The AI Investment Playbook: Who Wins Next & Where the Money Goes

Hello and welcome back to the Market Maker podcast. And this week we're going to talk about the business model of AI. So, what does that mean? Well, we're going to split this episode up with some of the major players in the business of artificial intelligence trying to better understand what's going on, what strategies these players should pursue, and ultimately who might win. But whenever you talk about the winners, you've also got to also talk about the losers. We're going to approach the AI question not from a technical perspective. That'll be well and above our pay grade, but from a business model perspective. So, we're going to focus on four company types. Those four being the upstarts, the one everyone talks about, the Open AIs, the Anthropics. Then the hyperscalers, Amazon, Meta, Microsoft, Alphabet. Then third, picks and shovels. Fourth, the product layer. If you're returning to the show having listened to an episode recently and you enjoyed it, don't forget to subscribe. Weekly episodes coming at you. And also we we love to see your comments. So, if you do have a question or a comment as we go through the show, drop it on the episode as you're listening. So, Stephen, how are you and where do you even begin with the business model of AI? >> Yeah, taking on the business model of AI in a 40-minute podcast is pretty ambitious. I think we was we were chatting off air and you were saying, "Look, are we are we even qualified to talk about the business of AI? Are we even qualified to talk about AI?" And because every second or third podcast that you'll probably have in your feed mentions AI in some way, shape, or form, is this not just a crowded market? But hopefully what we're going to do is with my business strategy hat on and try we're going to try and understand some of the market dynamics, some of the business positioning, where like some really, really first principles stuff about business models using AI as the context. And what would be really really helpful, you mentioned we get a lot of comments, which is absolutely fantastic, especially on YouTube channel. We're not experts at AI. We're experts at business and experts at finance. We're not experts at AI. So, if there's anything in this podcast that you want to write a comment and say, "Hey, you know, this is actually what this means." That is all good for the community. We are did an episode last week on the Seattle Seahawks and an American listener chimed in and said, "I love hearing two Brits talk about our national sport." You know, she was pretty impressed, but you know, we we went a little bit beyond our beyond our safety now, our comfort zone there. >> Okay, so what what's a a reasonable starting point then for looking at this from a a business case study? >> Okay, before we get on to the four different groups, the upstarts, the hyperscalers, the picks and shovels, and the product layer, I think it's worth just thinking about the markets. And there's two ways to make money in a market, right? As an investor and as a company. The first is to pick a market that is growing very very quickly. Pick a market where the total addressable market is growing double digits every year. So, the pie is effectively getting bigger. So, just by being present in that market, you hopefully are growing commensurate with the size of the pie. That's number one. The second is being extremely smart about where in the value creation chain your investment or your company lies. This is assuming that the pie isn't getting that much bigger, but you are pointing yourself in the most accretive, profitable segment of that pie. In that example, where the pie isn't getting much bigger, there is only a finite amount of profit within that value chain, right? So, it's a little bit more dog-eat-dog. It's a little bit more let's try and get it squeeze margins of of different players within within the value chain. It's not as nice a place to be. So, the first thing we need to discuss or think about is to what extent is the AI pie growing, right? And the AI pie, what I mean by that is customers, individuals and businesses, that are willing to pay significant and growing amounts of money for an AI product, an AI product layer. If that pie is not growing, if that total addressable market is not growing, then you will see this distribution of the existing pie across these different hyperscalers to upstarts, upstarts to picks and shovels, picks and shovels to product layers, and it's just what is what is in vogue at that particular time, and what, you know, where are the bulls and where are the bears. So, firstly, is the total addressable market of AI growing? Well, yes, but maybe not to the extent that we thought a couple of years ago. Obviously, if you take SpaceX's prospectus that we discussed a few weeks ago on the podcast, it mentioned a 26.5 trillion-dollar total addressable market in AI and data services. So, it's talking its own book and saying, "Look, everything is basically AI, and therefore everything is part of the total addressable market." >> Can I Can I just ask, how do you So, total addressable market, how do you get like almost the price discovery for TAM? Like, is there like a a party that authenticates it? Like, with the UK budget and its fiscal spending, it's like you have the Office of Budget Responsibility, the OBR, and they kind of sign it off. Yeah, this is legit. And this sort of these plans, these budgets are realistic. So, when someone says a TAM, like, how do How do you get the agreement of the baseline of what is the accepted TAM for a market like AI? >> Yeah, if I was an analyst looking at this, I'd probably try and get the estimates of the total addressable market from a number of probably management consultants, uh maybe accountancy firms that put out this type of research in order to to generate more business. And one of these pieces of work, one of these kind of meta studies of total addressable market for AI, has put AI total addressable market at 539 billion in 2026, rising to about 1.3 trillion by the early 2030s. So, just bear that in mind, bear those numbers in in mind, and bear that growth of 20 to 30% a year in mind when we go through some of the numbers that we're going to talk about regarding CapEx and things like that. >> When you look back as history for a guide, and there's been other technological shifts that we've had in human modern history, what was the TAM at those points, and how realistic were the forecasting to the actual realization of that marketplace in reality? >> Yeah, it's a really interesting question, and that you There's so much that's written about where How does AI compare to {dot} {dot} {dot} bubble, right? Or {dot} {dot} {dot} CapEx build-out. And lots of people talk about the dot-com bubble and you know, the irrational exuberance of the late 1990s where loads and loads and loads of capital expenditure, loads and loads of investment upwards of 500 billion dollars invested in laying 80 million miles of fiber optic cable got pushed got pushed through this very very bull market where everyone was getting a little bit over their skis and the concept of total addressable market probably went out the window because as you know and from a markets perspective there comes a time where things let get dislocated from their intrinsic value and their intrinsic potential. So obviously we know about the dot-com boom and then the the dot-com crash. But what that opened it's what that opened up to was all of this infrastructure that eventually laid the groundwork for some of the most valuable companies of all time. So when I'm talking about this pie and we're going to go on and talk about these four different classes of companies. When I talk about this pie all loads of money is being spent on the infrastructure to service this pie which is the total addressable market. It's just still not very clear which of these different players has got the right strategy. >> All right. Well look that that takes us on then to probably the most I guess the first category is the most sexy one because it's the one that we as use and users probably get the most use of on a day-to-day workflow basis. It gets the most news coverage. It's seems to be groundbreaking in terms of on the technological front the frontier nature of it. So should we start with the upstarts then? Is that the most logical place to start? >> Yeah, I'm going to start with the upstarts and when I say upstarts they're probably not upstarts anymore the likes of Anthropic and Open trillion-dollar companies, but they are the great disruptors and the pure play AI labs relative to the hyperscalers, the existing companies that do a lot of other things, and maybe even the product layer as well. So, these two companies, obviously they've had an unbelievable move from nothing to a lot in terms of revenue, in terms of brand recognition, in terms of business model size, etc. And, you know, if we had had this conversation 6 months ago, we would have probably still had question marks about Open AI and its quote-unquote business model. Uh but, we'd be pretty bullish on Anthropic as it's focusing more on the enterprise piece. Obviously, what happened a couple of weeks ago is Kimi came on board, right? You know, you heard of the Kimi K3? Tell me about the Kimi K3. >> [laughter] >> So, this was released by Moonshot AI, Chinese AI lab, in 2026. Kimi K3, 2.8 trillion dollar per trillion parameter open weight mixture of experts model. Lots of terminology there. But, what this is >> [laughter] >> The most important thing is that there are models that are coming on board not from Open AI and not from Anthropic that are getting to the getting to as good as the best models that are available from Anthropic and Open AI, except two things. They're a heck of a lot cheaper and they are what's called open weight models, which basically means that you can put them on your own laptop or you can put your own set on your own servers. And, what's been really interesting to see is when there is a market that is what we would call a duopoly, right? Where there are only two companies that are dominating this market. They can set prices, right? The consumers or the businesses that use these models become price takers. They just have to accept, all right, it's going to cost this much per token, you know, token cost, etc. Whereas when the market opens up as it has done over the last couple of years, last year with Deep Seek, this year with Kimmy K3, and similar type models coming out from the likes of Alibaba, it suddenly removed that pricing power from the likes of OpenAI and and Anthropic, and we're getting companies like DoorDash, we're getting companies like Airbnb, we're getting companies like Shopify saying, "Hey, hey, hey, I'm going to I'm going to keep my Anthropic for like the really, really intense, hardcore, technical, high computational, difficult stuff, but there's loads of what DoorDash co-founder Andy Fang calls lower-level work, which he is distributing to what was then Kimmy K2.6. >> What's the data security aspect of that then? Because if I'm a user of Airbnb or DoorDash, and off goes my inputs into the world of a of a Eastern-orientated company, which how does that how does that kind of fit within I mean, obviously I I saw an Apple advert during the World Cup again and again and again, and they've definitely lent into this this the security aspect of it as their kind of key USP. How much of it is customers will chase or companies the the priority to cost efficiency as opposed to data security. It's such a good question and we can talk about this in the con in the context of business models in a second. I think one of the advantages of these models is that they are open weight, which means that they are not they are not held on a cloud server where your data is being given to or is being lent to an open AI or an Anthropic. That's a closed weight model. Open weight model, you buy you buy the weights of that particular model and you put it on your computer or you put it on your servers. I think Airbnb said they are using a limited number of China origin models and they are hosting them on company managed servers servers and fine-tuned for specific uses. So actually that kind of risk of data transfer is not quite as high with these open weight models. >> Okay. Okay, I'm not quick completely convinced me yet but I like the way Was that Was that the language from the company? They were like we're only using a limited number of the Chinese origin models. Well, a limited takes a limited amount I'm afraid. >> But this is it. Yeah, you're absolutely right. And this goes back this is like absolutely pure play business model 101 and there's a brilliant book and if you haven't if listeners haven't read The Innovator's Dilemma by Clayton Christensen. It's it's it's what basically every single Silicon Valley CEO has read and it is absolutely brilliant and he gives this wonderful example about big companies trying to focus their strategies and his conclusion is you can't do two strategies at once, right? So he gives the example of HP, Hewlett-Packard. Bearing in mind this book is 30 years old, he was he was analyzing uh, printers printers, right? So you have two different types of printers back in the late '80s. You have inkjet printers and laser printers. Inkjet printers, much lower cost, lower quality, but should be on everyone's filing cabinet, right? It's the the consumer product. Laser printers, there are specific use cases. It's much higher quality. You'd probably only have one in an office, but you might need that one. So, what HP did is it split out its company, or it split out the laser printer division from its inkjet division, and allowed them to compete, right? They didn't try to pursue two strategies. They created two independent entities, and it was hugely successful because the laser guys didn't have to go, "Hey, we're trying to satisfy all people at all times." And the inkjet guys didn't have to pretend that they were anything but a lower cost, easy-to-use printing uh, printer. And this is such an instructive example for the likes of Anthropic and uh, and and Open AI. You need to pick You need to pick a strategy and back it, right? So, it it is easily possible that both the open weight, you know, much cheaper, maybe always slightly less quality, less customizable models are out there, and that's fine, and that's good for lots of different use cases. But it's also possible that an Anthropic can still be a trillion-dollar company only kind of focusing on the higher-end enterprise market and not trying to dilute its brand by going after all sorts of different target markets. Which maybe is what Open AI is trying to do. So, it's a really interesting time where you've got to pick your strategy and you've got to stick to it. >> One thing I did see was what looked like to me products coming out of OpenAI. So, is that a good thing? You think innovation utilizing the expertise knowledge that they have, or is that a bad thing? And as someone looking at strategy, you're going, "Hmm, they have not picked their race yet. They're still exploring, which in itself is kind of negative news, even though the products itself might be quite good." >> Yeah, OpenAI is is stuck in a strategic no-man's-land, right? I think Anthropic has made sure that it become It's done a lot of work to be the trusted enterprise-level, high-quality, security-focused, really deep integrations with large enterprise clients, lots of customizations, etc., and is starting to get, obviously, its its run rate revenue is growing very quickly, and it's starting to get that embedded stickiness. That's what you want, right? You want big contracts that are just so hard to untangle. This is like the kind of Salesforce model of of old. Anthropic's kind of getting there. Simply put, OpenAI has really good models, doesn't have any good products, right? It tried to launch its Sora AI video platform, if you remember. Uh it closed down a few months ago. I think it was it was costing about $15 million a day, well, between 5 and $15 million a day, and generating almost no revenue, right? It's been talking a lot about uh hardware, and we discussed on the podcast a few months ago the the Jony Ive, ex-Apple iPhone designer, coming on board in a a bit of a Sam Altman lovein. A product's been launched. A code is called the Codex Micro, which is actually not really an OpenAI product. It's an OpenAI kind of partnership uh with a with a with another company called Work Louder. And, you know, it's a little niche product for people that are using OpenAI Codex, their coding tool. What might move the dial one way or the other is their first mass hardware product, which looks like it's going to be some kind of Alexa-style microphoney speaker thing, which again is not going to get my heart racing particularly. So, you know, you've got to be thinking if you're OpenAI, "My gosh, I'm in a no-man's-land here. You know, my advertising hasn't really worked or hasn't really come online. My products that I'm putting out there aren't really working or aren't really kind of hitting the numbers they need to hit." So, super, super interesting. >> If I was an entrepreneur, and let's say I'm thinking about, you know, like you said, I'm thinking about the TAM, and I'm thinking about where to go, where do I sit within that? If I'm thinking along these lines here, like with OpenAI, first-mover advantage, is there any statistical evidence about the success of being first mover, and is there any correlation as to sector or area or something like that where it does prove to be good to be first, and in other areas not? Cuz it would strike me as technology changes so rapidly, and AI has compounded that compared to previous technological points of difference. Is that what's making it particularly challenging for OpenAI? >> Yeah, it's a really interesting one. So, again, in the book The Innovator's Dilemma, there there was a good study that basically said there is no enduring advantage of being a first mover. And OpenAI is seeing that. And we'll see And we've seen that time and time again, especially in tech products, right? Because whether you go by Moore's law or whether you go by some other, you know, tech um build-out uh principle, basically technology gets better, right? >> [laughter] >> It's kind of what technology does. So, if you're the first mover, you have created potentially a new category that lots and lots of other players are going to swim into. You've proven a thing, whether it's a technology, whether it's a market, and then you're just going to get other players coming in. It's exactly what Open AI had when Gemini and your Google's Gemini came in, started a little bit rubbish, got a lot better. There are There are definitely markets where being a first mover is advantageous. Maybe things like gold prospecting >> [laughter] >> and getting the rights to you know, new oil licenses and and things like that, where as soon as you got the thing, you've got the monopoly. But certainly in technology, being a first mover is probably not where you want to be. >> I feel like I can steal intelligence by pinching staff. I cannot Well, actually, I say that. You could just intervene in Venezuela or do something like that and take the oil, but I won't go down that road. >> [laughter] >> But um let's move on then. Let's talk a little bit about then the other side of this, those mature big massive tech companies, the hyperscalers, who like you said aren't so super focused on just being the frontier lab. They have many other tentacles ongoing that make the organization. So, how do you unpack that one, the hyperscalers? >> Yeah, so these are the likes of Alpha Alphabet, Meta, Amazon, Microsoft. There are others, but I'm going to talk about these four in particular. I'm also going to talk about Apple as well. And I've unfortunately had a shocker with my notes, which I've given you the answer to the quiz again. Did you look at it, Ant, or are you going to be honest this time, or cuz you got the quiz right last time. >> I got the quiz right and that right and I had no visibility at all. So, I need I'm going to I'm going to take full credit for that. I have seen this one. So, maybe you could have like a dramatic second pause when you say each name. Let the Let the listener dwell on it for a moment. >> Okay, so my quiz. Year-to-date share price share price performance of the following companies. I want you to rank them, audience, from most successful, most up, to least successful, most down. Year-to-date share price performance. Amazon, Microsoft, Meta, Alphabet, and Apple. >> Does it make any difference? Does it make any difference in your notes? You said amazing instead of Amazon. >> I know, it's weird, isn't it? I I I do do that. That's a very very strange >> Amazing. Wow, Jeff Jeff would be very pleased with that description. >> Yeah, exactly. My amazing Amazon. >> [laughter] >> My brain-computer interface is really working. >> I I I I think before people think of their mental answer to this, it reminds me of when I was little and I grew up on the seaside. So, on the seaside, there's always arcades. And anyone who's been to the arcade will remember back in the day, there used to be those 10p machines with the horse racing. And you put a 10p in and you back one of six horses. And then the first horse would bolt out and you'd go, "Oh my god, we're definitely not going to win." This is kind of what I feel like the hyperscalers have been like over the past two two and a half years. >> Yeah, it's a very very interesting analogy. There's so much to unpack here and it's it's both good and bad and strategically complex as well. So, the answer to the quiz. So, worst performer, Microsoft down 18% has had an absolutely shocking run over the last year. Lots of different reasons. Meta down 9%. Amazon, also known as amazing, up 2.2%. Alphabet up 4%, although obviously recently took a big slide. Apple up 24% year-to-date. The OG, it is as of the recording of this podcast, the most valuable company in the world once again. And it's barely touched AI. >> When when I saw your notes on this, I was like, hang up hang about here. Whenever anyone puts a statistic in front of your nose, you've got to go, well, hang about, I need some context here. And I was having a look at what Apple was trading, let's say beginning of 2023. It was at about just sub 150, 150, let's call it. It's now trading at about 340. Like 150 to 350. Uh Google on the other hand, who's one of the uh not up anywhere near as much as Apple this year, was trading at sub 100 and it got up to 400 before the recent sell-off. >> You're absolutely right. And it it and a lot it's it's really interesting to look at the market at the moment, and I'm sure that you and Piers discussed this. You know, there's so much bearish sentiment out there, but we're still only a few points off like record highs, right? >> [laughter] >> It's a really weird space to be in. I'm sure you've seen this before. >> Yeah, it's nothing just just keep calm, carry on. It's fine. >> Keep spending. >> Keep calm. Yeah, keep spending. Yeah, don't sell it all. >> Uh yeah. >> Um so, uh little bit on the hyperscalers. So, the big four hyperscalers from, you know, in the in our analysis, Microsoft, Amazon, Google, and Meta, spent about 400 plus billion on CapEx, capital expenditure, for this AI infrastructure build-out. In 2025, it's looking like it's going to be closer to 750 billion in 2026. Goldman Sachs estimates that the total hyperscaler capex will exceed $5 trillion between 2025 and 2030. Now, from a from a strategic perspective, it's really, really important to understand the fallacy of sunk cost, but also the concept of being in a strategic no man's land. So, when if you are playing this game, you play it to the death, right? There is not a lot of value to start the capex opening the capex spigot for artificial intelligence build-out, and then realize that you've got it wrong about 18 months later, and then you spend $20 billion, and you haven't really got anything to show for it because the rate of progress is so quick. >> What about Zuckerberg and the metaverse? Cuz he was all in, and then he backed out. So, is he the exception to that rule of when it has actually worked to remarkable success? Cuz the company looked like it was on its deathbed at one point, and then it's had the probably the biggest outperformance out of all the hyperscalers in the last 2-3 years. >> Yeah, it's a really interesting one. Obviously, metaverse came with the fact that there was no market for it, right? And it seemed like a little bit of a fever dream from the Zuck. And by the way, big companies pursuing exciting new things, they should be doing this. It was a strategic misstep, but there is general consensus that AI is obviously a real thing, and there is a huge total addressable market, and that total addressable market is getting bigger. And if you start playing the hyperscaler game, you really, really don't want to be the one that's starting to slow down. And you can always see this with Microsoft, right? The kind of slight underperformance of its co-pilots. You know, they're still spending billions and billions of dollars. There is a fallacy called the sunk cost fallacy, which you know, throwing good money after bad, doubling down even though you know you're onto a loser. Any gambler will understand the concept of sunk cost. Um, but in this environment, it's a very very difficult train to get off. Because if you try to get off it, you're left with you're almost left with nothing. Right? Whereas if you stay on, yes, you keep on spending, but there might be a pot of gold at the end of the rainbow. >> What was interesting is we had Google's earnings, I think on the 21st, 22nd, and their CFO, I think it was, I mean, they raised their CapEx forecast, their outlook going forward. And then when questioned, they were like, "No, we're going to keep on spending." And what was interesting particularly with Alphabet was it was their first quarter of negative cash flow ever. I mean, how how is that why markets then are having this kind of questionable moment about the how tangible we are at this point of the spend when you start to see a signal like that go off. I'm looking at a chart of the free cash of of cash flow for Alphabet, and it just goes up and up and up and up and up. And even in like two quarters ago, it was at record levels. It dropped quite substantially last quarter, and now it's negative first time in its history. >> Yeah, it's a really interesting one. If I as an investor, the the biggest difference between the hyperscalers and the the the the startups, the upstarts, is that the hyperscalers have unbelievably good business models, right? And they are only spending their free cash flow. Obviously, they are raising money as well because it's an efficient cost of capital. But, if they did decide to lower the rate of acceleration, which they you know, there's kind of hints that they're starting to in terms of this big AI infrastructure buildout, the the free cash flow would just come back, right? So, it's not So, it's not as if this free cash flow pre-CapEx has gone anywhere. It's just that it's all going towards this CapEx buildout. Now, investors, what investors have to judge, and this is super super important, what investors have to figure out is whether my dollar of free cash flow is going to be better spent on CapEx relative to dividends, right? Or to share buybacks. And for the last two or three years, the general answer has been, yes, keep putting our free cash flow into CapEx, because we care about the future and the future should be bright. The wobbles recently are just that kind of that tempering of those animal spirits and just going, hey, wait a second. All right, we're not quite seeing the ROI yet. We're starting to be a little bit more circumspect, and we are going to possibly overreact a little bit to CapEx plans in a way that a year ago maybe they would have underreacted to it. >> So, talking of cash flows then, let's talk about the ultimate cash cow out of the hyperscalers. So, let's talk a little bit about Apple and and how are they uniquely positioned? You mentioned year-to-date, they're up what? Almost 25%, and you're looking at Microsoft down almost the mirror image of that, down 25%. So, huge differential. I mean, it used to be invest in Mag 7. It's a win-win scenario. You know, just the AI theme of the moment 18 months ago was just getting to hyperscalers, you're going to be in the good. Now, there's a huge disparity between them. So, what what's put Apple in a unique situation to thrive in this particular moment we're in? >> Yeah, and and you you you you're absolutely right to kind of take a slightly broader perspective and instead of looking at year-to-date, look at the last 3 years. It is one of those things that Apple is now considered to be the bluest of blue-chip stocks, right? There is no concern about its credit rating, there is no concern about the visibility of its cash flows, there's it is almost as stable as you can be for a {quote} unquote technology company, and therefore the lows are not as low and the highs are not as high. It underperformed relative to the hyperscalers in the last couple of years, it's outperforming now, but but from a you know, much lower volatility perspective. And it just seems like they're playing a different game, but by no means the wrong game. It might be It might be that these hyperscalers end up succeeding massively and anyone that's invested in them in the last few years has done pretty well. But Apple's kind of said, "Look, all right, we're not going to play the data center build-out game. We're not going to play the large language model lab game. You know, we're going to We're going to do what we do best. We continue to build great hardware that 2.2 billion users use." Yeah, quarter of the world's population use have an have an iPhone or a or a Mac. And you know, these are the devices by which all of this stuff is going to end up in the hands of the consumer. So, in the same way as they charge 20 plus billion dollars of 100% gross margin to Alphabet to get Google on the home screen of Safari, they're going to do the same and they are doing the same for these large language models. They are going into AI in a in a in a smaller way with Apple Intelligence. And basically, they're kind of triaging. If there is something simple, a a simple AI request through Siri or whatever it might be, then it can be dealt with with the relatively cheap basic models that that has been created in-house. If complex, then we will utilize an Anthropic or utilize an OpenAI, although Apple and OpenAI aren't best friends at the moment. [laughter] Apple's suing the hell out of OpenAI. So, it just seems like Apple are playing a different game, and they're playing it extremely well. And when things start looking a little bit choppy, a little bit volatile, a little bit kind of squeaky bum time, as we say here in the UK, what better place than to go to Apple, which every couple of years sells, you know, you'll upgrade your phone, you'll upgrade your laptop, and that's it. The show goes on. >> It's interesting then going through like Gemini and the whole Google ecosystem and its enterprise value there for simplicity's sake, and then the logistical side and AWS beast that's Amazon, then Apple here that you've just described. This actually feels like there's actually quite a bit of diversification amongst the hyperscalers in terms of their own business pursuits and their their preferred model outside of just this one-dimensional AI play. >> You're absolutely right. Uh totally. And therefore, either you pick you pick an index or you pick an ETF that covers them all or you pick your one that you think's got the best strategy. >> Oh, I love it when people say this, and they're absolutely right, folks. You should If you're going to invest money, and this is a this is not investment advice, don't sweat about it. Don't try to be a hero. Yeah, they they you just get your ETF of preferred choice, sit back and relax, and I'll see you in 20 years rather than uh pick the winners. >> It's a little bit like if you if you do if you do finance somewhat for a living, and you talk a bit about this stuff, a lot of people end up asking your advice, right? And obviously, the stock advice to anyone is stick it in indexes, stick it in ETFs. Boring, S&P 500, you're going to be fine. I don't listen to that advice. >> [laughter] >> I love putting it in single stocks and seeing how I do. I've not done very well recently, to be fair. >> Yeah, yeah. How's How's How's uh Reddick coming on these days? >> Reddick's coming on fine. I'll tell you what, my uh my my SK Hynix is not doing particularly well, but we can talk about that in a second. >> All right. Well, look, let's let's move over. We got two more areas to cover before we conclude. So, the third area is picks and shovels. So, that that's a that's a point of terminology that might not everyone has heard of before, even though it's become quite common within this AI conversation. So, what's picks and shovels? >> Yeah, picks and shovels are the the kind of back-end infrastructure build out that is powering the AI revolution. We don't interact as consumers or even as enterprise buyers. We don't really interact with the semiconductors companies, with the chip companies, with the memory companies, with the data center provider builders, the energy companies that are providing gigawatts of energy to these big data centers. But, they're all there. And what's really important about the picks and shovels, and I'll pick on a couple of very very quick examples, what's really important is this is going back to the beginning where we spoke about this concept of the pie, right? And you can see over the last couple of years that money and hype and attention has been directed to different parts of the pie at different times, right? So, yes, OpenAI and Anthropic have got billions of dollars of investment at ever high valuations. Yes, the share prices of the hyperscalers have gone through the roof and then started to tail off. But obviously over the last year, the money that is allocated to this pie has been going more and more into the back-end infrastructure buildout. So, if you had invested in a Micron or if you had invested in a SanDisk or an SK Hynix or one of these memory or chip providers or even an Intel over the last period of time, these are not set Historically, these are not sexy companies, right? But they provided they were at one particular time the bottleneck of the AI buildout. So, the eye of money went towards these companies. Now, the big question, the big big big big question, and we'll talk about the sell-off in a second, the big question is is this pie going to increase, right? Cuz if the pie is staying the same size, you get a load of money invested in the AI infrastructure buildout. And then if the pie isn't going to grow, then that just stops, right? Demand for these for demands for these picks and shovels, these data center buildouts, these uh memory chips, whatever it might be, they stop, right? Because the pie is not getting any bigger. It is only if the pie continues to grow really really really quite quickly that there's enough money to keep the lights on and to keep the share prices booming not only at the hyperscalers, not only at the OpenAI and Anthropic, but also at these picks and shovel companies that are doing all of the messy buildout. >> So, one thing though, it's recent news over several months. You have Microsoft investing in OpenAI, and then OpenAI's got some sort of deals with Nvidia, and then Nvidia's got deals with Alphabet. If everyone's dependent on everyone else, isn't it within their agenda to keep the show going and keep pushing the TAM infinitely higher? Because if that fails, they they it's their incentive to cut deals in order to inflate the numbers. I mean, one would think that that's going to come at a a critical mass where literally the tide, you know, tide goes out and you find out who's actually wearing clothes, but >> Yeah, this is a little bit like a This is not a I repeat, this is not a Ponzi scheme. But there is Ponzi scheme being basically a fraudulent investment scheme where you pay returns to investors that have invested based on the contributions of new stooges. >> [laughter] >> This is not a Ponzi scheme, but if we think about it a little bit like a Ponzi scheme, you you continually need more money coming in in order to keep the wheels of this circular flow of financing going. And if the money stops coming in, either in the form of new demand because AI products are not doing as well as we thought they would, or in the form of debt financing, or in the form of new equity checks being written. If the money stops going in, then the whole thing falls apart, right? And the whole, you know, the circularity between these companies that are basically propping each other up, if money stops coming in, the whole thing collapses. If the pie stops growing, if money stops going into this pie, the whole thing collapses. So, when you hear news stories like we heard last week that Nvidia is in talks with Open AI to guarantee $250 billion in financing for a data center in Ohio, basically saying, "Hey, Open AI is going to be using this data center, but we're going to backstop it with our very, very good credit rating, right?" Um you start to get worried. And previously, the market shrugged it off. But when this news came out a few uh last week, Nvidia's share price went down 4 and 1/2%, which is a really, really big move. So, there is definitely this concern that A, there's a little bit of circularity in this in this build-out. And if And at some point, there might be an event, a default event, or some kind of event where we really, really start seeing a significant sell-off. Or we see some companies going bust. >> In terms of like smoke, uh no smoke without a fire, South Korea seems to have had some some pretty big puffs of smoke coming out from their local stock market, where it hasn't been uncommon in recent days and weeks to come in and see like the South Korean KOSPI down 10%, which is huge. So, what is there anything in the tea leaves there from the South Korean perspective and the makeup and composition of the equities involved in that local stock index? >> Yeah, absolutely. So, uh So, these are the big chip companies, right? The SK Hynix that I mentioned earlier on, Samsung Electronics, they have had significant sell-offs. And in fact, last week, the index, the KOSPI, the South Korean KOSPI index dropped by 7.4% triggering emergency trading curbs, which in your language is pretty significant, right? So, So, yes, that this could be the canary in the coal mine. These big chip These chip companies that have had such a rapid expansion are starting to really struggle, and there's a number of different forces at play here, whether it's the Chinese memory chip giant CXMT IPOing and surging 460% on its IPO, which means that more more memory chips are going to come online and it's going to drive the cost down, right? Or whether it's the open weight models, which maybe rely slightly less on this big AI infrastructure build-out, or maybe it's just the whole energy and momentum of this trade slightly going out. Again, I'll do what you do very well, Ans, and I'll say that although the Cosby decline on Tues- last Tuesday meant that the index has fallen about 25% over the last month, it's still 46% higher year-to-date, right? So, there is also this point, and again, you know markets much better than I do. There's also this point that there's just there's probably just a load of profit taking, right? It's just a little bit like, all right, okay. Let's peace out a little bit over the summer, sell a little bit. And then come back come back and set them up. >> Classical rule of like investing and trading, yeah. Don't don't get FOMO. Chasing the Cosby and then buying at the top, uh thinking where every man and his and his dog has been telling you about all the money they've been making on chip makers in Korea. And then you and the taxi man go in all in. And then we're down 25%. Story as old as time, that is. Just look at Bitcoin. Yeah, exactly. Just look at Bitcoin. >> [laughter] >> So, how about some final areas of interest as well here? I I'm remember seeing then in your notes the other, I guess, major area here is part of the build-out is the data centers and and real estate. I was literally just at a company called PGIM earlier this afternoon, Huge kind of private equity real estate business. It's kind of their area of specialism. So, how does that part or component come into the mix and what does that look like in terms of where we're at in this AI cycle at the moment? >> Yeah, it's a again, it's it's so fascinating and it it's going to be really This is why everyone loves talking about it because it's such an all-encompassing thing that is affecting every area of the market. And obviously, we've had this huge boom in real estate in the context of massive massive data centers, multi tens of billions of dollars of data centers being built around the US and increasingly around the world. There was a very interesting article by blogger Groundbreaker, Substack blogger, which has gone a little bit viral called The Second Derivative, Why No One Understands the AI Boom. And their argument was basically, "Look, this is not an equity boom. This is not an equity market boom like the dot-com bubble. This is more like the 2007-2008 mortgage-backed credit boom and then bust." So, there is loads and loads of debt being taken on to fund these massive data centers. But again, these data centers are only going to repay the debt if the demand, if the pie, sorry to keep using the word pie, used it a lot this podcast, if the pie gets bigger and bigger and bigger and bigger. And if, for example, lots and lots of the total addressable market gets taken up with open weight models that don't necessarily need quite as big a AI infrastructure buildout, then we might have a bunch of white elephants, big under underutilized data centers lying around. Which makes me think quite a lot of the old mainframe computers, right? Yeah, it's technology tends to get smaller, right? Not bigger. So, I wonder whether we're going to have all of these data centers and then increasingly we're going to be able to do just as productive a work and computation with a lot less power and a lot less compute required. And that's what this article is saying. There's going to be hundreds of billions of dollars of debt that may not be able to be repaid. >> I was just having a quick search online cuz I do remember I think it was Blackstone. And obviously going their strategy all in on the on the build out. And then I saw I was trying to click on to this article. It's kind of got a gateway though. The COO who's the kind of prominent face of Blackstone, John Gray. Uh current AI infrastructure boom differs from the previous investment cycles. So, of course it does. >> This time is different. >> [laughter] >> So, yeah, it's so interesting though. The amount of serious institutional big money that's within this that's that's sewn up in this this theme is too big to fail? >> It almost feels too big to fail, doesn't it? >> Mhm. >> [laughter] >> So, let you know, one of the things I think our listeners have enjoyed from some of our recent conversations is then we've covered a lot of ground here. So, once you start coming to the conclusion aspects, what's the main things to sort of think about or the way to summarize some of the things we've been talking about? >> Yeah, it's really interesting. I think artificial intelligence is real, right? And it will boost productivity. And it already is to an extent boosting productivity. And by the way, what I mean by productivity is that we can do more with the same amount of resources. I can do more podcasts. I can do more of my work because I've got these amazing tools, right? But what it might also do is it might destroy the profitability or the to use an economics term, the supernormal profits of large corporates or businesses that have had massive barriers to entry because of their scale and because of their knowledge base and because of their infrastructure, right? So if you are a massive company corp, you know, corporates have been the main wealth creating engine of the last you know, two two generations, right? Because they have been able to create in vast profitability over having barriers to entry and an entrenched market position. AI is going to boost productivity, but it might boost the productivity of much smaller businesses, individual people, freelancers, etc. And the value, the productivity boost will accrue maybe not to the main blue chip companies on the stock market, but it may weirdly enough, maybe to end on a positive note, it may accrue to normal people that can do far, far more with their finite amount of time and therefore can do things like start a business and use artificial intelligence to create all of the housing and the infrastructure around what it takes to start a business and run payroll and do marketing and things like that. So it really does lower the barriers to entry for starting a business. It lowers the moat of the large corporate. It doesn't necessarily mean that people will be more profitable that overall profitability will rise. That's not that not necessarily a bad thing as long as overall productivity rises. Maybe to conclude then, your thoughts, if AI continues down this path, intelligence becomes abundant, almost free, like you've just been explaining. Everyone's interested then, particularly younger people, or those in the early career, even people in their careers, about reskilling. You know, careers are going to be longer than ever uh in the current day and age, and probably given technological changes, there's going to be kind of different iterations of your career over time is to be expected. So, what skills specifically would you encourage a student or someone like that to think about when they think about their development for the next 3, 4, 5 years? Yeah, it's a really it's a really good question and a very very difficult one to answer because there are lots of glib things that you can say, right? Yeah, yeah, I really encourage you to show initiative and and have curiosity and all of these kind of things, right? And they're all important. I think I'll just I'll end with a comment from Dan Loeb, who is famous hedge fund manager, runs a thing called Third Point. His interview with Patrick O'Shaughnessy on Invest Like the Best is a great podcast I'd recommend to anyone who's interested in this kind of stuff. And he was talking about the advantage, the edge that you are going to have as an analyst, and he was talking about young people getting into the investing industry. And he's saying, "Look, you know, I want people I want humans, like I don't need models, like everyone's got models. Models are, as you say, abundant, intelligence is abundant from that perspective. I want people that are going to go to the fast food chain that they're analyzing, and going to order everything on the menu, and they're going to eat it, and they're going to order it all again the day later and see if it's exactly the same, same quality, same heat. Then they're going to talk to the manager, then then they're They're to talk to 10 of the customers, right? And show that level of curiosity, show that level of drive, and just human interest in the world. And that I think is what's going to make you stand out. You just got to kind of tweak get as a lot of people say, get up, leave your laptop, go out into the world, >> [laughter] >> and see where the opportunity arises. Cool. Well, look, on that point, we'll end it there. As we said earlier, any questions at all, any thoughts, any AI specialists out there, if you're working at one of these labs or hyperscalers, we'd love your take as well. Open exchange shovels, all the software product layers, so let us know. Don't forget to subscribe. More of these shows coming every week. Just a reminder, we have a bit of a deep dive into business case study or into an M&A transaction beginning of the week and at the end of the week a review looking summation of the global macro environment, of which obviously 2026 is a fascinating time to be involved with anything to do with the economy and financial markets. So, hope you enjoyed that. Stephen, thank you as always, and we'll see everyone next week. Thank you, Adam.