Deconstructing the AI Trade: Market Meltdown, Correction or Clean up?

0:00 Hi, welcome back. Well, it's been an interesting week, politically, economically, market-wise. And of course, last Thursday was quite a shock. It was a shock in terms of how much market went down and how much of the market going down was attributed to the AI trade. Now I've thought about the AI trade and I'm a little confused about what I'm trading. First, I'm assuming that this must be a long trade you're buying. And what example exactly am I buying? Am I buying Nvidia? Am I buying SpaceX? Or am I buying the hyperscalers, the the max the max seven?

0:38 So in this session, what I'd like to do is focus in on the AI trade because I think it might be time for us to create some clarity around what exactly the AI trade means. Now I'll I'll cut the cut the cut some slack for those people who still use the AI trade because for the last few years, it's been true that most AI companies, no matter what they do in this space, have tended to move together.

1:06 And when they move, they also move the market. So last Thursday, when the market was down, the Nasdaq was down 6%, the S&P 500 was down 4%. Of course, people fell back on the AI trade. Maybe that can explain why markets did what they did. While I understand the impulse, here's why I think it might be time to step away from lumping together all AI companies and treating them as a monolith and assuming they move together.

1:35 Think of all the players in the AI business. Let's start with companies like Nvidia. What do they do? They supply the chips. And those chips are acquired by the Alphabets, the Metas, the Microsoft, the hyperscalers, the Oracles, the Core Weaves. They use the data chips to build data centers. So you got AI builders. These these are the companies that are building the centers and the AI architecture. You have the suppliers, the chip companies, the power companies, the electrical equipment companies, and even the real estate companies involved in selling the builders the material they need to build the infrastructure.

2:16 And finally, all these data centers get leased out to the LLMs and Tropic, OpenAI, the companies that make the AI products. You've got AI builders who build the infrastructure, AI suppliers who supply the parts of that infrastructure, and AI businesses that essentially create the products and services. There's an external group, the AI customers, we'll we'll hold off on those. You know, it is true that all of these groups benefit when the AI market is viewed as really big. And for the last few years, that's all you needed. Big AI market, all of the companies move together.

2:52 But as AI ages, as a disruption, as it acquires more substance, I think you're going to see the interests of these companies diverge. And it might make more sense not to talk about the AI trade, but about the AI builder trade, the AI supplier trade, the AI product product company trade, or the AI customer trade. So let's start that process with a little bit of self-disclosure. I'm not naturally a curious person. I don't lie awake at night wondering what big macroeconomic problem I should try to come up with the answer with. I have other things on my mind.

3:30 So my interest in AI was accidental. It came about because I bought Nvidia in 2018 for and the reasons had nothing to do with AI. I wasn't that forward-looking. I bought Nvidia because it had dropped about 40 or 50% and I valued it as a really well-run chip company. And I found it undervalued. So I bought the company and your biggest successes are often accidental. The company did well and it outperformed the most of the companies in my portfolio, but was it a super winner?

4:05 Until November 30th of 2022. What happened that day? know, chat GPT got released and the AI buzz started and Nvidia took off like a rocket. And when it took off like like a rocket, I was faced with a problem. I mean, I you know, I've confessed that I'm an investor, which means I buy things that are undervalued, but the flip side of that is I have to sell things that are overvalued and something goes up 24 to that Nvidia did. I had to take a look at it and say, is it time for me to sell?

4:36 Now I won't bore you with my extended assessment, but in the process of valuing Nvidia, I had no choice, no choice but to confront what the AI business was going to look like and in pieces over the last three or four years, my understanding the business has become a little better. I'm not an AI expert. There are people who know a lot more about the AI business than I do. But I think I have a better grasp of what AI looks like as a business now.

5:03 You might also be aware, and this is a side part of the story, that over the last three years, I've sold Nvidia in four installments and all of my shares are gone. Have I left money in the table? Of course. Do I feel any regrets? Absolutely not. Because in the process, I've learned enough about AI that I can perhaps use those skills to buy something else in the space or perhaps Nvidia again at some future point in time.

5:31 So let me set up my argument. As I see it, there are four groups of companies in this space. You got the AI suppliers. These are the businesses that make the ingredients that that go into the AI architecture, the chips, the power, the electrical equipment, the real estate. You got the AI builders. These are the companies that build the infrastructure, build the data centers. You got the AI products and services companies. These are the companies that use the AI infrastructure to make AI agents, AI products.

6:02 Now when we think of AI products and services companies, of course, we think about the big LLMs, OpenAI, Tropic, SpaceX. But you could also throw in companies like Palantir or even Salesforce, which have used AI to create products and services or or augment their existing products and services to sell to their customers. And of course, you got the AI users. These are the the individuals or enterprises that the products or services are sold to. And why do they buy them? Presumably, if you're an enterprise, you buy them because it makes you more efficient, or maybe you buy them because everybody else is.

6:40 Now, the reason I like the structure is you can see the relationship between each group. The AI builders and suppliers, here's what happens. AI builders go out and build the infrastructure. Because it's long-term or it has a life over many years, it's treated as CAPEX. This isn't a conspiracy. It's basic accounting. So AI builders reported as CAPEX, and because it's CAPEX, it doesn't show up as an expense that year in your income statement, but it gets amortized over time.

7:10 Your cash flows will reflect the CAPEX as it happens. So the bottom line is when AI builders build their equipment, their earnings will go down. How much they will go down will depend on how how quickly they amortize, and their cash flows will be hit. Now, that CAPEX that AI builders put becomes the revenues for the AI suppliers. So when AI builders buy chips from NVIDIA to make their AI architecture, those chips are recorded as revenues by NVIDIA.

7:44 What does NVIDIA do then? It takes the cost of goods sold that it had in coming up with those chips, nets them out, shows it as profit. So AI suppliers, that same CAPEX that AI builders has now becomes revenues in income. So that's the linkage between builders and suppliers. Now, once builders build their infrastructure businesses, remember, it's been all negative, right? Lower the earnings, lower the cash flows. They now lease those infra, the AI infrastructure to the AI products and service companies.

8:18 The AI products and service companies might pay an annual lease, in which case it's an annual operating expense, or they may lock in their lease for 5 10 years, in which case you take the present value of those lease. It becomes more like debt. It's a financial expense. So the linkage between AI builders and AI products and services is this leasing agreement. And if you look at the AI products and service companies, those lease agreements create compute costs for them, either as operating expenses or amortized over time because they've taken it as a lease expense of financial expense.

8:52 Now what do they do with the infrastructure? They create AI agents, AI products, they sell them. And when they sell them to AI users, the custo, the ultimate users, they get recorded as revenues and net of the expenses, they become the profits for the company. Now what do AI users do? And this is a little out of the system. They use those AI agents presumably for because they, they make themselves more productive. It reduces their cost structure, it makes them more profitable.

9:22 Or it allows them to get more revenues because their products are are better. Or they're doing it because they don't do it and everybody else does it. Their competitors do it, they're going to be left in the dust for whatever reason. So they will spend, report the spending on AI products as an expense and hopefully they get enough benefits to cover their expense and more. So already you can see the structure of this process is going to create transactions where one side records it as a revenue, the other side records it as a cost.

9:54 But the timing of the payoffs is going to vary depending on which group you're in. The first group to get a payoff will be the suppliers. So very early in the AI process, we saw this. In video and chip companies and par, were the fir, first companies to see the re, the benefits almost immediately. Their revenues pick up, their earnings pick up. And they and their it's easy to understand those businesses. Initially the building business looks like looks pretty strange. They spend billions, in this case hundreds of billions of dollars in CAPEX. They have nothing to show yet. So all you see are the cash outflows.

10:37 And to the degree that you trust managers, you're hoping that they have a plan in place. So in the early phase of hope and hype, the suppliers have real substance. The builders have just this CAPEX, negative cash flow. The product companies are nascent businesses. They're still working on coming up with products. So they're pre-revenue often, of very small revenues, lots of expenses, R&D, product development, employee costs. They're basket case. They look like basket case companies. And the end users are usually just test users. This is still promise than product.

11:12 Then you get a second phase, which is the investment phase of AI. Your suppliers have now have businesses that are fairly established. They're hoping for continued growth, but that'll come about only if the builders keep expanding their AI architecture. But as the suppliers scale up, you're going to see growth rates start to drop even if they're doing well. On the builder side, some of that CAPEX that they've invested hopefully has come online. They start leasing it out. You're going to see the beginnings of a revenue stream from leasing the architecture. But it's going to be small relative investments. If you compute things like return invested capital, it's going to be a tiny number at this stage. You're saying what are you doing?

11:57 Is it even worth it? But it's still early in the process. The product companies, you're finally starting to introduce products. And it's small scale, but because it's small scale, as you get your revenues, the growth rate is going to be sky high. And those products now are sold to early adopters, experimenters, and in industries where you think AI is going to work the best. So in the investment phase, supplier growth rates start to come down, but they're making a lot of money.

12:27 Builders are starting to generate revenues, but on the AI architecture, they're more likely to be losing money or earning very low returns. The product companies are starting to use the architecture to create products which consumers, the customers they're going after, hopefully like. Then you get to the business building phase. By this point, your suppliers have well-established, well-formed models. What you see is pretty much what you're going to get, and the big challenge then becomes, can you keep scaling up given that you're a much bigger company?

13:02 The builders at this point should be able to lease out enough of their architecture that even if their returns on capital are not fully there, you can see where they're going. And if you are the market, you're going to see where they're going to end up with a return capital that matches their cost of capital, exceeds it, or at least make your best judgment. The product companies are now jocking for position, so they're trying to lock in more of the architecture, compute power as they prefer to call it, as a way of differentiating themselves from their competitors and giving themselves some kind of a moat for future growth.

13:40 And you use this now on the receiving end for sales pitches where the product companies, AI product companies are telling if you do this, amazing things are going to happen to you. And as you think about this process and you look at companies, you can see that depending on where you're in the process, the nature of companies in each group is going to change. So I'm going to pause right there and think about where AI is in the space. Clearly the hype and hype phase, I would say 2023, maybe into 24.

14:07 The investment phase, 2024-2025, it was amazing the speed with which it took off. But now I think in the business building phase, you have you know the the chip companies, the power companies, they have fully formed. They're still scaling up because the builders are still adding to the the AI architecture. But at least you're starting to see the num the earnings come from that from that architecture from leasing them up. The you know as you saw with the Entropic prospectus, the product companies are essentially locking in this compute power for five, six, eight, nine, 10 years and paying hundreds of billions of dollars to lock it up.

14:50 We're at the very beginning of the user phase. And I think that that's the part which is going to hold determine ultimately whether all of this is going to hold together. So with that structure in place, let's think about what companies across the groups have in common. First is they're all ultimately dependent on those on that customer group becoming larger, more lucrative, more monetized. Without it, everything else falls apart. So if the end market for AI products turns out to be small, the product companies are going to feel the first hit.

15:27 Then they're going to try to get out of their their agreements on on AI architecture compute power, which affects the builders who then are going to turn around and say, look, we're not building anymore data center. So you're going to see if the if the end market for AI products is small, you're going to see the ripple effect play through. If AI delivers on its promise as a amazing productivity tool where lots of people buy, all of the groups will benefit to different degrees.

15:56 In my view though, the AI architecture business, the AI supplier business, even though it was the first to be you know profitable is now I mean I this I may be getting ahead of its best days perhaps are behind it. It's if you look at history, this is the point where the power shifts to other groups in the space. Now the fact that all of these groups are connected and connected at the head, also might explain why you have so much circle of financing in this group.

16:31 Think of what Nvidia wants, right? It wants builders to be healthy, to be building more data centers. It's very concerned about the health of the builders, not because it cares about them, but because if they're not healthy, they're not going to buy Nvidia's chips. So if a CoreWeave gets into trouble, I wouldn't be surprised if an Nvidia stepped in and bailed them out. They won't view it as a bailout. It's a way of keeping your your your clients, the businesses that buy from you healthy enough to keep buying from you.

17:02 So the circle of financing is very much I think functional the fact that they're all interconnected. At the same time, there are competing interests as well, where what's good for one group is not good for the other. Let's take a simple example. If you're builders and I'm a shareholder in the hyperscalers, I'm worried about them building too much, right? Because they're spending my money, my shareholder money to spend, you know, to build hundreds of billions of dollars of data centers that might or might not be usable.

17:33 But if you're Nvidia, you want them to build those infrastructure. It's a mistake, but it 's a mistake you benefit from. So as you look at these transactions, you can see on the compute power, you know what's good for the building companies is they can get Anthropic to sign a 10-year contract for compute power because it locks in their revenues, makes them, if not riskless, lower risk to the builders. But from the LLM side, that 10-year contract locks you in. It creates debt before you're ready to take on debt.

18:08 So as you look at the actions in the space, you can already see we're at a point in time where things can affect one group positively and another group negatively. And I think that's part of the reason I'm make making the argument that we need to start differentiating across the groups. And I think the market is doing it as well. You're saying this is all in good. How do we even test this proposition? I'm going to try. This is what I'm going to do.

18:36 I'm going to start by looking at all publicly traded companies in the AI space. You saying why publicly traded? Because that's I need access not just to the the the financial data of the companies, but a market price of the companies. Now that is going to be a bit of an issue because in some of the groups that we talked about almost all of the companies are still privately owned. The AI product and services think of the big players, right? OpenAI and Anthropic are still private businesses.

19:07 SpaceX is public, but it's been public only for six months. So it is going to create a bit of a skew, but I have no choice. I'm going to look at public companies. Now to find the companies within each group, I would it would be it would be fine it would be great if I could go to their financial statement and they said percent of revenues from AI, but they don't do that yet. Or maybe they never will.

19:30 So here's what I had to do to find companies in each group. I had to go through new stories on who the biggest players were in each space. Now what does that imply? It also implies that their companies I will have missed. Either they're because they're small or below under the radar didn't make it into that new story. Third while I would love to tell you I found pure AI building companies, all they do is build data centers, that's not the way of the world.

19:57 Many of the companies in my sample in addition to doing AI stuff do other stuff as well. Take the big the mag the mag four that are big into AI, Amazon, Alphabet, Meta and Microsoft. These are the big the hyperscalers the building AI. But where do they get their money? They get them from existing businesses that are immense cash cows, right? Alphabet, Meta it's advertising. Amazon, it's it's their online retail slash logistics, you know, and cloud.

20:30 And Microsoft with both the cloud business and software. I can't disentangle those, so the AI effect can sometimes get buried by those other businesses. Finally, even with the companies I was able to find, you could argue that some of them are in more than one group. I'll give you an example, right? You take Meta, Alphabet, and Microsoft. They're builders of AI architecture. They're the hyperscalers. But they also have their own LLMs, right? Muse for Meta, Gemini for Alphabet, Copilot for Microsoft.

21:04 So they're in the product business as well. So some of these companies are in multiple places, and SpaceX is particularly tricky. I, you know, when I started, I thought Grok, it's an LLM. They're in the same space as OpenAI and Anthropic. But if you look at the revenue streams right now, they're behaving more like a hyperscaler in terms of building and leasing out rather than using it for themselves. With all those caveats in place, here are the companies I ended up with in each of my groupings.

21:34 I have about, you know, the AI chips, clouds, and supplies. Basically, I have about 10 companies. AI power, I have four. Next Era, Constellation, Bloom. I have AI real estate companies and construction companies. And I have AI electrical companies, right? And finally, I have some hyperscalers. Microsoft, Alphabet, Amazon, Meta, Oracle, CoreWeave. And for the product companies, I'm stretching. I had SpaceX. I threw that in there. Only public company among the big LLMs.

22:07 Palantir and Salesforce. Salesforce is Einstein. It's, you know, but you could argue that maybe even Salesforce shouldn't belong there yet. So it's kind of, you know, it's the AI product business is still building. Maybe next year, after OpenAI and Anthropic have gone public, the space will be more filled up. And here's what I did. I took these companies and I tracked them over time. At first started by looking at the at the years 2022, 2023, 2024 and 2025 and the market caps at the end of each year.

22:40 Why did I start in 2022? Remember November 30th, 2022 OpenAI gets launches ChatGPT to the public. AI kind of explodes as a buzz. I'm looking at the market cap of these four groups of companies over the four years. Collectively, let's start with the suppliers. I said they're the ones who will show the most substance early. Their collective market cap has gone from 2.3 trillion to 8.5 trillion. I know I know big chunk of that is Nvidia, but even without Nvidia, there's another two and a half to three trillion in there, right? Or more.

23:16 It's up 374% over the four years. You don't even need to do comparisons to the market say did of course it did well. The builders, which are the hyperscalers and with all the caveats I said about their other businesses, saw their market cap go from 4.3 billion to 7.8 billion. At the moment, the supplier business has a bigger market cap than the builder business. It's up about the builder business is still up about 180%. Amazing if you think about the fact that all the builders have for the most part to show for the last four years or for these four years is a lot of building. They don't have much in revenues, but the market seems to have cut them some slack. They're doing worse than the market, but the market is still pushing up their prices.

24:05 The product is still as I said, slim pickings. And if you look at the four years, they're up 364%, but of a much smaller base. We're not talking trillions, we're talking you know tens of billions. Collectively, you can see AI companies are up 249% over the period, which is much better than the market. Clearly, you know all three groups did well by and the group is done well. And it's been the AI buzz is clearly played out market cap.

24:39 Now let's take a look at 2026. Are things changing? Now the answer depends on what part of 2026 you look at. If you look at the first nine months, AI companies collectively were up 35%. See there the the the the bloom is still on the rose, but almost all of it came from the first two groups, chip and cloud and electric. You can also already see the divergence I talked about. In fact, in the first nine months of 2026, you can see that AI power, real estate, AI hyperscalers and AI products have all underperformed the rest of the market.

25:21 And then I focused in on the five weeks between the end of the third quarter and November 6th, last Friday. And now you can see the wheels starting to come off the bus, right? Because you can see the chip companies essentially give up not 22% of their gain, 26 minus 26% for electrical. But again, notice the divergence. The hyperscalers were down a lot less and real estate construction actually went up. The point I would make is we're starting to see their interests diverge much more significantly than they did in the previous four years.

25:59 In fact, year today, all AI companies collectively are up 16% better than the market. So even after last week's collapse, they're up. But almost all of it is coming from one business, the chip cloud business and perhaps real estate added in. The rest of AI is not looking as good relative to the market. So you can see that this is going to be exhibit one. Next time somebody asks me, what is the AI trade look like? It depends on what part of the AI business you are trading.

26:34 And I did also take a look at just last week, Monday through Friday, and I looked at the change in market cap per day. Because over the course of the week, you can see collectively and you know, in fact, just on Thursday, just that one day, AI companies, the the roughly 25 to 30 companies I have across the groups lost 1.8 trillion dollars in market cap that day. A big chunk of it came from the chip business, the chip and the chip and supplies business. But you can see that it's been quite an interesting week, and I'm using interesting in a very guarded sense for the AI business.

27:19 So where now? Now if you've been listening so far, expecting me to now tell you go buy AI, go sell AI, you're going to be deeply disappointed because I'm going to hit you with something that I am very fond of using when I think about big disruptions, like PC business, the internet. And it's what I call the big market delusion. What's a big market delusion? It's you start with the big market. Could be China, it could be the internet, it could be AI.

27:49 And you have a big market, people see the big market, they get greedy. And those people happen to be overconfident people, they start a business to try to take advantage of the big market. You with me so far? AI, big market, you start an AI business in any of the groups, right? Remember, you're overconfident. You find venture capitalists who happen to be also overconfident who they in picking winners, and you get a cluster of overconfident people looking at the AI market, the big market, and saying we'll win there.

28:22 The definition of overconfidence is you overestimate how much revenues you can get and how much value you have as a company. And this cluster is being repeated hundreds of times. Hundreds of people, or each of whom is overconfident doing this. And already you can see why I call this the big market delusion. Collectively, you can't be right. In other words, if I took your expected revenues across every company and added them all up, the market's not big enough to sustain all of you.

28:53 Will you eventually find that? Absolutely. It might take a year, it might take three years. What's going to trigger that recognition? I'm not sure. But when it comes, you know what's going to follow, right? There's going to be a cleanup phase. And it's going to Let's call it a correction if that's a word you use. As the market says, are we overreached? And pulls back. I think it's it's it's it's an interesting way of thinking of where we are in this space.

29:25 I think that the big market delusion kicked in big time in 2023, 2024, much of 2025. I'm not saying we're at the catalyst phase where we're going to correct. But I can almost guarantee you that there is a big market delusion in AI, that collectively companies are being overpriced because that's what markets do early in the process. So looks like I'm predicting a correction, right? But before you get crazy about that prediction, it will cause pain.

29:59 And that pain will be uneven. And that's the other thing. It not every AI company is going to get knocked on. Some might actually come out of this as superstars. And this big market delusion is not a problem. It's a feature, not a bug of big change. And yeah, I use an in others of you have heard me use this expression before was to ask people when they complain about the fact that, now how do we stop this from happening? This sounds crazy.

30:29 I said, do you want to live in a world run by actuaries? It'll be very realistic. You would never have an AI boom if actuaries ran the world. The truth is, and this is why I have been careful not to use the word bubble. Bubble suggests irrationality and something you want to avoid, right? If you want change, you have to live with the big market delusion, the chaos, the correction. And I think there will be one. But here's why I would make a big deal of my own prediction that there will be a correction. First, I'm not sure when it'll happen or in what form.

31:10 That's the problem. I mean, remember there were people who predicted the dot com correction in 1996 and they either went bankrupt before the correction actually came or the money they lost by staying out until the correction actually outweighed what they'd have lost if they just stayed in and gone through the correction. Part of the reason for that is when not quite sure what the catalyst will be for the correction. It could be some big event, a collection of small cuts. Nobody really knows.

31:45 The reason is the correction, the big market delusion itself is a behavioral phenomenon. It's born out of overconfidence and making an incorrect assessment of the size of the market because of that overconfidence, which means that to detect when it's going to happen, you're not going to find the answer in fundamentals. So you're going to probably have a better shot finding it looking at charts or technical indicators. Am I going to do that? No, because I'm not good at it. But perhaps you are, right?

32:16 So I think there's going to be a correction and if you are comfortable with your charting and technical analytical tools, who knows? You could call when that correction will happen. And assuming that that you see a correction on the horizon, what should you do? If you're an investor in long standing holds one of these companies that you think is going to get My advice is start selling portions because the uncertain you if you say it's all or nothing, you might never pull the trigger.

32:47 But every time there's a correction, you sell one quarter of your holdings. Far easier to act. And you give yourself at least the comfort level of saying I'm getting out slowly. And finally, when that correction kicks in, it's going to be dark. Markets are going to panic. That's what they do. Prices are going to get knocked down, not just for the companies that deserve it, but companies across the board. And if you've never bought an AI stock, that's going to be a chance to get in.

33:19 But to do that, you got to get a list of the companies in these spaces that you like right now. You might not like how they're priced, but you like them as companies. Make a judgment on what you think they're worth. Then I'll let you make your own judgment. I would value companies and come up with a with an estimated value. And wait. In fact, if you don't trust yourself to make the buy yourself, again, turn the control over. Make it a limit buy at a price. And if the market collapses, you'll be right there.

33:53 So as you look at the AI trade unravel potentially, we don't know. This could be it. This might not be it. What are the implications for Anthropic? I think it's actual scheduled offering date is the middle of November. We're getting awfully close. You know, you're saying what does it mean? You're already, you know, before last week, you already saw some leaks come out that said that led you to believe that they were trying to set the stage for an offering that they they announced they went back to the annualized revenue run rate, you know, 80 billion was in October, which is, you know, was viewed as good news by some people, bad news by others.

34:32 But ultimately, an IPO, you price companies. You don't value them. What does that mean? It's It's and supply. It's determining what price you can clear the market at. It's a mood and momentum game. And last week, the mood turned a little darker. And if that continues, then I can almost guarantee you bankers are going to watch out for themselves and they're going to price the company to reflect that dark move. Remember, bankers guarantee a price. They don't have the capital to back up a trillion dollar offering if it's massively overpriced.

35:06 So I think the pressure is going to build for them to lower their 1.5 to 2 trillion dollar original estimate down to a lower number. Now, you think what about the valuation front? Last week, I valued In fact, earlier in the week, the the day of the election, before we knew the election results, we were around 1.45 trillion. You think so what's changed? A lot, right? The elections happened and one of the the byproducts of the elections is it looks like data centers are going to get more difficult to build.

35:40 With those data centers being built in Texas, it might be that the contracts are already set. Those data centers get built, but new data centers will get more difficult. How does that play out in valuation? Remember that sales to capital which basically tells you how much I'd assume that compute power would ease up, that it'd be easier to get. Maybe that assumption is a bad one. So I'm going to change that from $3 of revenues for every dollar of capital to 2.5 of revenues for dollar in capital.

36:09 I'm requiring to invest more. It also looks like the regulatory environment is going to get much more hostile to AI. What's that going to do? That drag that margin switch I originally estimated to be 5% let's push it up to 7 and 1/2. I'm not going crazy here, right? Changing the sales to cap increasing the drag with those two changes put in, my estimated equity value drops to 1.25 trillion. You see, how can it change that much?

36:40 When you value a young growth company, the tail's going to wag the dog. The number is going to swing around in both directions. Numbers come out. So, what's the bottom line here? I think that I started off with that point about the AI trade. I think we need to move away from it to maybe the AI builder trade, the AI supplier trade, the AI product trade, the AI customer trade because each group has its own characteristics, things that are good for one group and bad for the other. And during a crisis, there might be some groups that are affected more than others.

37:15 And I also think it's, you know, it's time that we hold off on making assessments about AI that are so broad that we refuse to even look at the facts. So if you start out with the presumption that AI will never work, that this is just a scam, your mind is already made up. There's, you know, you will never buy an AI company. And that's going to cost you. But on the other extreme, an AI is the greatest thing ever. You're dazzled by the technology.

37:44 You think it's a win. You're again going to stop asking questions. Notice the common theme. The key to good investing is to continue to ask questions. And if your views get too deeply ingrained, you cannot move from them. You will stop asking questions. I hope you found this session useful and I thank you very much for listening.