The Realistic Warning on AI Stocks | All-In Podcast

Hey everyone, me Kevin here. In this video, we need to talk about a critical shift in artificial intelligence. Towards the end of the video, just to give you a quick breakdown of what this video is going to look like, I'm going to talk about this latest $750 billion news with Nvidia uh and what the actual impacts of that realistically are since there are time frames associated with that important for you to know. I'll talk a little bit about Iran and the Federal Reserve since those are going to be some catalysts that we need to talk about for later this week. But I think more importantly, what I'd like to start with is what's going on with artificial intelligence and the three layers of profit that are really, in my opinion, falling to just one layer of profit. Now, we've been talking about this for a few months now. We've talked about how there are three layers of profit in artificial intelligence. There is the level that everybody knows, the the large language model. There's your LLM layer. Uh and then above that you've got your software layer built on top of that and below that you've got your infrastructure compute layer. Those are three le levels of potential profitability from artificial intelligence. Lately most of the enthusiasm around uh where profits are going to go have been sitting in that LLM layer. That's where the idea or sort of the investing mindset has been and in the compute layer. The application layer has actually gotten destroyed. Software, right? Software stocks have plummeted. LLM hopes for whether that was the integration into the SpaceX IPO or the coming anthropic IPO or the potentially coming OpenAI IPO. There's been a lot of enthusiasm around raising money via IPOs for that LLM layer. And then of course the compute layer has been just explosive since uh really 2023. Whether it's now the memory cycle that we're seeing or originally what we saw the GPU cycle with Nvidia and then of course all the downstream components whether it's a vertive whether it's a Dell assembling and selling at lower margins those server racks but doing so much in terms of volumes these are the three core layers. The problem that I'm finding is that we started talking about these three core layers months ago and warning that my opinion was that LLMs are going to commoditize and after LLM LLM's commoditized I am also of the mindset that the hardware or the compute layer over the next decade will also commoditize. This is why I say I do not want to invest in the LLM layer or the compute layer. And I've been pretty clear about that on this channel for a long time. I'm not interested in the compute layer. I actually think the compute layer, you know, at the end of uh let's say call it 2033 will probably look a little bit like um a big bag, right? You spent a lot of money on chips that at the end of a decade have likely depreciated substantially. And I don't know that your ROI has actually paid for itself especially with the level of hype that has gone into uh some of the Neo clouds or uh even chipmaker stocks or even what we're seeing with memory now uh which we know memory is cyclical. We know memory is a commodity and we know that at some point that cycle stops even though the CEO of SKH suggests a while demand is infinite we may never catch up to demand. Usually those sort sort of comments if we will are actually themselves indicators of us getting closer to a peak. Uh but let's analyze for a moment what the all-in pod talked about. I want to start with sort of my thesis. So I gave you my thesis. Now I'm going to add commentary to what the all-in pod talked about. The all-in pod when they start talking about something I do feel like it reverberates around the suits pretty rapidly. So, I don't actually think of the all-in pod as um this is this is not to be a slight on them. It's just it's just a consequence of of their reach. I actually think when they're starting to talk about it, the suits already, you know, they've already integrated it into their investment thesis. It's it's almost sort of too late. And so, when they're talking about it, it's you better already be with that or you're going to get left behind. Again, that's not a diss on them. what you're saying. You know, we've been talking about these three different layers and actually how in my opinion one of the most desirable is finding the software application layer with pricing power over the next decade. I'll give a quick example. Quick example of course could be a uh Palunteer or another one could be an Axon where your mode is actually how you're integrating either corporate or government data. use Axon, for example, if they're the exclus exclusive data collector uh via body cams or police reports or 911 calls or whatever, then they have a data mode that makes it really hard to replace their application layer, right? Those modes are very useful. So, what did they talk about on the all-in pod? Well, this is what they said. uh they said that or Chimat started with suggesting that LLMs appear to be commoditizing faster and the Chinese sort of deepseek/Kimmy moments we're seeing suggest that uh companies like Anthropic might be in an a valuation preservation race right now because the value that they're extracting today via uh you know clawed subscriptions or whatever might not actually be sustainable for a 5 to 10 year period. Uh I completely agree. David Saxs though has a fantastic counterargument and he actually says what you have to do with anthropic or open AAI is you have to look at um co-work or cloud code or you've got to look at um uh oh what's the what's the version that um uh it's it's whatever the open AAI version of uh of the the coding software right you have to look at these application layers and see that openai and anthropic are already racing ahead of the commoditization of LLMs. They already recognize that that's coming and that's one of the reasons why we keep getting new application level releases from OpenAI and Anthropic which is making a lot of software companies fearful like the whole Figma and C uh you know Canva debate over uh-oh is Anthropic just going to take our sort of moat do it a little bit better and then replace the software companies. Sax's point is that it's almost like OpenAI and Anthropic know the commoditization is coming and the more they can get into the software layer, the more profitable they will end up being long term. And I have to say, fantastic counterargument. You what you really actually now have though is a consensus amongst all of the four on the all-in pod. Even though if you watch the video, you'll walk away going, "Oh, well, three of them think that LLMs are going to commoditize and one of them doesn't." I think that's the wrong way to walk away from this. I think the way to walk away from this is all of them agree LLMs are going to commoditize. David Saxs just argues it's the fact that Anthropic and OpenAI are building that application level to get ahead of that commoditization actually still makes those interesting companies and interesting investments. Very interesting point and I agree. So what we end up getting is this debate over how LLMs will really just end up looking like varieties of Coke. Do you want uh Coke from Coke? Do you want Coke? You know, cola from Pepsi or cola from another third party company, whatever. Uh and the the biggest concern for companies is actually not that LLMs will commoditize because they will. It's actually how do you end up preventing your data from getting distilled if you're an AI company. So for example uh my real estate startup we actually think it's very very difficult to distill real estate data uh because there's so much that goes into an ARV uh such as renovation cost the comparable costs getting access to all of that data is part of a moat in addition to your training weights that can't easily be distilled because you're only getting a resultant value rather than what went into creating that. And since homes aren't exactly like commodities, it's a lot harder to apply that to a whole lot of other hosts of of homes. Point of that being, I actually think it's harder to commoditize uh a real estate data. And it's one of the reasons why real estate artificial intelligence has kind of not really gone anywhere. I mean, if you think about it, what you see is you'll get sort of generative AI where it's like, help me remodel this room and make it look like, you know, tell me what it would look like with this kind of floor or paint. That's great. but actual real estate valuation AI. There's very little competition in it, which I think is a fantastic benefit to my real estate startup, House Hack, uh also known as Reinvest. Um you can see that over at house hack.com or reinvest.com. But anyway, the you know, there are lessons in all of this that we should touch on and then we'll talk about some of the other economic movers. The lessons in this are if it's becoming more mainstream that LLMs are commoditizing then what we have to remember is any of these new exciting LLM deals or announcements uh open AAI anthropic the IPOs whatever when we look at these companies we should be looking at how much of a moat do they actually have in software and price the valuation of those companies based on their software mode not on the LLM subscriptions or the LLM modes or the growth rates for free users which could be a bunch of Chinese distillers or Taiwanese or Philippine Philippines Philippines distillers anyway. Uh that's no offense to anyone from any particular country. It's just saying that's that's where people think a lot of the distilling accounts are being created. Uh and really ignore all of that. Instead value these companies based on not whether they use an LLM or create an LLM. assume that every software company is going to bucket together large language models in almost a perplexity computer style software and they'll all just use the model that they need at the time for the strength that they need at that given time. Which means the LLMs are no different than oil or gold or corn. They have a purpose, but they're a commodity. And it'll get more and more into that direction, which means and and it, you know, sorry it took so long to kind of get to this this takeaway here, but what that means is of the three layers of compute, and we've talked about this in our course member livereams, of the three layers of compute, two of them, or of the three layers of AI, I should say, two of them uh will likely commoditize. The hardware layer gets commoditized. The LLM layer gets commoditized. What actually survives in artificial intelligence as an investment? Software companies where they have a proprietary edge on data such as again Axon with those body cameras, Palanteer with those lock-in contracts once you get people into those. I mean, I guess you could leave at any time you want. It's sort of like Hotel California, right? You can check out anytime you'd like, but you can never leave. You get the idea. You know, once you're in an an ontological model or you've set your entire database or your company up in one of these, the pain of moving is really hard. The pain of moving from a cyber security company that has now helped you install endpoint device management on every single phone or laptop that all of your employees have and you've managed that. You finally got everybody on board. The pain of moving to another provider is really hard. So you have to ask yourself who has the moat in data when it comes to artificial intelligence. Uh and those are the people who are have a unique advantage uh in in understanding how real estate functions for example. I think we've got a really big edge there. Uh and how we're collecting data and how we're training. I think we're really ahead of the curve in terms of AI and real estate. But ignore me. I'm not not just trying to talk about myself or or house hack. Those are just little notes that, you know, when when I study things, obviously I like to apply them to what we're doing as well to make sure we're staying ahead to make sure we're not getting distill attacked or whatever, right? But also, it it should make you look at companies that have uh even a business mode of data like in it. Is into it losing customers because of AI? Of course, they're losing mostly that free to to no pay or low pay uh level of customer. they bluntly acknowledge this, but the higher level uh customers, maybe those uh enterprise or business levels that are larger and have a lot of their data already with, let's say, into it, QuickBooks, they're actually spending more because they want those AI features. And so evaluating that pricing power is going to be exactly what the market has to shake out over really the next decade. It's one of the reasons why when people look at a company like Service Now and say, "Why is your pricing power waning?" And then you look at a company like Salesforce and it's like, "Huh, your pricing power is stable." And then you look at a company like Palanteer and you're like, "Oh, your pricing power is growing." You you could see those different layers in also the stock performance. Now, in fairness, all of them have sold off uh frankly because so many people are concerned about ah crap, you know, what happens if Anthropic just ends up replacing all of those. But again, they got to get their hands on the data. So, uh bottom line for this segment, and I'm going to briefly talk about some of the other topics. Bottom line on this segment, uh, let's let's try to noob versus pro this. It's really just a a a tool to signpost what's going on here. Okay, so the noob says OpenAI and Anthropic are going to take over the world. Anything they touch is going to go up with it. You got to own OpenAI stock and Anthropic stock because they're going to the moon. They own the LLMs, the frontier levels of intelligence, and every kind of chip that they buy, whether it's an AMD chip, whether it's an Nvidia chip or an ASIC from Broadcom or Marll, these stocks are going to the moon. I don't care if you're water cooling them or building the rack like Vertive or Dell, you're going to the moon. The pro looks at this and says, "Ah, we are probably going to see LLMs commoditized, especially with openw weightight models, which means the LLM side of AI is probably going to collapse in margin. Pricing power goes down. Uh, the compute side does well as long as there's a shortage, but once we run out of like this shortage, when demand starts meeting supply, we won't have massive pricing power in chips anymore. They're replaceable. Everybody has their own once they're up and running and the supply to make them is there. And so the real moat, the real pricing power in artificial intelligence likely comes from the software layer, specifically software companies that have a moat or some unique way to access data, evaluate data, or train data that nobody else is touching. If you can incorporate that into your portfolio for the next 10 years, that that mindset, I actually believe you'll make a crapload of money. # no guarantee is not personalized financial advice. Now, a couple of other things. Obviously, we've got a pause with uh with Iran in terms of strikes. I don't think this is like a durable ceasefire. I don't, you know, while we could see a bounce because oil and uh the tenure is down a little bit, I don't think this is extremely durable. Uh so, we'll have to see uh how this uh this plays out. Uh I think there are a lot of risks this week in the market. We've got the potential of a resumption of fighting Iran. Uh we've got the uh Fed meeting on Wednesday about a 34% chance that we'll see a rate hike. I don't think we'll see a rate hike, but the market's going to be nervous about that. Plus Microsoft and Meta earnings. There's a lot there's a lot coming this week. So, um you know, a lot has to go right for this week to go very green. Uh, and then of course, um, next Monday I'll be back from vacation and, uh, I think then we go to the moon. All right, folks. I appreciate you being here. Thanks so much. Hopefully this was useful and insightful. If it was, consider subscribing and we'll see you in the next video. Goodbye and good luck.