The Memory Collapse JUST Started | Here's WHY.

Memory stocks are getting hit hard and they're taking the NASDAQ 100 down with it. Part of the reason that led to this 8.9% sell-off in Samsung stock has to do with expectations versus reality. But a whole another piece in my opinion has to do with the industry and the suits starting to talk about this. A shortage cycle or a a structural shortage in memory stocks does not prevent cycles. Now, I want to be clear, this isn't like, oh, everybody should be a bear on memory. There are some really important components and this is a very large research piece and we're going to go through the most important components of this uh right now. So, we're going to go through all of it. I've already read through all of it and highlighted. It's going to be great. But we've got to understand first before we hit the numeric piece about how the physics of a bottleneck don't prevent the cycle. And the real question is where does the cycle end and what are the warning signs for that end? Do we have any of those warning signs? Now we're going to talk about those. But all of that makes the most sense when we think about expectations versus reality first. So for example, Samsung stock just cratered in the Korean stock market 8.9%. That has now led uh Micron stock to drop about 8%. You can see most of this uh is just now occurring and still falling off a cliff at the time we're uh you know discussing this segment. SanDisk down about 11%. And it's taking other stocks down with it. The NASDAQ's going down 700 uh down to 703. We're down off 735 in our alpha membership last week. We argued 735. We're going down from here. We might be going down sub 700 soon. And that was a projection we made last Monday morning. And it's been straight down since last Monday morning. Now, that's not to be bearish. There are a lot of things going on. Part of it has to do with the memory sector. The other part has to do with Bessence's failed credibility. Topic for a different video. The other has to do with Jackson Hole, also topic for a different video. Although we do have a coupon code J-Hole for the Meet Kevin membership. Massive changes to the benefits and pricing structure coming. So you want to get in before that because you're going to lock in some lifetime uh access and some longer term access than than other people will. So get in before the 27th uh and uh of course then you've got a trade war with Donald Trump. So those are all in and Canada. So those are all things that could be accelerating some of the pain that we're seeing right now. We don't want to ignore those and pretend that those aren't also accelerants to what's going on. But let's be clear, uh, Iran and a trade war and Bessant and Worsh, those things don't bring Samsung stock down uh 8.7% in a day in Korean trading. This is very specific to the memory sector, which is exactly what we want to harp on right now. So, what happened with Samsung? Well, Samsung had earnings and the expectations got missed. It wasn't so much so that expectations just got missed. Okay, we missed on earnings a smidge. We I got it right here. We missed on revenue by 29%. Nominal. We missed on EPS by 1.7 7%. Also nominal. Does that really justify an 8.9% miss? Not really. This is something to know though for companies like Nvidia. Markets expect that market is going or that Nvidia is going to beat expectations. And so the way you have to look at Nvidia is not oh Nvidia is going to beat earnings. It's that we already expect Nvidia to beat earnings. You know these are the earnings expectations. Okay, we expect those to be low. The market's expectation is earnings are going to come in here. But markets expectations are built on historically they beat earnings by X percent. And so even if you miss like you could beat on guidance but you miss those market expectations stock craters. But it wasn't just the miss on earnings with Samsung. It was also something really important that they pointed out which Micron could end up going down the similar road. Buyback expectations. Markets had been expecting 100 to 200 trillion Korean Juan of buybacks for the company. They came out and said, "We're going to do between 90 to 110 buybacks and dividends." That's the low end of the expected range. And that really pisses markets off. So just to visualize that a little bit more, let's get the whiteboard and then we're going to go into some of this Namura research research on Micron. So if markets are expecting 200 to 100 trillion Juan, this is like 72ish billion to $140 billion of buybacks. If that's what markets are expecting, typically you look at the midpoint and the midpoint consensus estimate was 140 trillion. That's what people were looking for. This was the range and this was deemed realistic. They ended up coming in between 90 to 110. Uh some listed as 104 to 110. Uh but the point is it's way lower. The other thing that they're doing is rather than just doing buybacks, they're focusing more on dividends. And this has an optics issue. When a company buys back stock potentially high, they look like a freaking clown if the stock goes down. Now, obviously, the company can't predict exactly what's going to happen with their share price, but let's just put it this way. If you're Salesforce and you take on $25 billion dollar of debt to buy your own stock at $191, when the stock goes down to $160, you look like a freaking clown. And that's where I looked at the financials and I'm like, I'm a buy Salesforce in the 160s because they look like a freaking clown. And I think they're buying back the stock because they're really bullish. And even though it's down in the short term, it's going to go up in the longer term. And it has, you know, it's at like $210 now. It's really rebounded on this. But the optics are terrible. when you buy a stock for one level and and then it goes down even in the short term by doing a dividend. They're not buying back the stock at a certain price. So that could be a little bit of a tell that the company thinks we're closer to a cyclical top. So another thing that you know I've been talking about on the channel or or alpha membership is that everybody wants to get really rich in the stock market. The best time to really place your bets in the stock market is when you're closer to the bottom. If you're closer to the top, like let's say we're here, it's a lot harder to time the near top to the top than it is, you know, the bottom of the top. But the problem is like is the bottom here or is it here? Is it here? Right? So, it's like finding the bottom is harder and you got to have the balls to get through that pain. So, uh, you know, we could be more near a near top, right? Anyway, so that gives us a little bit of color on Samsung. Uh this uh expectation really got missed. And it's not just the optics that they don't want to do as many buybacks, but it's also this expectation that this number is way smaller than expected. Micron has expectations set by UBS that they could make enough money to retire as much as 40% of the company outstanding stock. And that comes from UBS because the company itself said, "Hey, you know, we're going to return most of our free cash flow to shareholders." And so then UBS uh made calculations and they're like, "Oh, they could buy back 40% of the stock." And then of course, you know, Twitter and FinTeit and Reddit just runs with the headline, which is, "Oh my gosh, they're going to buy back 40% of the stock." No, probably not. But the point is, if Micron now comes out and pulls a Samsung and says, "Hey, we're going to buy back 10% of the stock," that would be a really good announcement from a point of view of like, "Oh, they're doing big buybacks, right?" But it would be really disappointing from expectations. Now, part of the reason that Samsung did buy uh dividends to to great to a greater degree than buybacks has to do with their uh their other subsidiary companies and companies that have exposure to Samsung electronics that can only hold a certain percentage of shares, otherwise they get hit by the regulators. Some of that is just really weird technical regulatory bull crap. And so, that's how they're trying to brand it. I personally think they don't want to be suffering the optics of buying their stock back too high and so that's why they're preferring the dividends. But it makes for a good excuse to blame the regulators. Always makes for a good excuse. Uh okay. So now let's go into the no more research piece because that also applies to what's happening here in this memory selloff. And this will help us understand is this temporary or what are we up against. Okay. So what I wrote over here on this piece is econ 101. Okay. This time is not different. And and I don't want to again sound bearish. It's just a matter of when. Right? But at some point it all does come crashing down. Uh and what they're basically arguing is exactly that. They customers amplify the swings in chips. uh and memory stocks when chips are scarce over orderering or even double orderering to make sure they can secure the allocation. And when that scarcity eases, they'd rather part with their deposits and not necessarily fulfill all of the contracts they've committed to than risk being in a situation where they wish they had the contracts and they don't. Uh and so that's where they say the phantom demand when the cycle turns can be immediately cancelled and again and you get those non-refundable deposits that companies like a Sandis keeps you know Sandis got a lot of upfront capital by the way for and this happens with a lot of the hypers scale the neoclouds as well you know whether it's um an NBIS or whatever they get a lot of upfront money so they can sort of keep investing in infrastructure and people see that upfront money and and think, "Oh my gosh, this company has so much cash flow." But forget that that's actually a prepayment for contracts years out, they get cancelled. You're not getting those extra deposits again. So that does create a potential downside when those cancellations come. We're not saying those cancellations are coming. We're still in a massive shortage right now. So this is in the future. Uh so actually, let's take a look at where we are right now. This is where Nura thinks we are in the cycle right now. They think we're over here in boom. And they actually suggest that the shortage stage is not likely uh to turn down at all wi-i which is remarkable. They actually think we're going to stay in shortage for quite a while. They argue that the latest reading of what they call the CSI, which you know I call it crime scene investigation, but uh this has to do with their shortage index is at 103. And essentially they argue anytime you are above 100 you are in shortage for chips and anytime you're below 100 you have too many chips. And so we've been in these too many chip cycles before like for example March of 2023 or you could go back to 2014 or whatever. It's a regular sort of boom and bust cycle, right? Shortage pressures may fluctuate or ease at the margin but the evidence does not point to an approaching regime change in supply slack. In other words, they think that the shortage situation is going to continue. So that's actually bullish on shortage, right? This is why I'm saying like this isn't to be bearish. It's to look at facts. Let's look at some facts here. Let's look at some more. For most part of the past four decades, the semiconductor cycle has followed the global economy. But now they actually argue the semiconductor cycle is becoming its own source of basically GDP growth. that in my opinion is actually a risk factor. It's a risk factor because what's happening is you are accelerating the economy that was actually kind of bleeding towards recession. AI has really accelerated this economy into boom mode and that's great but if the AI trade slows then then the underlying economy has less of that sort of stimulus that has been getting. Uh, and so this is a a good point though that typically semiconductors follow the economy. GDP does well, people buy lots of chips and computers. Now it's people are buying a bunch of chips and computers and GDP is doing well. It's most of our GDP is driven by the AI trade. It's why Goldman Sachs thinks we would be in a recession if it weren't for um this uh uh you know boom in artificial intelligence. Global economy is in the midst of the largest investment boom since the 1990s. Uh and in Korea at least, housing prices are soaring again. Strong liquidity, twin super cycles of liquidity and housing. Physical bottlenecks are leading to massive pricing power. Unusual combination of multi-year AI demand and a supply base that is adjusting only slowly. AI spending no longer follows the replacement cycle for consumer electronics. Uh and those commitments change little within the year even as prices rise. So Nvidia just raised prices for their server chips by about 15%. And it's leading to expectations of oh my gosh this is going to lead to bigger margins at Nvidia. My expectation is a lot of those price increases are solely because of memory which that makes people really bullish on memory that higher prices can sustain for longer. In fact that's actually exactly what Namora research says. The latest NORA CSI reading stands at 103.8 8 in July, close to the highest reading in the sample, which suggests the chip market remains in deep shortage territory and that we could sustain high chip prices for longer than we could in past cycles. In other words, they're starting to make the argument, guys, guys, guys, this time is different. This cycle is going to last even longer than in the past. And part of that could be true. I actually think a lot of it has to do with Claude. Now, I'm not trying to oversimplify when they when I say these things. Uh I try to give you just my mindset of like here's why I think something is happening. I think Claude for coding was such an innovation that we mostly because of memory expansion, right? We kind of broke through the memory wall by getting access to a lot more memory for coding purposes. We were hallucinating way less finally in coding in larger code bases or for cyber security because we broke through at least the first blockade in the memory wall. There's still some memory wall obviously and and there are mass of massive shortages remain for high bandwidth memory. you need a lot more DRAM which is a commodity uh to make high bandwidth memory which is in steep you know shortage probably expecting high bandwidth memory prices to double next year it's a phenomenal product and it's very expensive and it's critical in those GPUs but I think a lot of the big memory boom came from coding in Q1 like March and then into Q2 April that was kind of our clawed moment the problem with that is I don't think we're going to have a clawed moment every year. I don't think it's a coincidence that Anthropic wants to IPO right after the clawed moment. And look at the chart, folks. Look at the damn chart. This is the memory chip shortage indicator. And it literally peaks right here at the moment of claw, the clawed frontier explosion for mostly coding. But then you also get the GPT catchup with codecs. Sorry for the crappy handwriting. This is why I usually like to type. The handwriting is just not that great. Uh but uh whatever. I don't think that's a coincidence. So I don't think it's a coincidence that Claude Co is probably what peaked us. I don't think it's a coincidence that Anthropic wants to IPO on the back of that because their year-over-year numbers now are going to look the best. their ARR at $65 billion as of the last day of July is really delicious for, you know, promoting an IPO. But are we going to maintain those growth rates? Well, you'd probably need another AI uh co-work. But where is that going to come from? Maybe it's going to come from healthcare. We've got a lot of talk about Mona with imunotherapy drugs for cancer now and their partnership with Merc uh which is very impressive. Really big fan. Let's keep going on the article though. Uh so the latest Nuro CSI reading for July is close to a record at 103.8. Recent monthly movements have been small which indicate there is no confirmed turn in the shortage cycle. So I want to be clear here. Namura this is why I said at the beginning this is not intended to be like a bear piece. Namura is telling you guys there's no sign this is turning around yet. And so that's why I write in my notes, right? But we are coming after the clawed explosion. So I think anthropic helps us measure a cycle peak unless something else comes to take over. What are some things that could take over? I think there are three things. Robotics, AGI, or some like major boom in healthcare AI could all happen. All of those things could happen. I personally think AGI is not going to happen. I personally think robotics are 15 years away and do healthcare AIs need to run on frontier models or can they run on you know Alibaba's Quinn? I don't know. Uh but that that is you know something to consider as well. So the model points to gains for pricing in memory stocks through 2027 and beyond. So like Namuras, notice how like they start with like, hey, this time's not different. We're still going to have a cycle, but this cycle can last longer than previous cycles. Okay, this is where I get into my micron numbers. So I know I shouldn't do a chart like this, but I do it anyway because I think this is a chart that represents investor psychology. Usually when you have earnings at a company that explode like this, okay, these are the percentage growth levels of of earnings per share growth at Micron earnings per share exploded like 900%. Uh you know 25 to 26 like this is an insane growth in earnings per share. This is just my quick little spreadsheet, right? The problem is, in my opinion, because like technically, you know, you're comparing to a low base, you could use a log uh chart and really smooth this chart out to get rid of some of this extreness, this exponential growth here. But the reason I like this chart is because I think this is your chart of investor psychology. the people yoloing into Samsung stock or Micron stock in June when everybody was leveraging up to the boobies. What ended up happening? People were leveraging up because they saw this nominal growth. But those numbers don't last. Even though Micron's revenue is expected to go up for the next four years, they're going to make more money on top of more money on top of more money. Their growth rate converges to zero. And that is a problem. So when does the stock peak? Well, does it peak closer to where we are now? You know, the the postcloud moment. Uh does it, you know, peak after the anthropic IPO? You know, this is crystal ball kind of stuff. But the point is the markets care about expectations. Remember how we started about expectations. The market is already expecting pricing to come down from memory. That doesn't mean it's bad for Micron stock, right? The markets already know memory prices are going to come down in the future. We know the growth rate, the second derivative, the growth rate of memory prices is turning negative. The question is not that it's going to come down. The question is how long is the cycle going to last? That's what Namora is saying as well. They're like, look, right now we're in massive shortage. This time is not different. The cycle will end. But that's not the problem. The problem is when we have the boom, the usual boom cycle looks like this. Can this cycle instead of looking like this look like this where you plateau longer? Okay, that right here, this fat so to speak, that is your reason to be bullish that the cycle can last longer. If these expectations get compressed because Anthropic stops spending as much money for example because remember high bandwidth memory this is kind of important small sidebar most enterprise artificial intelligence in my opinion and what we're seeing with like what we're doing with our artificial intelligence at reinvest most artificial intelligence for inference does not need as much memory as the frontier levels of training. This is in my opinion logical. The frontier models and training the frontier models, they need those Blackwell GB, you know, 200, 300s, the Vera Rubin stack. They need the latest and greatest massive memory stack to keep those frontier models going. Openweight models and enterprise applications, they don't need that much. They can operate on substantially smaller chips, older chips, in-house chips, openweight models that are more compressed. The more we use those, the more we compress the need for high bandwidth memory. It's still going to go up in price. It's still in shortage, but the more these openweight models come, the more you you reduce that plateau. And that's how the market trades for memory stocks. The market doesn't trade on, oh, memory prices are going to go down in the future. Yeah. No. Every jerk off in their mom knows memory prices are going to come down again in the future. We know the second derivative is going down. The whole cycle is dependent on what keeps this getting fatter. If robotics come out and all of a sudden we need to train, which eventually we will, all of these vision-based systems on uh you know for what's it called? um GPU vision um sorry how robotic vision that's where I'm going with all of the robotic vision that needs to be trained is going to use a lot of hardware a lot of GPU a lot of high bandwidth memory but if that's 15 years away we can have a big down cycle before that if that's a year away and the Optimus robot is coming hey we could scurve this up again that's how memory stocks trade and that's what makes it hard to to be exposed in debt to memory stocks because if you're margined up on memory stocks, you are setting yourself up for oopsy dupsies. Uh just like situational awareness did, you know, they were they were a a real lesson here. So, let me go to some of the other pages I have right here. Okay. Yeah. Yeah. Yeah. So, technically, we are right here with pricing. So you can actually see that the midpoint expected range here of export prices to a CSI shock is that we still have time where memory prices go up, the second derivative is going down because the line is flattening and then soon the first derivative will go down, right? The first derivative goes negative. Uh memory pricing starts coming down usually, you know, 2 to 3 years after the cycle starts. the cycle really started in about Q4 2025 and by 2029 you're just outright negative. That's already built in. It's just how fat can we make it. That's what memory stocks price off of right now and that makes them hard to to invest in with debt. Uh so just be very careful because the volatility will be massive because you're trading expectations. That's why Samsung stocks down 8.9 or 8.7%. AI and high bandwidth with memory have become large enough and sufficiently wafer intensive to tighten deliverable supply chains across the broader market. This is actually important because really what's happening is memory has become so critical that all components related to chip manufacturing are getting more expensive and pricing power is is going through the entire stack. Okay, what else do we have here? today's constraint harder to reverse wafer allocation basically it's taking more time uh and these shortages can persist longer again they think between through the end of 2027 and beyond however they do flag that historically double ordering becomes very normal and when that double ordering gets cancelled you get a massive reversal in momentum and that's not ideal now could that actually benefit Nvidia that's another argument I could make the argument that Nvidia could actually benefit from memory stocks coming down because they could keep uh GPU prices elevated but actually uh make more money because they're spending less on cost of goods sold for the memory supply. It's possible. Uh this is okay. Then we talk a little bit about Moore's law here. Typically uh chips are a source of deflation because transistor density doubles every two years or slow or so. Every generation of miniaturaturization yielded chips that were faster, more numerous and cheaper per unit. This is why semiconductor prices come down for decades. Uh and they are a persistent source of deflation. However, because of the way high bandwidth memory is manufactured that it takes so much of this DRAM that you stack together, uh you're creating a structural shortage that's just going to last longer than previous cycles. Still going to be a cycle, but going to last longer. In the first half, year-over-year, memory prices increased 135%. And if you just look at the second quarter postcloud, memory prices were up about 400%. Now, I think that's actually a really important note to make is that the highest gain in memory prices actually came immediately after and sort of during the anthropic claude um moment, if you will. That's important because it reiterates this concern of what is the catalyst that keeps it pumping. Let's go see if we could find the price increases. They're listed somewhere here in in the Micron statement. and I'll pull them up, but just from memory, we went from like, oh yeah, prices went up 110% year-over-year or whatever it was to prices went up 400% or something ridiculous. Uh, let's see here. Oh, okay. Yeah, yeah, yeah. Okay. Well, this is this is the Micron example. Ready for this? Sales of DRAMM increased 343% primarily due to low 260% increases in selling prices and mid 310% increases in average selling prices for NAND. So not exactly 400% but my my point is Q2 this is the micron Q2 piece Q2 was even more extreme than Q1 like substantially more extreme. These numbers if you pull up the Q1 were at like mid 110 range is what they wrote for this. So most of that price increase that they're talking about in the no more research piece uh aligned with the second quarter not the first quarter which is again important because it helps you pinpoint why claude. Okay. Uh what else do we have here? Uh actually I think oh I just got to the end of my no more piece. That is my no more research. So what are the bottom lines of that? Like what what does that mean? Does that mean you know this this stock selloff is durable? We're going to hell. No, it doesn't mean the cycle's over. We got a lot of crap going on in the market today. It's going to create buy the dip opportunities this week, you know, thoroughly with Jackson Hole coming up. We'll be live streaming that obviously. Uh and of course we'll talk some memory bottom lines here. Just a quick note, you can use that coupon code J-Hole. That expires on August 27th. That'll be a pretty big expiration because we're including some more benefits for members who join before that date. Uh and and then there'll be a big price increase as well. So remember, joining could be a tax write off and you get our uh research and our alpha report uh and trade alerts uh before along with the access to all the courses. It's pretty awesome. So check that out over at mekevin.com. But bottom line for memory, where do we sit bottom line on memory? Bottom line for memory, you're going to be in a shortage for a while. Uh but at some point we need expectations to keep pumping the cycle to actually enable those like 40% micron buybacks to happen. Otherwise the companies will end up leaning conservative. Say well let's not have bad optics at buying our stock back too high. Uh we'll do some dividends instead. uh we'll keep a little bit more capital for debt payoff which is smart or uh you know capex. The issue is all these companies are building out uh machinery and capacity based on their expectations of future demand not anal or in my opinion not realizing how every single memory related company is trying to throw money into expanding capacity to capture that demand that we have today. And that's usually how you end up getting into an over supply because it's not like they coordinate all of their buildouts. Uh even even if you know some brand that they do, they don't. So uh anyway, I think that's a useful primer on what's going on with memory cycle. If you're really bullish on memory, in my opinion, it should be because you expect that fat portion of the curve that that plateauing to either last longer or you see scurves that that maybe I don't. again robotics, healthcare, AGI, I think these are a little further out than markets expect today and and that creates some downside risk for these plays. Uh SanDisk is sort of lumped into this even though it's a storage play. Uh you can check that uh uh my full Samsung or SanDisk video. Uh search it on YouTube, you know, meet Kevin Sandis and you'll see. But that gives you a full-on breakdown on memory. >> Why not advertise these things that you told us here? I feel like nobody else knows about this. We'll we'll try a little advertising and see how it goes. >> Congratulations, man. You have done so much. People love you. People look up to you. >> Kevin Praath there, financial analyst and YouTuber. Meet Kevin. Always great to get your take.