re missing the point. They’re not in the world they need to be in. I think, though, converting that into a successful business is gonna be tough to do. And of every 10 AI companies that have started, there’s gonna be one that looks like Amazon 30 years from now. It might not even be a name you recognize today. And people then are gonna ask you, if you’re 25 years from now and you missed it, “How come you didn’t see that company?”

I think there’s a huge amount of hindsight bias that leads then to ROMO and then to FOMO. And you’re saying, “What’s ROMO?” Regret over missing out. You regret the fact that it... So what do you do? If you’re afraid of missing out now, so you buy SpaceX at $2 trillion, saying, “I can’t afford to miss the next Amazon, the next Facebook.” So I think that hindsight bias and the combination of the regret over missing out and fear over missing out are kind of playing out in what you see. I mean, let’s face it, we haven’t had a high-profile big IPO in a long time. Markets came back after the 2022 drawback, but IPOs did not. So this will be a year where three huge IPOs come out, and there’s a lot of pent-up FOMO out there that’s going to go into these IPOs.

So I’m not surprised, I’ll be quite honest. I wasn’t surprised at that first day. I’m surprised at the second day, but that’s kind of mixed up with the war ending and oil prices dropping. So there’s a lot of other stuff happening as well, capital coming off the sidelines. But I’m not surprised that the money is coming in to SpaceX, but it’s based on pricing. It’s based on fear of missing out rather than people telling much bigger stories than I am and assessing a higher value for SpaceX as a company.

Kai: Yeah, I mean, I think what you just said is interesting because it ties to your work with Brad Cornell on the big market delusion, right? The idea of a huge market, whether it’s $26 trillion or even some fraction of that, combining with overconfidence on the part. Oh, yeah, of course, we’re gonna win, right?

Aswath: Overconfident entrepreneurs plus overconfident venture capitalists, which is kind of the standard description I would give for any set of entrepreneurs. You have to be overconfident to be an entrepreneur. You have to be overconfident to be a venture capitalist. “I can pick winners.” It’s a feature, not a bug of the people who self-select to do that. And you put overconfidence in a big market together, you can see the big market delusion play out in the sense that you overreach, then there’s a correction. And a big chunk of the overreach is, “I wish I hadn’t done that.” But it’s a feature, not a bug of big markets, so I’m not surprised that it’s happening with AI. And there will be an overreach, there will be a correction, but that doesn’t mean every single one of these companies is overvalued. The correction will be in the aggregate, but there’ll be a few winners that come out of this space.

Kai: Right. So the problem was just collectively, if you were to invest in OpenAI, Anthropic, and xAI via SpaceX —

Aswath: If you were to invest in AI —

Kai: ...the capital raise is more than the —

Aswath: Yeah, and you know what? There will be people who do it. There are ETFs I’m sure that’ll be created, which will be AI ETFs, and you’ll be investing in the AI space. You are, in a sense, investing and hoping the big market delusion is not a delusion. I wouldn’t go there, but there will be people who will be tempted to do it.

Kai: Yeah. I mean, there’s another aspect of this collective dynamic around the capital cycle. So I had Edward Chancellor on last month. And so that’s another interesting question because in your SpaceX valuation, you say that the AI segment of SpaceX has the lowest economics, the most fierce competition, and the unit economics of AI due to the high cost of goods sold is just not as favorable as the other businesses, and it’s unclear whether they’ll get scale economies. But at the same time, it is the biggest area of reinvestment for the company, right? So SpaceX and Elon are pouring all their resources into —

Aswath: Not just the company. It is driving the economy. I mean, that’s the real big difference between the dot-com boom and bust and the AI boom. We don’t know whether there’ll be a bust, but history suggests there will be a bust. The dot-com boom bust, there was no huge capital expenditure in that cycle. In fact, there was very little traditional CapEx or even R&D driving it. People started apps. They started basically going online. This has been the biggest, I think, infrastructure run up I’ve ever seen of a business. You can go back and compare to the automobile business 100 years ago.

Kai: Yeah. Railroading.

Aswath: This is — the amount of money that’s being put into AI CapEx is immense, which means that when the correction comes, the pain will be more intense. And herein lies a second problem. The dot-com boom and bust was almost entirely equity funded. You’re saying, “So what?” Well, when the bust came, there were shareholders who lost sixty, seventy, eighty, or even ninety percent of their money. You felt sorry for them, but the loss was restricted to the shareholders.

The problem with the AI CapEx boom is not only is it immense, but a big chunk of it is funded with debt, the debt coming from private capital rather than banks. And there is a very real chance that if there’s a correction and companies start having problems, that problem is gonna show up as distress and default, and that pain doesn’t stay restricted. It spills over into the rest of society. I’m not saying it’s gonna be 2008, but 2008 is an example of what happens when lenders overreach, when they lend money at too low a rate, and the correction comes, the pain spills over. So that is my concern with this big market delusion, is the potential societal cost of having to deal with debt coming due that you’re unable to pay is much more painful than your share price dropping 90% and you feeling the pain.

Kai: And what do you think about the big tech companies, the so-called Magnificent Seven, whose businesses have transitioned from more asset light to more capital intensive, utility-like, as a result of the AI build-out?

Aswath: I think they’re the least exposed to default risk. I’m not worried about default, the distress there, but they’re changing their entire characteristics. It’s like you used to hang out with this 150 pound lightweight, and all of a sudden he disappears. He goes, and he comes out as a 250 pound muscle-bound. It’s a very different character you’re gonna be hanging out with, right? I mean, it’s... He’s gotta eat protein six times a day. He takes... I mean, that’s what’s happened with these companies, is they’re changing their characteristics of the company. If you’re a long-term investor in these companies, as I have, I’ve invested in Amazon since ‘97 off and on. And I own five of those Mag Seven companies.

I’m coming to terms that these are different companies. The way I have to read their earnings reports has to change. It’s not just about looking at margins and new businesses. I’m looking at CapEx. Where’s the CapEx going? How is it depreciated? Things I didn’t think about with these companies seven, eight, nine years ago. I’ve got to think about it. Doesn’t mean that I can’t value them, but I’ve got to value them differently than I did because they’re changing their characteristics as companies. And my worry is these are not companies used to this traditional CapEx-driven investment. They’re companies that got sloppy and lazy ‘cause they could grow with very little reinvestment.

And they’re now doing something they’ve never done before, build huge factories, infrastructure investments, which take 10 years to depreciate but could become obsolete in five. It’s a very different game, and I’m not sure they really know what they’re getting themselves into, which means that I have more acceptance of what Apple is doing as opposed to the others, because Apple’s saying, “Look, we’ve never done this. If we decide to invest tens of billions of dollars in CapEx, we’re playing a game we never came to play, we’re not very good at. We want to stay in our lane.” So I think what Apple’s doing is a very different way of approaching the AI market. And while there are —

Kai: Extremely restrained in terms of spending.

Aswath: More restrained, in fact.

Kai: For a lot of the CapEx.

Aswath: Absolutely. And I think a lot of analysts and investors are down on Apple for that reason, which is, “How come you’re not jumping in with both feet?” And I think they have a basis for their strategy reflective of Tim Cook’s personality, which is, when he announced that he was gonna step down in September, a lot of — I wrote a blog post, and the title kind of gives it away: “An Ode to Restraint.” Because I think we undervalue restraint in business, and this might be one of those markets where restraint is a good feature. Let other people make the big mistakes and then step in and say, “Hey, now that we have learned from Google’s big screw-up, which cost them $15 billion, I know what not to do.” And I think the worry of being first there and being able to provide the AI product and services is driving these companies to overinvest and perhaps jump into spaces they’re not equipped to be very good players in.

Kai: Right. So on the note of changing characters of businesses, or potentially in that vein, one of the big themes we’ve seen has been the rise of companies like Micron. So first Nvidia and then Micron. Semiconductor stocks have historically been commodity-like, cyclical, kind of lower quality businesses. And of course, they’ve risen tremendously. Micron’s a trillion-dollar company now. You’ve done some work, actually — I read one of your older papers on evaluating commodity businesses and cyclical businesses. And so the question I have here is, obviously, at this point in the cycle, these companies are being driven by these macro themes, namely the AI CapEx, and they’re downstream of that. So the question is, do we think that AI is a structural tailwind that has changed the characters of these businesses? Or do we think that we have now the risk of valuing them at a cyclical high, peak earnings, and they may be due for a crash? How do we think about where we sit now with this sector of the economy that has become, I think, a quarter of the stock market is now chip stocks in the US, which is surprising.

Aswath: We’ve talked about three groups of companies, right? We’ve talked about LLMs, we’ve talked about big tech, and we’ve talked about chip companies. You know what the common theme is that’s emerging in all of this? It all depends on what that final AI product and service market is going to look like. If it’s going to be ten trillion with margins of forty percent, all of these companies, all three groups, are going to be much better off, right? They might not all win, but they’re going to be in a much better space. If the final market turns out to be three trillion with gross margins of twenty percent, there’s a heck of a lot of cleaning up to do with all three groups.

There’s no way around it. It’s almost like with these groups, you have to take a stand on what you think the AI product — So if you’re older and you’ve never used ChatGPT, and this is a completely foreign space for you, you need to get acquainted with the space if you’re a portfolio manager. Of course, if you’re an average person, go live your life. You said you don’t have to value SpaceX or invest in Micron. But if you want to play in this sandbox — and as an active investor, you have no choice but to play in this sandbox — then you got to be part of this conversation. And some of that conversation is playing out in the social slash political realm.

The best case scenario for AI, that ten trillion dollar market, will happen if it replaces people. I mean, here’s the bottom line. If AI is a tool, it’s gonna be a much smaller market than if AI replaces people. So the stories we’re telling about ten, fifteen, twenty, twenty-five trillion markets are actually terrifying stories for the rest of the world. Why? Because if that story comes true, half of all white-collar people are gonna lose their jobs. And what are they going to do instead? Who’s going to come up with the income to buy the products and services? I mean, there’s an entire debate going on, if AI works as well as it’s supposed to and replaces people, how do we deal with that as a society?

‘Cause people lose their jobs, not only do you lose your income, you lose your life’s meaning. There’s a whole set of... I mean, go back to the ‘90s when factory workers lost their jobs, and we were blasé about... I mean, I remember listening to people saying, “Learn to code,” and... I can’t even believe you would say something like that to a 55-year-old steelworker. And one of my favorite German words is schadenfreude, and I think that there’s an element of schadenfreude because what happened in the 1990s to factory workers is essentially threatening to revisit us, but this time it’s gonna be bankers, consultants, white-collar workers who are targeted. And to the extent that those factory workers are still around, you know what they’re gonna turn around and say? “Learn to plumb.”

Kai: Nice.

Aswath: Right? AI can’t replace your plumber. You have a leak, you can’t call AI and say, “Fix my plumbing.” But the scary thing is the big stories you tell that can justify AI, if they come true, are going to create some insane costs for society that we better start thinking about right now. ‘Cause you can’t wait till half of white-collar workers lose their jobs and then say, “What now?” This is something we need to be thinking through. I don’t think anybody is, because it’s almost like you don’t want to open the lid of that box and look to see what’s inside. But all of these groups are being driven by this AI fever dream at the moment, and we have to think about what if that fever dream is right, and think through the consequences. And then ask, what if the fever dream is wrong, and think through the consequences, ‘cause everything’s on the table right now.

Kai: Yeah. So I think you wrote a blog post on this in March on the doomsday scenarios.

Aswath: Yeah.

Kai: What was interesting in that piece was you put together a framework with two dimensions. So it wasn’t just the magnitude of how successful AI was and how big the markets were. The speed also mattered, right? The path from point A to point B matters a lot. Does it happen overnight, or does it take ten, twenty years for the S-curve to form and for diffusion to occur? And in that piece, you said not only is the labor market impact going to be different in those two scenarios, but also the effect for investors — since it’s an investment podcast — on the companies and, you know, obviously we see software stocks selling off in the wake of AI disruption fears. So my question here is around, as we think about the speed of the adjustment, from the standpoint of an investor in, say, the non-AI stocks, so the rest of the economy — what companies are truly being disrupted, which are maybe oversold, and how would that play out as you dial up and down the potential speed of AI transformation?

Aswath: You know what? I get my data from Capital IQ for individual companies. I would love it if Capital IQ also had columns on what percentage of the workforce in these companies is white collar and what percentage is blue collar, right? So you take a GM, or even a Kraft. So let’s take a GM. High percentage is blue collar, small — You take a McKinsey, it’s almost all white collar. And so one way to build a truly meaningful total addressable market is to take the collective salaries you pay all of these white-collar workers and say, “My limiting case is I replace all of them.”

I mean, I know it’s a nightmare scenario for those people that are replaced now. I think the wild card in that post that I did not bring in, that I should have, is the cost of delivering these products and services. That’s why I brought in the fact that the high-end AI stuff, which is what you need to replace your white-collar worker — I kind of ignored the cost of that, because I did not factor in that if it costs you $600,000 to replace a consultant, the AI product you create, then only consultants you pay more than $600,000 are going to get replaced. So that’s gonna be a crimp on how much that story can overreach. The Citrini story has to be revisited by saying, unless the cost of delivering AI products and services becomes much lower as we go through. And the pathway to a lower cost is not obvious to me, because what’s driving up the cost, a lot of it is physical cost, like power. So something has to break in this or change in this for it to change. Model —

Kai: Efficiency.

Aswath: Yeah.

Kai: And I think — So I actually have an idea.

Aswath: Yeah.

Kai: I have an idea of how we can get at that.

Aswath: Yeah, go ahead.

Kai: So I did a paper a little while ago where I looked at job postings for different companies. So Ford and GM, which you mentioned. And you can look at over, say, a trailing twelve-month period, what the composition of those things are. They’re all mapped to ONET codes based on the BLS database. Each ONET code is then collapsible or expandable, I guess, into the specific tasks that those workers do, each of which you can assign a score as to how much LLMs can automate each one. So you can come up with a score. You also have average, median, and percentile salary ranges for each of these employees. So you can say, “All right, what percentage of GE’s workforce or GM’s workforce is auto workers, technicians, software engineers, so on and so forth?” White collar, blue collar. And then which of these people are highly compensated versus more kind of low, middle tier. And then you could certainly quantify it, actually, what percentage for each company is in this quadrant of highly paid white-collar work, because that seems to be, at least based on this framework, the part at most —

Aswath: I think that part is gonna be easier to come up with than the cost part. You know what I’d love to see? I’d love to see how much it costs Anthropic to deliver that hour of Claude — in terms of how much comes from power, how much comes from water, how much comes from paying for the data, because some of the high-end stuff... I mean, and because Anthropic is now big enough. I mean, when they started, they just stole stuff, right? They just took stuff out of that — you know, acted like everything was in the public domain. In fact, about three months ago, I got a letter from Anthropic lawyers, or the lawyers involved in a lawsuit, saying that a lawsuit had been settled and that Anthropic had used 12 of my books in creating these bots that obviously they hadn’t paid for, and that I was entitled to a payment, and to put in — so I had to fill out the paperwork.

I usually don’t fill out the paperwork for these, but I did anyway because, you know. I think that those days are done. So one of the things is, if 80% of your cost is coming from having to pay for this high-end data — because now the data will not just be numerical data, it’ll be data on what does a consultant do, so as a McKinsey consultant, what do you do? I think that’s going to be an issue which will determine, as you get bigger, will those costs decrease? Will there be any economies of scale, or will this cost keep increasing? It’ll be like Spotify, right? When you sign a new subscriber, you pay by stream. You never get the benefits of scaling up a Spotify that Netflix does, because of the way you pay for content.

I’d love to be able to delve into the inside, and the odds are, as these companies start to report financials with more detail, the kind of data I’d like them to report will be very different than the kind of data that I was looking for in a traditional software company or manufacturing company. It’s about the costs and how they’re changing as you scale up, because that’s gonna be a big part of how this story evolves, is what percentage of that white-collar high-end workforce gets replaced will very much depend on how this cost structure changes over time. I mean, it’s interesting that OpenAI recently announced price cuts. Because that suggests that there’s a different business you can go after, right? Which was what the DeepSeek kind of opened up, which is 85% of AI stuff doesn’t require Nvidia chips and insane amounts of data. It’s really machine learning carried an extra step.

It’s been around with us for a long time, and the reaction that DeepSeek was saying, “Why are you guys paying for these expensive chips and insane amounts of data?” And OpenAI might be saying, “Look, maybe that’s a much better market to go after, a low-end AI market where we don’t charge these premium prices. The products we offer are less powerful.” But if you’re shopping in a grocery store, you don’t need high-end AI. You need very low-end AI to assist you on your grocery shopping.

Kai: Right.

Aswath: So that’s why I think this business is still evolving. We don’t know how it’ll play out. I think it’s great. But that’s a different market, you know.

Kai: Yeah, there’s totally different markets. There’s markets for cancer research and then markets for making a recipe on your phone. There’s also the question around IP monetization, right? Like, is it trade dress —

Aswath: And last week’s government decision on Claude actually opens up another question. Are these LLMs better off letting individual niche companies create bots, AI bots for their niches in medicine? Because then when there’s a problem, it’s just that one niche. Right now the problem is you are the LLM providing these products for hundreds of businesses. You could, after a day like Friday, say, “We’ve essentially cut off our entire lifeline, and these businesses are not even gonna adopt you anymore because they’re scared of the risk of adopting you and not having anything to do.” So there’s another tension that’s gonna play out here, where AI companies, the LLM companies, have to decide whether they will provide the engine for other people to create bots, which outsources that issue entirely, right? ‘Cause then if you create a bot that has problems, the government will come and shut that niche bot down rather than ChatGPT. So I think there are a whole host of unresolved issues here that are going to be played out in the public space, like it did — like you saw last week.

Kai: Yeah, I mean, super interesting, and obviously it impacts all the world’s largest companies. So as an investor, it’s kind of something you can’t ignore. So I wanna make sure we leave some time for the final section here on value investing. This is kind of my pet topic, I guess. I was raised a value investor. I worked at GMO, a value shop in Boston. And so you’ve been generally critical of traditional value investors, arguing that they’ve lost their edge over the past couple of decades due to becoming too — and I quote — “rigid, ritualistic, and righteous,” end quote. Can you maybe explain what you meant by this?

Aswath: No, actually, let’s step back. The conventional wisdom is value investing used to work well. What’s that based on? What is that conventional wisdom based on? There’s anecdotal evidence we offer, right? Look at Warren Buffett, look, Ben Graham, et cetera. But the overwhelming evidence for value investing, and this is ironic, came from academics. By doing what? Those graphs that Fama-French, in ‘92, showed that low price to book stocks did better than high price to book stocks. Low P/E stocks did better than... So people said, “Look, value investing works. Low P/E and low price to book stocks do better than —”

But the problem is that’s not value investing. I could create an ETF with low P/E stocks or low price to book stocks, which would have delivered the same excess returns with none of the additional baggage that value investing brought in, right? You’ve got to know the management, know they... You’ve got to look at moats. Let’s go back to the 20th century. If you look at active value investors, value investors like Brandes — I mean, you remember all the offshoots that went out around the country and created value investing houses. You hired analysts, they did research. The average value investor in the 20th century actually underperformed a value index fund in the ‘20s. So this legend of “it used to work, we used to make a lot of money,” is based either on anecdotal evidence or by looking at these generic studies of low P/E, low price to book stocks. And I think the problem is value investors drank too much of their own Kool-Aid.

They started to believe the fact that “we won for a century, we must be doing something right.” And I think that led them to this, “We’re the chosen ones. We do all the right things. We’re gonna continue to be rewarded for doing the right things.” That’s the righteous part. You think you’re entitled to an excess return, to earning more than other people because you do your homework, you’ve done your research, you read annual reports like you were taught to do, page one to page 400 if you’re looking at the prospectus. “I do my homework. I buy companies that trade at less than what you generate from assets in place.”

And that’s a key. Traditional value investing was not about buying stocks which traded at less than value, but trading at less than the value of assets in place. Sounds like a weird side story, but it’s basically, what do you have? Traditional value investors were very untrustworthy of value from future growth. You can see it all the way back to Ben Graham, which is, it’s all assets in place. You want to buy something for 50 cents on the dollar if you can find the dollar. So the righteousness came from their history, which then led them to a period where they’ve underperformed. And the problem is when you’re righteous and you underperform, you know who you blame. You never blame yourself. You don’t take responsibility. You blame the rest of the world. And for 20 years — and I’m not saying every value investor does — many old-time value investors have blamed the world, right? There’s this entire discourse around passive investing that I’m writing — a blog post called “Indexology” — emerging from this question of should these big trillion-dollar new companies be part of the S&P 500?

And that’s become part of a post on how people have essentially, especially a lot of active investors, many of the value investors, have made passive investing the villain of the piece as to why they’ve underperformed over the last 20 years. And you’ve heard the stories, which is, when you have 55, 60% of money going into index funds and ETFs, fund flow will basically mean that the largest cap companies will always have momentum behind them, and the momentum will then carry them further, even if the fundamentals don’t support it. So their argument is this is all passive investing’s fault because they’ve kind of broken the relationship between fundamentals and price. There are a whole set of errors in that logic, but the righteousness basically means you never take responsibility. It’s always somebody else’s fault, and if it’s always somebody else’s fault, you can never look inward to say, “What are we doing that is not working?” For instance, should this focus on book value be abandoned? A historic focus, right? Traditional value investing is very book value focused on the false belief that book value is somehow a proxy for liquidation value. I mean, I would wager that 98% of companies’ book value has almost no relationship with what you’d get if you liquidate the company.

The rigidity comes from the sets of rules. Value investing is very rule-driven. Why? To stop human beings from using judgment to override what the numbers should be. And that has two problems. One is, because you’re rigid, it essentially means you can never see nuance. And the second is you’ve set yourself up for disruption, right? ChatGPT can take that Ben Graham book and do what used to take value investors weeks or months to do and do it instantaneously. So that’s the rigid part, and essentially it’s becoming less and less attractive. The ritualistic part basically comes from the things that you’re supposed to do as a value investor.

Like what? You’re supposed to have read Ben Graham’s Security Analysis. It’s like being a Christian, and you’re supposed to have read the Bible, even though you haven’t read it, you claim to have. And when pushed, you do little pieces, you’ve memorized the Bible even though you haven’t read the rest of it. You’re supposed to go to Omaha once a year and listen to two octogenarians, at that time, sitting on a — I mean, I have an immense amount of admiration for Warren Buffett because he has a philosophy. He’s very clear about his philosophy, and he stays consistent with that philosophy. Hey, I love Charlie Munger even more, because the guy spoke his mind on everything if he was asked.

But I think the notion that you cannot be a successful investor unless you’ve read the Berkshire Hathaway annual reports every year for the last 30 years, and you’ve read Ben Graham’s Security Analysis, you can’t... I mean, those are rituals. They don’t make you a better — I mean, I think you can be a great value investor without ever having read Security Analysis and without believing that every word that comes out of Warren Buffett’s mouth is gospel. It’s not. He’s human. He makes — He has his own blind spots, and being aware of the blind spots doesn’t take away from his greatness. In fact, it adds to his greatness that he’s been able to overcome those blind spots. So the rigid, ritualistic, and righteous might sound like a harsh description, but unfortunately for a vast majority of investors who call themselves value investors, it’s a fair description of how they’ve operated, and it’s an explanation for why they have no way out of the hole they’re in unless they start to break bad habits.

Kai: So, on the positive note then, you have actually given some prescriptions for how these investors can break these bad habits and perhaps evolve their philosophy and strategies in a way that conforms better with the new economy and the way the world works today, and maybe does away with some over-reliance on kind of obsolete accounting rules. So I have a quote here from you from the same paper. You say, quote, “To rediscover itself, value investing needs to get over its discomfort with uncertainty and be more willing to define value broadly, to include not just countable and physical assets in place, but also investments in intangible and growth assets,” end quote. So I think that kind of summarizes a lot of the themes that we discussed today around trying to evaluate companies like SpaceX, right? That might be earlier in their life cycle. That may have more of its value in future growth as opposed to assets in place. That may be driven largely by stories and by narratives and be intangible intensive, say, around the IP, around reusable rockets. What would you add to this in terms of how the value investing school — and I know it’s a very heterogeneous group of individuals — but how they should be, what lessons should they be keeping and not just throwing away? What rules should be retired as they think about how they can revitalize or reinvigorate the class of investing that is value?

Aswath: The first is never say never, right? Too many value investors: “I will never buy SpaceX. I will never buy Tesla. I will...” At the right price, if you’re a true value investor, you should be willing to buy companies with bad corporate governance, companies with different voting shares, ‘cause it depends on the price. So the first is, recognize that any company can be a good investment at the right price. Conversely, any company can be a bad company at the wrong price. So this notion of good companies are good investments, let that go. That connection that you make between companies and investments has to be more nuanced.

Second, stop looking for conspiracies. Especially — I mean, I find so much of the value investing discourse kind of focused in on some accounting choice the company made and then blowing it up. I’ve seen 100 posts on how depreciation of AI CapEx is the big wild card. Come on, guys. If that’s what you’re focused on, you’re missing the big picture. Depreciation for a young growth company is not even making my top 20 list of worries. Could it affect my cash flows? Yes. Is it going to affect the earnings? Absolutely. But this is not the big driver of whether AI is going to make it or not.

And if you don’t trust people, that’s a perfectly appropriate factor to bring into your assessment. Because when we invest, it’s an article of faith, right? That the management can be trusted not to cook the books. But I think to distrust everybody because you’ve been cheated once means that you’re gonna be looking for accounting scandals where they don’t exist. You know, I won’t name names, but there are people who uncovered the Enron fiasco, and every company they look at is becoming another Enron. They focus in on a footnote that makes them uncomfortable, and they spin it, and this entire company is a scam. So I would say never say never, and don’t fall down that conspiracy hole, because once you do that, you’re gonna lose sight of the things that matter.

Kai: Thank you. So you’ve been super generous with your time. I have just one more question.

Aswath: Yep.

Kai: The standard closing question for you, which is, what’s one thing you believe about investing that most of your peers would disagree with?

Aswath: What do...? One thing I do believe about investing that... That it’s the errors that you make. I think the biggest problem in investing are the big mistakes you make that you refuse to acknowledge, right? I think that we’re all gonna make them. The nature of investing is you’re gonna be wrong a lot of the time. What gets you into trouble is when you’re wrong and you dig in and you double down, you triple down, you quadruple down. And a lot of other people might believe that as well, but I think that that’s one factor. The other is, I think the least persuasive evidence you can show me that you’re a successful active investor is your historical returns. ‘Cause I know how much luck and standard error and... I mean, I think that one of my favorite books is Mike Mauboussin’s book on separating luck from skill and how difficult it is to do in investing as opposed to basketball.

You can’t be a lucky basketball player and make 15 out of 20 three-pointers. It’s not gonna happen. But you can be a lucky investor and beat the market 15 years out of 20 all the time. So people think that if they back-tested something of their track record, they’ve got a conclusive way of making money. You know, it doesn’t work. In fact, I’ll close with a story from yesterday about this bettor on, I think, Kalshi or DraftKings or wherever he bet. He bet $1 million on Spain beating Cape Verde, which is this tiny... The payoff was only like eight percent. It was like you misinvested in it because it was such a slam dunk. Spain was expected to win. And this guy had made a lot of these high probability winning bets. He’d pick a team which had a 90% chance of winning, he’d put a lot of money on it, and walk away — until yesterday, where they tied, and he lost the $1 million. And it cuts to the heart of: there’s a subset of investment strategies where you will win most of the time, but when you lose, you wipe out ten years of returns.

And the problem with historical track records is you might be — I mean, one classic investment strategy where this happens, for instance, is to sell calls, out-of-the-money calls. Most of the time you’re going to sell the calls, collect the money, and you’re gonna keep it, right? It’s an out-of-the-money call. You say, “What’s wrong with it?” The problem is, all you need is a couple of stocks to explode out of the box, and there goes ten years of excess returns. So be wary of historical track records and investors claiming to beat the market because eliminating luck is really tough to do.

Kai: Thank you. That’s really wise. Much appreciated. Thank you. Well, thanks so much for coming on the podcast. I really appreciate it.

Aswath: Thank you for having me.