The Real Reason AI Stocks Are Crashing (Not the Bubble!)

0:00 The US stock market just lost more money than we did in the 2008 financial crisis over a matter of seemingly a couple of days. It's been the worst memory crisis, the worst stock market crash in quite some time because the numbers have gotten so high. Memory has run up to astronomical levels. But what's interesting here is this is backed up against some counterfactual evidence. SK Hynix, that Korean company that could seemingly do absolutely no wrong, they recorded more profit than they did revenue this last quarter. So, the numbers are great, but the stock market is saying, "Wait a second, something's not right here." So, we have some insight. In this episode, we're going to unpack this. Why the market is reacting the way it is and evaluating whether or not it's right or if it's a pretty serious overreaction. I'm currently sitting in the middle of a hotel in the middle of nowhere on some spotty Wi-Fi trying to get this episode out to you guys because I mean, this is this is an incredibly important topic. This is feels very timely and important to navigate. So, EJ, let's unbundle what's happening here in the market. I mean, worse than 2008 is crazy.

0:58 >> Yes. So, I'm seeing a lot of commentary in the markets right now, which is the AI bubble is finally popping. It's unwinding. It's been happening for 4 weeks. the Korean stock index itself is down over 30%, which is just the largest drawdown they've ever had. and so, I want to explain what the damage is and then let's get into whether this is right or wrong because I honestly have some pretty strong opinions as to why this is completely wrong and why it's like one of the bigger opportunities right now. So, what is the damage? What you're seeing on the screen right now is the Korean stock market. And the reason why I'm showing you the Korean stock market versus the American stock market is because they're influencing each other quite a bit. Now, Korea has three of the biggest or rather two of the biggest memory manufacturers.

1:38 SK Hynix, which is the number one, and Samsung. And they provide pretty much all the memory that is required for the GPU makers, the CPU makers, or whatever type of AI infrastructure you can dream of. So, it's a very essential component. Now, the issue is it's down 32% and that is because there's a a of memory providers that have been dumping completely. So, if I pull up SK Hynix right over here, over the last month it's down almost 25%.

2:03 If I pull up SanDisk, right? SanDisk Corporation creates a different type of memory for AI. They're down 50%. This is the the darling stock, by the way, which was up like initially crazy. >> It's a meme coin, man. >> Yeah, I know. It was but it also pumped like a meme coin. It like I think it was up like something ridiculous, like 4,000%. >> And even if you look at the year chart >> Yes, let's look at the year chart.

2:24 >> Just to see what that >> Yeah, the year chart is still up 2,200%. So, even though it's down 50%, it's still >> >> 2,200%. >> Oh, this is so unbelievable. >> And so, the question on everyone's mind is why are these stocks, specifically AI stocks, specifically memory stocks dumping on the back of some amazing news, which is these companies are still pulling in more revenue, more profit, as you mentioned earlier, than they've ever done in a single quarter. SK Hynix released their quarterly earnings literally yesterday. I was reading it last night before in preparation for this episode. And Josh, to your point, their revenue increased by 354%, but their profit margins increased by 555%. They made more money in this quarter than they did in the entirety of 2025, last year. So, the question on everyone's mind is do these stock price movements make sense at all? The thing that like >> First of all, the answer is no, because I read through this earnings report, too, and I was immediately confused because how on earth do you have more net profit than revenue? It's 118% net margins. It's like unbelievable profit margins. It's an incredible business. It seems like it's doing remarkably well, and yet the market seems to just be kind of done with this. It's like It's like the the Toy Story meme where the they just kind of throw the toy out, like, "I'm done with this toy." But the numbers don't make any sense at all. Can I tell you why?

3:47 >> Yeah, please. Like, can Like, I'm trying to understand. Like, I'm reading through these notes here. I'm like, okay, why what's wrong here? Okay, so there's two reasons why I think SK Hynix or memory stocks, AI stocks in general, are dump picks. Number one, they analysts at these different firms on Wall Street or wherever set targets, right? So expected revenue growth for a lot of these companies. Now, SK Hynix technically missed their revenue target by around 1.7 billion dollars. You're seeing this on the screen right now. But what I would like to draw your attention to, Josh, is this. You see that?

4:20 That is 250% of revenue growth from the previous year. And that is almost 600% in operating profit from the previous year. So whilst they may have missed some random analyst projections, right? They've still excelled and compounded at a much more rapid rate than any other company in the world. It is extremely impressive. These things are printing money. But the question will should then be, why were the analysts even predicting this target in the first place, right? and I'll give you an answer for this. SK Hynix specializes in this thing called HBM, high bandwidth memory. We've spoken about this a lot on the show, right, Josh? Now, they are the number one provider of HBM. They only dedicate their chip fab capacity to create HBM.

5:06 Now, if you look at the other competitors, Samsung, Micron, they do HBM, obviously, but they also do some other cheaper memory for like your mobile phone or for your computer. It's called DRAM, and you're basically able to use that for other different types of gadgets. Now, because SK Hynix is so focused on HBM, they have run out of supply. So they can't possibly sell anymore. So they're selling off on the best news ever, which is they've sold out all their entire supply for this quarter. And so they have to move on to the next quarter supply in order to like get higher profit margins. So if you look at Samsung, if you look at Micron, they have dumped, but they've dumped less because they're selling more DRAM.

5:45 Have they made as much money as SK Hynix? No, because HBM is priced higher, uses more wafers. So, the whole thing basically is ridiculous. Like, SK Hynix has created a really good product. They have sold all of the product that they could potentially make in that single quarter, and because they've done that, they have now not been able to reach a specific target that this random Wall Street analyst has set for them because they've dedicated all their chip capacity to this specific thing. So, let me know if that makes sense, but basically it's it's crazy. If I had to like summarize it probably into >> three points. It's like, okay, the first one is the stock market rating. Like, a lot of the analyst ratings. Where basically, you could you could think of it like if you if you give your kid $100 for straight A's and they bring home like four A's and then an A- minus, and everyone like freaks out cuz they're like, "Oh my god, no, this isn't what you promised." That's the first thing. The second thing is the record earnings surprises. Like, basically, like you mentioned, they've fully sold out of their inventory for 2026. Yes. There is no more capability for them to sell more, by the way.

6:45 In 2027, there's no ability for them to sell more or sell it at a higher margin because it's already pre-sold. So, therefore, you eliminate a lot of the upside surprises, and there really is only downside surprises possible in the case one of these deals don't work out how they expect or things fall through. So, the upside is kind of capped in terms of surprises, downside is not. And the third is there is this two times leverage ETF that started trading just a couple weeks ago, July 13th. So, I mean, of course, being a Korean market, a lot of people it's funny like there's this this thing it's kind of known with the Korean stock market where they are the most aggressive gamblers per se.

7:19 They like to take on the most risk, and this two times leverage fund I'm sure fed right into that. So, there was a lot of leverage baked into the price of these stocks, and any sort of sell-off event creates this cascading liquidation event, and I'm sure we saw a lot of that as well with a 2x leverage stock. So, the convergence of those three things >> that, Josh, there's some news from Juken and this morning that JP Morgan. So, so you mentioned the leverage ETFs and you're right. Like this has led to a lot of the dump because those people like couldn't afford the stocks that they were buying.

7:48 They were too leveraged up. And JP Morgan this morning reported that the leverage ETF drawdown, the liquidation specifically is about 90% complete. So if you wanted to kind of like extrapolate, you'd probably see this bottoming sometime soon. So like this this drawdown can't go on forever and we're probably nearish a point where it's going to reach its bottom before like everything starts to like settle and maybe kind of like recoup. >> Yeah, well it seems like I mean we had this earlier in the year where there was that big sell-off. I remember Bill Ackman famously saying like hey, this is oversold, the market is wrong, you're overreacting. We are probably getting close to something like that now. Again, not financial advice, who the hell knows, but there are some signs that things are shifting. And I want to shift our attention to China now to talk about what's shifting over there cuz there is some You could say that China played a fairly large role in this and will continue to play a fairly large role going forward. What you'll notice is that we're not really talking much about the United States stocks. Like this is very much an international This is a global marketplace now because everyone is so interdependent on these supply chains and China has a very big one with memory. And there is a company I'm going to try to pronounce this right, Changxin Memory Technologies. Yes. CXMT is the ticker basically. And they had themselves a public IPO, a public debut in which they traded up 466% in one day, which is instantly the most valuable China-listed company ever, which is more than Alibaba or Tencent.

9:16 And they raised about $9 billion. So you're thinking, who on earth is this company? I've never heard of CXMT. Well, they're the world's number four DRAM maker. Now you'll notice we normally talk about the top three DRAM makers. A new entrant has entered the category and I have to ask, EJ, is this like a little concerning because there's more distribution of people who are able to make this memory. I mean, over the last 4 years, I believe, they've gone from a 1% market share to a nearly 10% market share, and it seems like that number is going up only.

9:48 They have the backing of China behind them. You know the Chinese CCP is going to be really pushing for them to win. Is this playing a role into the memory problem as well? >> Yes. And it's not as much of a problem as people make it out to be. So, let me actually ask you this question. Of the non-Chinese memory makers, so SK Hynix, Samsung, Micron, who would you think is the biggest region that they're selling all their memory to? Is it the West or is it China?

10:15 >> I would assume it's the West because we have all of these GPUs. >> You'd be right. It's overwhelmingly the West, and the issue there is there's not enough supply to meet the West demand, right? So, guess who is starved of memory? It's China. >> Mhm. >> China's starved of memory. >> open-source guys, huh? >> Yeah, those open-source guys. So, let Listen, they're not getting access to any of the American chips. Nvidia has a trade restriction. They can't sell them frontier chips, and they don't get access to any of SK Hynix and Samsung's memory chips because they're selling it to the West. Micron is obviously selling it to the West as well. So, they have to kind of do their own thing. That company is the number four memory provider now, CXMT, because, of course, Chinese AI labs like Moonshot creating Kimi K3, Jipu creating GLM, they also need memory for their GPUs to train their own AI models. So, CXMT stu- stepped up and basically IPO'd and went up 500% in a single day, making them the most valuable company in China. Their valuation, I think, right now is roughly around the price of Micron or the market cap of Micron. And they did that in like a single day, just like the craziest IPO ever. Now, the reason, again, for why this is the case is we are starved of memory in AI. It's just a very simple thesis. You need memory to remember everything that you type and talk to Claude and ChatGPT about. You need memory to keep your agents running 24/7, and that memory demand isn't just a linear line. I'm trying to not figure out what this looks like in the camera, but it's not a linear line. It is a completely exponential line. And if you look at the demand growth for any of these memory supplies, it literally looks like this. And you know what else looks like this? The profit margins and the revenue that we're seeing. So, it whether it it misses targets by like a billion dollars or not, it does not matter. one thing, but there's two other news items why China is causing stocks to crash, Josh. The other one, have you heard of this company called ASML based in the Netherlands?

12:10 >> yeah. Might might have heard that. Like singular company that the entire world is built upon. >> Yeah, yeah. Do you remember they create these like 300 million-dollar machines which are used by TSMC? >> Yeah. Yeah, exactly. EUV, extreme ultraviolet lithography. And they use this >> they shoot little pieces of of light at tin and then the tin turns into light that doesn't exist anywhere else on the planet. >> a banger episode. this crazy scientific company?

12:33 >> Yeah. >> >> Exactly. That was a good episode, by the way. For the OGs here who know what we're referencing. >> Go listen to that one cuz ASML's a crazy company. >> an awesome company. Anyway, so this company, it's one-of-a-kind, it's based in the Netherlands. They create these 300 million dollar machines. And I think they pump out a couple hundred a year. They're so hard to make. They have teams and teams of people trying to create these things. It is incredibly difficult to do and it's very secretive.

12:57 They have not released any blueprints. Such that it has been super hard to replicate this. They've tried many times in the West. Elon Musk has tried. You just haven't been able to do it. And it is pinnacle to have these machines to create next-generation AI chips. So, Nvidia, very close to ASML. Now, China, a company in China announced, very surprisingly, that they've been able to replicate a version of these 300 million-dollar machines. It's called DUV. It's called deep ultraviolet. So, it's not quite extreme, but it's deep ultraviolet. And I have to stress like it is a prototype machine. This hasn't been scaled. And let me ask you this, Josh. How many of these machines do you think they're creating for the rest of this year or in a year that they're targeting for a year?

13:37 >> Dude, not many. It's like low hundreds. >> Dude, no. It's five. Five. >> Oh Okay, that's way less than I thought. >> aiming for 10. So like >> Oh god. >> It's a nothing burger, but the market saw this news and were like, "Oh crap, China's about to flood the market with EUV machines. The the cost of all these GPUs is going to go down. We're going to have so many more GPUs. We should just dump Nvidia. We should dump AMD. We should dump all these memory stocks. It makes no sense. It is a massive overreaction.

14:04 >> Yeah, it seems like we have this baked-in trauma. I mean, there's like the the bear thesis is kind of like the solar panel idea where China famously they subsidized and then flooded the market with solar panels, completely collapsing prices everywhere. And because China's able to manufacture things at scale, they're able to kind of compete at a margin that other companies cannot. And the Chinese government is willing to back these companies and subsidize those companies in order to destroy the demand in other marketplaces. It's how China has always won. They've used their manufacturing capability and that connection with the government to subsidize these things to reach low prices that other companies cannot compete with. This is not the case for solar. This is just not really true and it's it's like a solar is this static technology. It is this commoditized thing, whereas memory is very dynamic. There's many different types of dynamic memory. There's many different ways of making it. There's very many different like custom architectures for it. And that's just not really how it works. You can't build a memory company to subsidize the prices of and and lower the cost relative to all the others because one, the demand is so high and two, there's so many different types. I mean, DRAM is kind of like tap water. That's kind of what they're going for. HBM is that premium bottled stuff. That's like that blue glass bottle that you see all the time.

15:18 Like they're totally different things and yet the market is reacting to this news as if they are the same. And I think that disconnect is probably where we feel a little optimistic and feel like perhaps this is a little bit over sold. Now, maybe from here we get into the kind of unbundling of this thing and talking about where the money is going cuz it's not just leaving the system. It is kind of shifting places. There is this unbundling of the AI trade happening. Maybe you could shed some light onto kind of where that's headed to now. The relative way to think about why the money is going >> in AI is going from the hyperscalers.

15:53 It's going from the top AI labs such as Anthropic, OpenAI, Google, Meta. They're spending copious amounts of money. I think the the figure for this year is something crazy like what was it like 250 billion dollars or something like this on AI capex alone. Like well Google just recently reported their quarterly earnings. Yeah, I think it might be more than that. It feels low when I said it. But Google's quarterly earnings reported that they've for the first time since they IPO'd, so 21 years I believe, they've gone negative cash flow, which means they're spending more money than they are taking in. The balances have been depleted, right? And the craziest part about this is they're doubling down even more because they see the opportunity. Now, think about it.

16:36 Google's doing this, Amazon's doing this, Meta's doing this. These guys aren't stupid people. Like they will will only be doing this if they see that there's real revenue coming through. Amazon CEO Andy Jassy famously said this in his previous quarterly earnings. He said, "We are investing all this money because we are literally getting revenue back from it almost immediately or 6-month delay. So, it makes sense for us to just keep compounding this, right?" So, the money is going from these companies into the semiconductor companies and that's what we're seeing.

17:04 That's why SK Hynix had a record quarter where they made the most money that they ever had, more than they did in 2025. So, the unbundling is this free cash flow going from the hyperscalers to the semiconductor companies. So, if you wanted to like kind of look at a layer to potentially consider investing in or kind of like being focused on. It still is, as boring as the answer is, semiconductors in general. The other thing that I think is playing into this, Josh, not to to bring up China again, but like we have to because they've been so relevant this week, is open source. A big critique from people right now is, "Huh, if I have an open source model I can run at home and is cheaper to run, why on Earth would I need to be spending millions and millions of dollars a year on AI? Why do I need all these GPUs?"

17:43 Well, that is fundamentally wrong, and Gavin Baker actually did a really good job explaining this. What he basically said number one, these models, these open source models, aren't cheap to run at home. If you look at Kimikate 3, it costs like between 1.1 to 1.2 million dollars to run effectively at scale. So, like it's not available to the average consumer. Not getting that on your home PC. >> Not at all. Number two, and this is a really good point. He said the hyperscalers or the the cloud service providers, like Google, like Microsoft, like Meta, like Elon Musk's SpaceX now, which is lending computer Anthropic at 1.2 billion dollars a month, they locked in really cheap contracts. Those contracts expire at the end of the year.

18:23 What do you think they're going to do after those contracts expire? They're going to re-rate it like 2x. And he makes the point here that like the spot prices for GPU rentals have not slowed down. They're 2x higher than the contracted rates that they were at the start of their contract. So, the point is whether you have an old GPU, whether you have new GPUs, whether Nvidia releases Vera Rubin in abundance, these GPUs are in such over demand that the prices for these things still go up. So, every fundamental building block for this unbundling, Josh, is just money going to semiconductor stocks and semiconductor companies, and I don't see any other way out of it right now.

18:55 That's what it looks like. >> Yeah, and as we talk about the GPUs, I mean we just had that that fun visual on screen here which shows you the anatomy of these chips and how nearly 50% of the costs are associated with this high bandwidth memory, and everything else takes up for 50%. So, the most important object in the world right now is a GPU. The most critical component of the GPU which accounts for about half of the cost is this memory. So it's like okay well we have a seemingly infinite demand for GPUs therefore infinite demand for memory therefore infinite demand for all of the supply of all of these companies like when they get re-rated it should go up. Right that that math just seems like it checks out.

19:34 So that's it feels like a little confusing as to why this is happening. Again there's a lot of external factors there's a lot of kind of overreactions baked in but it seems like as we're just looking at this like kind of pragmatically all of these numbers are checking out. Also it's insane as I'm looking at this that Nvidia sells these chips for $40,000 at an 84% gross margin like oh my god good for you man good for you. But there is this interesting like inverted capex thing happening where traditionally in technology for the last two three decades all of the funds have gone from the bottom up. So it's been from the consumers from the enterprise paying into these huge margin companies like Nvidia to raise their profit margins.

20:12 And for the first time we're having the reverse effect where all of the companies that have collected all this money over time like Google are now spending it for the first time ever in its history faster than it has made it. And the downstream effects of that seem to be pretty huge. And I mean we have Lisa Su she's here looking reading at Yahoo Finance actually who is just sharing the idea that AI adoption is faster than any of us thought and there is no end in sight at least from what we can see of where the demand for tokens is going to stop where the demand for all this compute is going to stop. And as of right now there's still this like tremendous shortage. If anyone was producing more memory I think it would just get eaten up right away and I think that's kind of the conclusion of this episode generally speaking is that like hey there's no end to the demand curve in sight and the more GPUs the more memory the more power we can apply to all this the better off everyone's going to be and the more hungry everyone's going to be to generate more tokens.

21:04 >> Yeah and just to be clear a point that Lisa Su makes in this clip that we're showing, on screen right now is it's not just general AI demand that is causing, GPU prices and memory stock prices to, or memory demand to to accelerate. It's also this thing called agentic AI. AI agents in general have exploded over the last couple of months. And guess what these AI agents need to be able to access tools, to do these tools, to orchestrate all the tasks, to run 24/7. You know, you see all these fun viral examples on Twitter where, you know, you set and forget a prompt and you go to come back and like you have a full triple A game like we saw this week with Claude Opus 5. It all requires CPUs. CPUs require a lot of memory. So, the point is as AI agents scale, you're going to need more of these fundamental things.

21:51 >> That's funny you should say that because speaking of agents, we have a sponsor for the show, Ledger, which we have to shout out here because if you are building with AI agents, you're worried about security. An agent with unchecked access is a brilliant but probabilistic teammate with access to important secrets. Ledger's Agent Stack fixes that with four open-source tools. DMK skills and CLIs for Ledger Wallet, Ledger Enterprise, and Ledger Enterprise multisig that let your agents work with crypto while you stay in full control.

22:17 You could tell your agent, "Rebalance my wallet. ETH to 40%. Do it in the fewest swaps." It does the math, drafts the swaps, and stops. The transaction lands on your Ledger signer and nothing moves until you approve it. Agents propose, humans approve, Ledger signers enforce. And, I mean, this is the best part, obviously, it works with cloud code, codex, and all those are all the places that you are actively working. and it's open-source ready today, available on developers.ledger.com.

22:41 Also linked in the description. And thank you so much to Ledger for supporting this episode. Now, let's close this thing out, EJ. We got to let people know, what do you do with all this information now? Like, okay, we've given you kind of this like idea. Here's what's happening. Here's the market demands. Here is the sell-off that's happening. We are in a currently in in that's worse than 2008, although it doesn't feel like it, probably because we speed ran it and because everyone's already up so much.

23:05 but like what what what do we do here? What do you do with this information? Okay, so here's my grounded take and you know, that's rich coming from the show where I'm quite quite optimistic. But I'm going to try and be grounded. The AI hype definitely drove markets to pretty insane valuations. And I'm not denying that. But what I will say is the future of AI and the economic value that it'll generate is just like the wild west right now. It is incredibly hard to predict and even conceive. There are many theories out there, many skeptics out there. So, the bet you need to make if you're listening to this is do you believe that AI LLMs, Claude, ChatGPT, that AI agents, that GPUs are going to be in absurd demand going forward. Do you think the demand for AI products, are you using AI more over the last couple of months? Like, you know, answer to that question. If you believe that scales out. Remember, it's only like something I said like 5% of the population that's even using AI LLMs beyond just a Google search. If you believe that scales, then you believe and bet that these semiconductor companies that are creating the fundamental materials, you know, we showed this on our screen earlier, to build these chips that are required, whether you like it or not, to run whatever types of AI models, whether this is open source or closed source, then you're betting that these companies are going to be in more in demand. And you're betting that these profit margins and revenue is going to increasingly grow.

24:31 And guess what? Those are the fundamental drivers of whether a company's stock price is going to go up. And I think that the market is completely unjustified right now. And I think we're going to look back on this in a year. I'm making my stake, I'm making my claim, I'm making my prediction, and we're going to think that these stock prices were at absurd valuations. >> Yeah, well, here's kind of how I think about it, too, is like on a personal note, I I'm more of an investor than a speculator. And that is why it's very easy to feel constantly optimistic. It's like I very firmly believe in the idea that we are going to need a lot more tokens, a lot more compute, a lot more energy over a long period of time. How long it takes to get there is unknown, but that doesn't really matter. If you have a low time preference where it doesn't matter if this takes 6 months or 6 years or 60 years, you just kind of directionally know where it's going to go, then making these bets and dealing with the volatility makes things much easier. This directionally feels like a trading opportunity. This is people who are selling off their profit. This is people who are positioning themselves to make a short buck. That doesn't need to actually be the case if you believe in the long term. And I think that's probably where we can wrap up this episode today. So, with that, yeah, thank you all for watching. That's the state of memory. It was crazy sad to find out that we lost more money recently than 2008. And we don't feel like it cuz clearly we've been printing a lot more dollars and those those numbers need to go up a lot higher to feel the same thing, but that is kind of where we stand. There is this discrepancy between memory stock prices and the actual demand for these items. And yeah, I think that's pretty much it. So, if you enjoyed the show, please don't forget to share it with a friend who might also enjoy rate us on your favorite podcast player. Leave a comment if we are too optimistic or if we need to dial things back a little bit or or if you disagree and why and what stock you are investing in and choosing to gamble on.

26:09 do you have any closing thoughts before we wrap this >> No, that is it. Thank you so much for listening and we will see you on the next one, guys. >>