The AI Trade Everyone's Getting Wrong

0:00 Samsung just became the most profitable company in the world. But nobody actually understands what they make. I feel like when I think about Samsung, I think smartphones or computers, but the reality is is that almost 96% of their company's profits come from a single entity, and that's memory. And when we think about memory, we've talked about this in the past, there are three main providers. Two of those three just so happen to exist in Korea. So what are two of the most profitable companies doing in the same country? And what are they doing to unlock all this value? And then what are the implications of this across the entire market? That's what we're going to get into today on this episode. EJ, what's going on with Samsung? How is this even possible? I feel like Samsung was the company that like people buy Android phones from, not the most valuable company, profitable company I should say, in the world.

0:42 >> Yeah. No, I I remember as a kid like a bunch of my friends had like Samsung phones and that's all I knew them for. So like when I heard the name come up repeatedly over the last year and a half or two years for AI, I was kind of like, okay, what is this company doing different? Now firstly some of the headline numbers. Samsung recorded their quarterly profit Q2 at 58.5 billion. Now this was surprising to many analysts because they were expecting it to be 55. So you know they overachieved by around $3 billion. But the larger headline reasoning behind this is they absolutely crushed Nvidia who only came in at a measly $53 billion. Now look at this. Oh my god, >> it's a crazy chart. So, so just obviously I'm justing, but the point I'm trying to make here is why has Samsung been able to achieve so much profit? You mentioned earlier that 94% of their profit over the last year and a half has come from one single department in their company and that is AI memory. Samsung is currently the number two largest provider of something known as HBM, high bandwidth memory. In order to understand why this is such a useful commodity, every single GPU that Nvidia sells, every single TPU that Google makes, any type of AI chip or AI hardware quite frankly requires a ton of memory. And emphasis on the word ton because typically consumer devices, your laptop, your computers require just kind of like a standard pretty predictable amount of memory. But AI memory in particular has been quite exponential.

2:18 So, Samsung is one of three major providers and because they own a relative monopoly on this industry, they've been able to hike prices up massively. Now, I know I'm throwing out big numbers here, but to help you understand why this is such a gargantuan amount of money to make over 3 months, we're roughly talking about $650 million per day or $27 million per hour. So, this is more money than the most valuable company in the world makes. Yet Samsung isn't the most valuable company in the world. And in Korea, right next to it, in the number two position is SKH Highix, which is relatively their sister company or not their sister company, but their sister company in terms of memory, who is the number one AI memory provider. So there's some really unique modes being formed in Korea. And it's weird because typically in the west, you largely have the monopolies in tech.

3:10 And now we're seeing it in Asia, specifically Korea. Another fun fact, if you actually scroll up to the top of this artifact, we distilled it down to seconds. That's $7,500 per second that this company makes, which is remarkable because going back down to that other chart, I mean, we see the little bar on the bottom here that shows Samsung Q2 2025 at $3.4 billion. For those of you who are a little confused about years, 2025 was last year. That was a year ago, they were making 3.4 billion. And now they've surpassed Nvidia at 58.5. So, like, what on earth is going on? what is going on with this memory? And I think the essence is like you mentioned EJ is like every computer needs two things. It needs the workbench which is the memory and then it needs the filing cabinet which is the NAND. And there's only three companies that make virtually all of this. Each company kind of has their specialty. There's DRAM first, which we commonly reference as the workbench.

4:00 It's the RAM inside of your gaming PC. So if if you've ever built a gaming PC or if you've ever been on Newegg shopping around, you'll recognize this term called DDR5. That's the fast memory. That's the temporary RAM a lot of people call it and it gets wiped when you turn off your computer. Then there's NAND which is more closer to like the filing cabinet. It's your SSD. So if you have an iPhone, chances are you have NAND flash inside of your iPhone. That is a little bit slower but much cheaper per gigabyte. And then there's HBM, high bandwidth memory, which is the AI stuff.

4:28 And that's the reason for everything taking place right now. And those DRAM chips, we've described this before. They're basically stacked up and they're 12 to 16 stories high, kind of like a skyscraper of memory. and it uses this incredibly complicated thing. Now, SKH Highix, one of these two companies we've mentioned, they own about 60% of the entire HBM market. So, any AI chip that gets built, chances are it's gone through them. And that's why so much of this monopoly has started to form and that's why so much of these profit margins has begun to expand is because there are three companies in charge of this bottleneck that controls the entirety of the AI market. And you can see here kind of a visual of what this looks like. DRAM next to NAND next to HBM and Samsung and SKX in particular just so happen to both be the Korean superstars that are making this this reality possible right now. If there's a single sentence that I think would describe why I guess people are so bullish on these companies. It's AI is basically a black hole for memory in general more so than any other consumer or enterprise tech wave that we've seen before. So to give you an idea when you write a prompt and you submit that every single AI model chat GBT cla doesn't matter references or rereads the entire model weights. Now the model weights are what these companies are spending billions and billions of dollars to build to build a better intelligent model. It rereads it every single time.

5:50 So you need a lot of memory and every model by the way kind of 10 to 20xes the requirement of memory needed to build that model. Now, at the scale that they're building these models, 15 trillion parameter models, 20 trillion parameter models, whatever that might be, that is going to keep going up. So, there's a black hole for memory there. The craziest part is that's not even the largest part. When you talk to your chatbot, when it remembers things about you, when you ask it to remind yourself of the thing that you spoke about it in a different conversation, that's all temporary storage, which is on the nan flash side of things, and that is equally important. So the the overall effect of this is more memory required from all three of these companies that were that were well these two companies mainly Samsung and SKH highex but also from the likes of Micron in the US and that's why it is pretty crazy to say that memory is no longer a commodity and I think this is an important point because typically or critics will point out an AI memory bubble is a bubble and it will eventually pop and the reason why they're saying that is justified on historical facts which is AI memory has gone in boom and bust cycles constantly over the last couple of decades. In fact, 3 years ago, SK High was doing so badly that they almost got bought by Josh, can you guess the company?

7:03 >> No, don't tell me Samsung. >> Micron. >> Mic. Oh my god. >> Micron owned SKH Highix. >> They came this close because it was it was the pico bottom of the memory cycle. They couldn't sell memory. They had overinvested in this thing called high bandwidth memory and they never knew if they were going to sell enough units and Micron almost purchased them but they doubled down and they didn't sell and now they as of this week were the number one most valuable company in Korea but Samsung eclipsed them with their recent profit earnings.

7:38 It's pretty insane. >> That's unbelievable. And there's this really crazy thing if you go down back down to the visualization right underneath that we have the wafer math of how this actually works because HBM is the AI memory. who need HBM and it's really demanding and it actually destroys a lot of other consumption ability because according to this 1 GB of HBM consumes the factory capacity of roughly 4 GB of regular DRAM. So every wafer that becomes AI memory is a wafer that never becomes laptop or phone memory. And that's why we're starting to see prices from a lot of these consumer grade products going up. Like for the first time ever, Apple just increased their products across the board on the laptops probably on the iPhone this year. Why does that happen? And it's because HBM is consuming four times the amount of memory consumption per gigabyte than traditional memory otherwise would. And as a result, this creates pretty hefty margins. And we have this pretty cool chart to reflect just how good these margins are. Because again, when you have a monopoly, you set the pricing. You determine how much money you actually stand to make. And we have it compared relative to other industries. So, in a grocery store, for every $100 of chips sold, they collect about three of that as profit. Very slim margin on grocery stores. Car maker is a little bit better at $7. Apple, which is the traditional tech margin when you sell hardware, is about 30%. Samsung is 52 and SKH Highix is 72, meaning $72 of every $100 sold goes to their top line.

9:07 It goes right to the balance sheet they're able to use and spend that on R&D and profit. So they have this like tremendous amount of profit margins that they're able to play with and I think that's what's making them clearly so profitable and Samsung's memory workers are getting bonuses now worth six times their annual salary which is really incredible to see as we were talking about this before the show how generally speaking Koreans like to gamble a little bit more than the average person and if you are a Korean who is interested in gambling and also interested in just making a tremendous amount of money if you either bet on SKH or you start working at SKH You either got a six times multiple on your investment or a six times multiple on your salary as a bonus. And this just seems like this unbelievable gift to the country of Korea. Oh my gosh. Here's a fun fact.

9:52 The luxury sales market has tripled over the last 4 months in Korea specifically because everyone's getting rich off of SK Hex. saw a story about folks over in Taiwan who are borrowing up to $60,000 from their bank at crazy interest rates just to buy TSMC. The the point is there are crazy things happening in the AI markets in Asia specifically and Korea is no different. It's interesting you raised the the point around margins. I was looking at the price increase or the cost of memory that SKH Highix and Samsung have been selling over the last 6 months. Now, in Q1, they hiked their prices up 90%. Just imagine that for a second. Just just consider that. Imagine >> that's so brutal.

10:40 >> Buying something that you use regularly in your phone, in your computer, and suddenly the price of it doubles purely because one component of your device has gone up almost 100%. But it gets worse. In Q2, it went up a further 50 to 60%. That's why Samsung just reported the best earnings ever, 19x better than last year. And more money, by the way, than they've made in the last 40 years they've made in one single year. So that's the point I'm trying to make is these prices are have been hiked up pretty aggressively. And so your question might be, well, it's not going to go up any higher. Well, Samsung announced or are going to announce in their quarterly earnings that they're going to hype hike it up another 20% in Q3. So, the point is simple. Memory is extremely scarce. There are three companies in the world that can produce this memory that is required for AI. 2/3 of them reside in Korea. And Samsung and SKUX hold the monopoly for something called high bandwidth memory specifically. And they can price it however way you want. Now you may look at me and listen to this and say, "Well, why can't we just create more memory?

11:49 Why can't there be other companies that do this?" It is a very closely customized process. These companies, SKH Highix, the CEO of SKH Highix hangs out with Jensen almost every other month or every other two weeks actually to try and figure out how best to create memory for the chips that they're building. All these other hyperscalers are doing the exact same thing. They're begging. Remember there was a story we spoke about a few months ago Josh where Google and Microsoft had flown their execs over to Korea to convince them to sell them more capacity. They even offered to pay couple a couple hundred millions of dollars to pay for their equipment that is required to expand their fabs. It is absolutely insane what's going on. And you're right as a skeptic if you're listening to this and you are a skeptic to think you know what like this is a bubble it's going to pop soon. But also if you look at the fundamentals like companies such as Micron and SK Hanx the PE ratios are still astoundingly low because they don't have enough supply atoms can't get moved. So hearing this price increase, it kind of makes me it makes me like, you know, happy for you, happy for the investors, glad everyone's making money, but I think about the impact that it has on my life and like the actual consumer because this is starting to trickle down into the real market and it is affecting me in a way that is pretty annoying. When we look at the price comparisons of what things cost today as a consumer because of this demand crunch, the numbers are like pretty brutal. just to get a 32 GB kit of RAM, which is traditionally not that expensive and a small component in a computer, it's two to three times more expensive this year than it was last year. And now, if you want to build a PC, over a third of the cost of that computer, it's just the memory. And the same thing is true with a lot of Apple products. Like we mentioned, the MacBook Air now has increased $200. It went from $1,100 to $1,300. The MacBook Pro, even more. If you bought a MacBook Pro just a couple weeks ago, it was $1,700. And that same MacBook Pro today is now 2 grand. Same with Mac Studio, which has increased $1,300 because why do people buy Mac Studios? Well, they buy them to run local inference on their machines.

13:52 They have a tremendous amount of memory. That memory is costing a lot more. Now, it went from $4,000 to $5,300. And the same with their lowcost MacBook Neo, increasing $100. So, the price of everything is continuing to increase. If you are a consumer, there's probably some pretty strong forces that be to convince you to maybe keep your phone for another quarter or two or another year or two or another cycle because these prices are just going up up up.

14:15 Now, the question I have for you, EJ, is is this going to continue and how long will this continue? Like, can they afford to just continue to move these prices up and up and up and just expect everyone else to absorb that price or is there a moment where they're going to have to start cutting those margins a little bit because the market's just fully tapped out? we cannot afford to continue to spend this high on memory.

14:34 >> There are many different ways to analyze this. There are frameworks. There are Wall Street traders that are like spending hours and hours trying to research this and figure out hm will this memory demand keep going up? There's only one proxy that measures that are more people using AI. >> What does that curve look like? If you believe as you're listening to this that more people will be using AI, spinning up multiple AI agents per person and increasingly use it across work, leisure, and whatever they do in their day-to-day life, then you are betting that memory demand increases exponentially. But then you might be wondering, well, what if they produce enough supply to meet that demand over the next year? Well, number one, you're not going to get that supply coming on board until 2030. These chip fabs take so long to build. They are meticulously designed and they are hard to flood the supply right now. Now, typically in a boom and bus cycle, you could overflood the supply because you can only make so many mobile phones. Apple and Samsung, ironically, were the ones making these mobile phones. So, you can kind of predict the consumer hardware.

15:41 You can't do that in the GPU game because so many people want that. So, it depends on if that demand outpaces the supply of memory. And right now it's looking like it's 3 to 5x over the next couple of years over the actual supply. So technically we're going to be constrained. The other perspective I have on this is we will look for alternatives in terms of memory suppliers in China. And Apple's been doing this. They're shopping around this company called CXMT in China that produces similar DRAM and HBM. But here's the sticker. They're oversold to the Chinese AI labs and they also don't have any supply. So, Apple is trying to strongarm them and the US government to allow them to buy from a competitor in China. and the final point is what if we create a new model architecture that doesn't require as much memory.

16:31 Now, the skeptic will jump at that and they'll say, "Hey, memory demand is going to crash. Prices are going to like crash like like this is the bubble popping." I would argue the opposite. If memory becomes cheaper, guess what? more use cases get unlocked, more AI agents get run, and the world economically becomes way more productive. So, does that sound bullish to you? Yeah, I guess so. >> Yeah, we've been so bullish, and for good reason. Like, Micron's up what, two and a half times.

16:58 >> We made the call at the end of last year. It's up 150%. >> We've been on it. I will say the Luminous portfolio has been doing absolutely incredible over the last 12 months. Like everything that we have predicted to varying degrees has done pretty well. But now I have to ask the question. It's like okay SanDisk up 6 and a half times Micron up two and a half times SKH Highix two and a half times Samsung 1.5 multiple yeartoday all these companies have done incredibly well but just recently all of those companies are down more than 20% from their recent highs in the stock market which if we're speaking technically that is a bare market. So we're technically in a memory bare market right now which doesn't feel like it based on what we're talking about. And the weird paradox here is that like Samsung just announced their earnings earlier this week. And in fact, they beat estimates, which was unbelievable. They had this record earnings report. That's how we found out that they're more profitable than Nvidia. And yet, the stock sold off 9% in a single session. And SKH Highix fell off 15% in a single session. So, these stocks got crushed. And there's this interesting thing at play where like, okay, these memory companies are saying that they have record earnings. They're going to continue to increase prices.

18:05 They have increased amount of sales coming down the pipeline that are going to fulfill all this demand. And yet the market is saying, "Whoa, whoa, we got to slow down. We are taking this down 20%." Now, there are some things that may have spooked the market here. Meta has hinted at capping their AI capex. Not that they've built anything fun with it anyway, but they've they've kind of hinted that they want to lower the capex. And I guess that's where we're sitting now is that this this weird point of uncertainty where the companies are saying, "Hey guys, we're all good on this end. We have incredible margins. We have an incredible business. There's tremendous amount of demand." But the stock market is saying like, "Whoa, hold on there. We've gone up a lot. We might have to check you. There's some unknown unknowns that are coming down the pipe that they're not entirely sure about."

18:44 Can I give you my honest and boring take as to why the markets are crashing? Tell me, why are we down 20% right now? We just came to the end of pretty much quarterly earnings, and this is just a sell the news event. I think there are a lot of funds around the world that are leveled up on these stocks and they kind of want to get in at a better price. Now, if you're listening to this and you're thinking, "Easy you sound delusional." Fair play. Maybe I am. But I think that if you are any kind of a long-term investor in AI, it is so strikingly obvious that we're going to need memory. And if you've done any kind of in-depth research around how to produce that memory, you're going to need these three companies, right?

19:28 That's the the way that it sits right now. And I think that's I think people are getting concerned. People are getting worried. They're wondering how on earth can we make this much money over the same stocks month after month after month. How can the the profit margins can't keep going up? But they kind of can keep going up as long as memory supply is underpriced or underpaced for AI demand. And that's currently what we're seeing before. And we've never seen this before. That's the whole point. Like these companies are more in demand than they have ever been.

19:57 Now a fun fact is like with every cycle there is a boom and a bust and eventually they will be a bust I believe. but that's only at a point where these fab start over supplying the entire market. And at the last poo bottom when memory was super cheap do you guys want to guess which company or companies were aggressively pushing their prices even lower to buy as much memory as they can? >> Oh there the shoe was on the other foot.

20:24 It was Apple. M it was Apple. >> Apple had the monopoly on SKH and Samsung and forced them to sell DRAM and HBM for the cheapest prices possible. So the point I'm trying to make is this might seem super aggressive. Samsung skex hiking up prices. They're just doing what other companies did in a capital efficient market and that's what we're seeing right now. I think this will be a nothing burger in a couple of months time. That's my bet. Yeah. And you have to imagine like there probably is some scar tissue like you mentioned.

20:51 We have never seen this before in terms of demand but we have seen this before in terms of price where these kind of memory super cycles do happen. The last one being in 2017 to 2018 where Micron peaked at about four to five times price to earnings ratio and then proceeded to fall 60% while their earnings were were still rising similar to what we have now. So I think people are just a little concerned a little scared. This is probably a short-term outlook. Longer term like you mentioned things are looking pretty solid. In fact, coming up, we have a test of this and a pretty big one because SKHX, as we mentioned earlier, is a Korean company, but not for long. On July 10th, they are being listed on the NASDAQ under the ticker SKNY, raising about $30 billion. So, this feels like a really pivotal moment in the memory world of how the American market is going to perceive SKH Heinix bringing their listing onshore. And we're going to see kind of how the market reacts. And this is this is an interesting one. EJ, I'm curious for your take. is this something you're planning to participate in and invest in? Like we have a a sneaky IPO here.

21:55 >> Yeah. In short, yes. I'm bullish on memory. I own Micron. I own DRAM, which is kind of like a basket that owns each of these top memory companies individually. Like if you're a US citizen, you can't necessarily get access to SKHX or Korean stocks in general, but you can do it through these kind of like basket things. DRM is one. But yeah, I'll be buying SKHNX when they do IPO here in the US. I've been waiting for it. fun fact, >> guess where I think 2 to3 billion of this $28 billion ADR is coming from.

22:28 >> Leopold Ashen Brener. That guy's back at it again. This dude does not miss our last episode, which we recorded on him, which you guys should go look at. It has like 120,000 views now on YouTube. it talks about how his portfolio made no sense. He's like, "Why are you shorting Nvidia as your largest position?" Well, turns out Nvidia is down 20% since recording that. Who would have thought? And now he is participating in SKH Highix as the IPO as one of the largest seed investors in this entire thing. So like the dude is on it. I mean, as much as people push back against his decisions, he has really >> not missed.

23:02 >> He's killed it. >> Yeah. He's done such an amazing job. >> Yeah. He he has one of the his ears are closest to the ground in terms of what is actually going on in the Frontier AI labs. and it's so funny because you know some might say there's some questionable practices going on but I remember I think a few a month ago he announced major investments through his fund in some of these Australian neoclouds and then they signed a major partnership with the project lab. If you are blindly kind of like following this guy, I would urge you to be careful, but it's promising.

23:37 a question you asked earlier is we're going to see this as a real test in the market in the west at least as to whether the memory trade is over or whether it is in demand. this round, this SKHEX IPO which is launching in a day's time if you're listening to this right now is currently rumored to be 4x overs subscribed. So Forex there is insatiable amount of demand for memory stocks in general and that comes from capital institutions but it also comes from retail as well but you can't get that amount of money without having institutions that have thought long and hard about this pension funds etc that are buying this and holding for a while in play. So that's what we're seeing right now and I think that this is going to be a pretty huge success. I think these stock declines over the last couple of days is a nothing burger. I think it's just a new quarterly earnings and hedge funds are doing their thing.

24:28 They're trading their >> probably right and if we if we go back to the artifact there's a there's a super cycle or sell signal which I think is probably a decent place to wrap up this episode in deciding like hey what where is everything going here and I think the the reality that we've kind of converged on I would say between the two of us let me know if I'm wrong is that longer term things are looking great.

24:49 there is no limit to the amount of memory demand that's happening. there are a very small subset of companies that are capable of building this type of memory and there's not a very clear path for new entrance to actually come and make a meaningful impact on their business. So as long as this demand continues to rise which we assume it will due to just the sheer demand volume of AI and tokens that is looking good. On the short term, people are a little bit afraid. They're a little unsure and uncertain because there's been such a large runup and because there's probably some scar tissue from these previous boom and bust cycles when it comes to the memory companies. So there is a lot of uncertainty, but this makes sense to me and we've seen this over and over again in the market where like there are a lot of knee-jerk reactions that are incorrect and overdetermined. And the reality is is that if you have a slightly longer time frame, a slightly longer time scale, then you can be right eventually. And perhaps you're not right in the next month or two. If you bought it a month ago, you're looking pretty dumb right now. You're probably feeling like, "Oh man, I bought the wrong company." But I think the reality is is that 6 months, 12 months, 18 months, 24 months in the future, these things are still going to continue to be pretty strong businesses. And I think that's probably the reality we could leave it at.

26:00 >> Yeah, I agree. I'm curious what the folks listening to this think. I know AI memory isn't the sexiest of topics to talk about, but it is increasingly become the most important component to this AI infrastructure capex spend race. it is the most costiest and priciest item that is driving up all the prices of your consumer goods. What are your thoughts? Do you think we are overly optimistic? Do you think that memory profit margins are going to keep going up? Do you think memory stock prices are going to keep going up or do you think this is an insatiable bubble that is eventually going to burst and we are all currently delusional and Josh and Eas are marking the top? Curious to hear from you guys on what you think.

26:41 thank you so much for listening to this episode. we are trying to cover different types of companies that just aren't based in the west and also kind of explain how and why they're important and how you guys can maybe potentially get exposure if you are interested in doing so. we're always excited about talking about these topics, but if there's anything that we've missed recently and you want to hear more about, let us know. we've noticed that a lot of capital markets episodes that we've been making have been super interesting and engaging by you guys. We love the comments. wherever you're listening to us, Spotify, YouTube, Apple, let us know what your thoughts are. Give us a rating, give us a thumbs up, and tell your friends about it as well. And I think the final thing to say is Limus is in the market for partners that we want to team up with and we want to talk about your product and your feature. If there's something that might be interesting that you think Josh might like or if you know of someone that has something that they might want to publicize, let us know. We are in the market. We have hundreds and hundreds and hundreds of thousands of people every single month that listen to us and we would love to promote and talk about your product. but aside from that, I think that's it.

27:47 >> Yeah, it's it's almost like you've had 200 episodes of practice on that sign off. That was pretty good. No notes, no additions. Thank you all for watching as always. Very much appreciated. Share with a friend if you think they would enjoy. And as always, we will see you guys in the next episode. Next one's the roundup. We got a big week coming for the roundup. My god, we got couple new models coming down the pipe. We got a new >> just a couple. Yeah, just it's a good one. It's a good one. You're gonna want to tune in that's dropping on Friday. We got a newsletter coming out as well. And I'll see you guys on that episode. Thank you so much for watching. See you guys.

28:14 Peace.