원문 보기
Memory stocks are taking a major hit, dragging the Nasdaq 100 down. Part of the reason Samsung's 8.9% selloff is the mismatch between expectations and reality , but another important reason, in my opinion, has to do with the industry and the fact that executives are starting to talk about it . A cycle of scarcity or structural scarcity in memory stocks does not negate the cyclical nature of the market.
Now I want to clarify. This doesn't mean, "Oh, everyone should play memory degradation." There are really important components, it's a very large research paper, and we'll go through the most important parts of it now. So, we'll work through all of this. I've already read all of this and highlighted the main points.
This will be great. But we need to understand first, before we get to Numera's material on how the physics of " bottlenecks" does not prevent cycles. And the real question is: where does this cycle end? And what are the warning signs of this ending? Do we have such signs now? We will talk about them.
But all of this is best understood if you first think about expectations versus reality. For example, Samsung shares just crashed 8.9% on the Korean stock market . This caused Micron shares to fall by about 8%. You see, most of this is happening right now, and they're still flying into the abyss as we discuss this segment.
SanDisk is down about 11%, and it's dragging other stocks down. Nasdaq falls 700 points to 703. We fell from the 735 mark. Last week in our Alpha membership we argued that after 735 we would go down. We could drop below 700 soon, and that was the forecast we made last Monday morning, and we've only been going down since then.
This does not mean tuning into a bearish trend. A lot of things are happening. Some of them are related to the memory sector. The other part has to do with the lost trust in Bessant, that's a topic for another video. Another one related to Jackson Hole, also a topic for another video, although we do have a J Hole coupon for a Meet Kevin membership.
Large-scale changes in the structure of benefits and prices are approaching . So you should join this because you will be able to get lifetime access and a longer access period than other people. So, make it to the 27th . And, of course, there's the trade war with Donald Trump. So, that's it...and Canada.
So these are all things that could be exacerbating the pain that we're seeing right now. We don't want to ignore them and pretend that they are n't also fueling what's happening. But let's be frank. Iran, the trade war, Bessant and Warsh—these things don't make Samsung shares fall 8.7% in a day on Korean exchanges.
This is a specific situation for the memory sector, which is exactly what we want to emphasize. So what happened to Samsung? Well, Samsung published a report, and expectations were not met. It's not so much that expectations simply weren't met, okay? We just missed our profit projections. Here, I have the data here.
We missed out on 0.29% of revenue, nominally speaking. We missed out on 1.77% of earnings per share, also nominally. Does this really justify an 8.9% drop? Actually, no. But it's worth knowing for companies like Nvidia. Markets expect Nvidia to beat all forecasts. Therefore, you shouldn't look at Nvidia through the prism of "Oh, Nvidia will beat earnings expectations." And because we already expect Nvidia to surpass these figures.
You know, that's what earnings expectations are. Okay, we expect them to be low. Market expectations are that earnings will be at this level. But market expectations are based on the fact that historically they have exceeded earnings by X%. So even if you beat forecasts , you may not meet market expectations and the stock will crash.
But it's not just about Samsung's lost profits. There was also something really important that they pointed out that could end up similarly for Micron. Expectations regarding share buybacks. Markets expected the company to repurchase 100–200 trillion Korean won. They came out and said they would do a buyback and pay a dividend of between 90 and 110.
That's the lower end of the expected range. And this is very annoying for the markets. So, to better visualize this, let's take a whiteboard and then analyze Nomura's research on Micron. If markets expect 100–200 trillion won, that's about $72 billion to $140 billion for share buybacks. If this is what the markets expect, they usually look at the average.
And the consensus forecast for the average value was 140 trillion. This is what people expected. This was the range, and it was considered realistic. In the end, they got from 90 to 110. Some indicated from 104 to 110. But the point is that it is much less. Another thing they do is this: instead of buying back shares, they focus more on dividends.
And from an image perspective, it looks ambiguous. When a company buys back shares, perhaps at an inflated price, it looks like a complete clown if the stock price falls. Of course, a company cannot accurately predict what will happen to its shares, but let's put it this way. If you're Salesforce and you take on $25 billion in debt to buy back your own shares at $191, and they fall to $160—you look like a complete clown.
That's why I looked at the financials and thought: I'll buy Salesforce when the price is around 160, because they look like clowns, and I think they're buying back shares because they believe in growth. And although they have fallen in the short term, they will rise in the long term. And so it happened, now they cost about $210.
They really played that fall. But the image is terrible when you buy stocks at one price and they fall, even in the short term. By paying dividends , they are not tied to buying back shares at a certain price. So, this could be a signal that the company believes we are closer to a cyclical peak. Another thing I've talked about on the channel or in Alpha membership is that everyone wants to get rich quick in the stock market.
The best time to bet on the market is when you are closer to the bottom. If you are closer to the top, say, like now, it is much harder to guess the peak than the bottom. But the problem is , where exactly is the bottom: here, here or here? So finding the bottom is harder, and you need nerves of steel to get through this pain.
So we might be closer to a local peak, right? In any case, it gives us some insight into Samsung. Their expectations were not met, and it's not just about the image due to the reluctance to buy back shares, but also because this figure is much lower than predicted. Micron has expectations from UBS that they can make enough money to buy back up to 40% of the company's outstanding shares, and this is data 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 UBS did the calculations and decided: "Oh, they can buy back 40% of the shares." And of course, you know, Twitter, FinTwit, and Reddit are picking up the headline: "Oh my God, they're going to buy back 40% of the shares." No, most likely not.
But the point is, if Micron now follows Samsung's lead and says, "Hey, we're going to buy back 10% of the shares." That would be a really good announcement in terms of, "Oh, they're doing big buyouts, right?" But that would be very disappointing given the expectations. Part of the reason Samsung has paid out dividends more than it has repurchased shares is because its other subsidiaries and firms with stakes in Samsung Electronics can only own a certain percentage of the shares or face regulatory sanctions.
Some of this is just weird technical regulatory nonsense. And that's how they try to present it. I personally believe that they don't want to give the impression that they are buying back their shares at an inflated price. That's why they prefer dividends, but that's a good excuse to blame regulators.
This is always a good excuse. OK. So now let's move on to the Nomura study, because it also relates to what's happening right now in the sell-off in memory stocks. This will help us understand if this is temporary or what we are facing? Good. So, what I wrote here is the basics of economics, you know?
This time it's the same. And again, I do n't want to sound pessimistic. It's only a matter of time, right? But at some point, it all collapses. And they, in fact, claim exactly that. Customers amplify fluctuations in the chip market and memory sector stocks when chips are in short supply. This is called over- ordering or even double- ordering to guarantee you get your share.
And when this deficit decreases, they would rather sacrifice their deposits than fulfill all the contracts they agreed to, so as not to risk finding themselves in a situation where they regret having these contracts. And so that's where, as they say, ghost demand, when the cycle reverses, can be immediately canceled.
And again , you get those non-refundable deposits that companies like Sandisk keep. You know, Sandisk got a lot of upfront capital, by the way, and that happens with a lot of hyperscalers and clouds, you know, whether it's Nbase or whoever. They get a lot of upfront funding to continue investing in infrastructure.
People see these upfront funds and think, " Oh my God, this company has so much cash flow." But they forget that this is actually a prepayment for contracts for years to come. If they are canceled, you will no longer receive these additional deposits. So that creates a potential downside when those cancellations come.
We are not saying that these cancellations will happen . We are currently in a state of huge deficit . So this is a matter of the future. So, actually, let's look at where we are now. That's where Nomura believes we are in the cycle right now. They believe we are here in a boom phase. And they actually assume that the deficit stage is unlikely to decline at all.
Which is quite strange. They really think we'll be in deficit for quite some time. They claim that the latest value of what they call CSI—I call it " crime scene investigation," but it's their shortage index— is 103. And they're basically saying that when it's above 100, you have a shortage of chips.
And if it's below 100, you have too many chips. So, we've been in these chip glut cycles before. Like, for example, in March 2023, or you can mention 2014 or any other year. It's a normal cycle of ups and downs, right? Deficit pressures may fluctuate or ease slightly, but the data do not indicate an imminent regime shift toward supply-side easing.
In other words, they believe that the deficit situation will continue. So this is actually an optimistic forecast for the deficit, right? That's why I say it's not to be pessimistic. This is to look at the facts. Let's look at some facts here. Let's take a look at some more . For most of the past four decades, the semiconductor cycle has followed the global economy.
But now they are actually arguing that the semiconductor cycle is becoming its own source of GDP growth. In my opinion, this is actually a risk factor. This is a risk factor because you are actually accelerating an economy that was just about to slide into recession. AI has really put this economy into boom mode, and that's great.
But if demand for AI slows, the underlying economy will have less of the stimulus it was receiving. And this is a valid observation, because the semiconductor market usually follows the economy. When GDP increases, people buy lots of chips and computers. Now people are buying a lot of chips and computers, and GDP is growing.
A large part of our GDP is driven by the field of artificial intelligence. That's why Goldman Sachs believes we would be in a recession if it weren't for this AI boom . The global economy is in the midst of the largest investment boom since the 1990s. And in Korea , at least, housing prices are rising rapidly again , there is strong liquidity, dual liquidity and real estate supercycles.
Physical limitations lead to a colossal price advantage. An unusual combination of long-standing demand for AI and a supply base that is only slowly adapting. AI spending is no longer tied to the consumer electronics refresh cycle. And these obligations change little over the course of the year, even despite rising prices.
So Nvidia just raised the prices of its server chips by about 15%, and that creates expectations of, "Oh my god, this is going to lead to higher profits at Nvidia." I expect a significant portion of these price increases to be driven solely by memory, which makes people very optimistic about the memory market.
Higher prices can last longer. Actually, this is exactly what the Nomura study claims. The latest Nomura CSI in July is 103.8, close to the highest value in the sample, suggesting that the chip market remains in a state of deep shortage and we can sustain high prices for longer than in past cycles. In other words, they start to prove: “ Guys, this time it’s different.
This cycle will last even longer than in the past ." And that may be partly true. I actually think a lot of it has to do with Claude. I'm not trying to oversimplify when I say these things. I'm just trying to explain my approach to you: "Here's why I think something is happening." I think Claude for programming was such an innovation, mainly due to the memory expansion, right?
It's as if we've broken the memory barrier, gaining access to a much larger amount of memory for programming purposes. We finally started hallucinating much less when writing code in large codebases or in cybersecurity, because we had overcome at least the first barrier—the “ memory wall.” Of course, the " memory wall" still exists, and there is still a huge shortage of high-speed memory.
Much more DRAM, a commodity , is needed to create high-speed memory, which is currently in great shortage. We should probably expect prices for high-speed memory to double next year. This is a phenomenal product, it is very expensive and critical for these GPUs. But I think a lot of this memory boom was driven by programming in the first quarter, like March, and then in the second quarter, in April.
It was our kind of " Claude moment." The problem is, I don't think we'll have that kind of "Claude moment" every year. go IPO right after its " Claude moment." Now look at the graph, friends. Look at this damn graph. This is an indicator of a shortage of memory chips. And it literally reaches its peak at the very moment Claude appears.
Claude's explosive growth is largely due to writing code. But then you also get the GPT catch-up from Codex. Sorry for the terrible handwriting. That's why I usually prefer printing. My handwriting is just not very good. Well, but it doesn't matter. I don't think this is a coincidence. So, I don't think it's a coincidence that working with Claude probably brought us to the peak.
do an IPO against this backdrop. Because their annual figures will look their best now. Their ARR of $65 billion as of the last day of July— that's pretty tasty for, you know, promoting an IPO. But will we be able to sustain this rate of growth? Well, you're probably going to need another such "AI collaboration" moment.
But where will it come from? Maybe it will come from the healthcare industry. We're hearing a lot about Moderna with their cancer immunotherapy drugs right now and their partnership with Merck, which is very impressive. I'm a really big fan. But let's continue with the article. So, the latest Nomura CSI indicator for July is close to the record—103.8.
Recent monthly fluctuations have been small, indicating that there is no definitive reversal in the deficit cycle. I want to clearly dot the "i's". Nomura—that's why I said at the beginning that this wasn't intended as pessimistic material. Nomura is telling you guys that there are no signs that things are changing for the better right now.
And that's why I write this in my notes, right? But we are already after the explosion of Claude's popularity. I believe Anthropic will help us determine the peak of the cycle, unless something else takes over. And what could prevail? I think there are three things. Robotics, AI of the general level (AGI), or a major breakthrough in medical AI—all of these could happen.
Personally, I don't believe in the emergence of AGI in the near future, I think robotics is still 15 years away, and I don't know whether medical AI needs advanced models or whether solutions like Alibaba's will be enough. But this is also worth considering. Therefore, the model indicates an increase in memory manufacturers' stock prices until 2027 and beyond.
Notice how Nomura analysts start by saying that this time things are different , the cycle will continue, but it may be longer than previous ones. Here I will move on to my numbers regarding Micron. I know I shouldn't show such a chart, but I do it because it reflects investor psychology. Usually when a company's profits skyrocket like this , you know?
These are the percentage levels of earnings per share growth at Micron. Earnings per share soared by about 900% between '25 and '26. This is just crazy growth in earnings per share. This is just my quick note, of course. The problem, I think, is that technically we are comparing this to a low base. You could use a logarithmic plot to smooth out these extremes and remove some of this exponential growth.
But I like this chart because, in my opinion , it shows investor psychology. What happened to those who went "all in " on Samsung or Micron stocks in June, when everyone was taking maximum leverage? People took out leverage because they saw this nominal growth. But such numbers don't last forever. Although Micron's revenue is expected to grow for the next 4 years.
They will earn more and more money. But their growth rate will approach zero. And this is a problem. So, when will stocks peak? Will it be closer to the present moment ? You know, that post-cloud era ? Will the peak come after the Anthropic IPO ? This is a question of fortune telling on coffee grounds.
But the main thing is that markets care about expectations. Remember how we started talking about expectations? The market is already pricing in the decline in memory costs. That doesn't mean Micron's business is bad, right? Markets already know that memory prices will decrease in the future . We know that the growth rate, the second derivative, the growth rate of memory prices becomes negative.
The question is not whether they will decrease. The question is how long this cycle will last . Nomura also says this . They say, " Look, we have a huge deficit right now." This time it's the same. The cycle will end." But that's not the problem. The problem is, when we have a boom, the boom cycle usually looks like this.
Could this cycle, instead of looking like this, look like this, where the plateau is longer, you know? This, in fact, this "thickness", so to speak, is your reason to be optimistic. That the cycle may last longer. If those expectations are compressed, because Anthropic, for example, will stop spending so much money, because remember, high-speed memory is important.
A small digression. Most enterprise AI, in my opinion, and what we see as we work with AI at Reinvest, most AI doesn't need as much memory to infer data as it does to train advanced levels. This, in my opinion, is logical. Advanced models, during their training, need these Blackwell GB, you know, 200s, 300s, Vera Rubin stacks.
They need the latest and greatest massive memory stack to support these cutting-edge models. Open-scale models and enterprise applications don't require that much. They can run on much smaller chips, older chips, proprietary chips, open- source models that are more compressed. The more we use them, the more we reduce the need for high-speed memory.
The price will still increase. The deficit is still there . But the more of these models with open weights appear, the more you reduce that plateau. And that's exactly how the market trades memory stocks. The market doesn't trade on the basis of, "Oh, memory prices will drop in the future." Yes, every fool and his mother knows that memory prices will fall again in the future.
We know that the second derivative becomes negative. The whole cycle depends on what allows this to become bigger. If robotics comes along and suddenly we have to teach what we're going to end up doing , all these vision-based systems for, uh, what's it called? Um, GPU- star. Um, excuse me, what about robotic vision?
That's what I'm getting at. All the machine vision that needs to be trained will require a lot of hardware. Lots of GPUs , lots of high- bandwidth memory. But if this happens in 15 years, we may face a major recession before then. If this happens in a year and the Optimus robot is ready, we can go up sharply again .
This is how shares of memory companies are traded. And that's why it's hard to hold debt in these stocks: if you use leverage, you risk making a big mistake, like Situational Awareness did—it was a real lesson for everyone. So, let me move on to the other pages that I have here. Okay, yes, yes, yes.
Technically, we are right here with pricing. You can see that the average expected range of export prices under the CSI shock suggests that memory prices will continue to rise for some time. The second derivative is falling because the line is getting flatter, and soon the first derivative will start falling too , right?
The first derivative becomes negative. Memory prices are starting to drop. Usually, you know, two to three years after the cycle starts. The cycle actually began around the fourth quarter of 2025. And by 2029, the indicators will become frankly negative. This is already included in the forecasts. The only question is how big will this profit be?
This is what currently determines the share price of memory companies. And that makes them difficult to invest in debt. So be very careful because the volatility will be huge because you are trading expectations. That's why Samsung shares fell 8.9 or 8.7%. AI and high-bandwidth memory have become so significant and require so many wafers that it has caused supply chain constraints in the broader market.
This is actus that memory has become so critical that all the components involved in chip manufacturing are becoming more expensive. And the price advantage extends to the entire chain. Good. What else do we have? Today's restrictions are harder to overcome. Distribution of plates. In general, it takes more time.
And this deficit could last longer. Again, they believe this will continue until the end of 2027 and beyond. However, they note that historically, double orders are becoming the norm. And when these double orders are canceled, there is a massive momentum reversal, and that is not a good thing. Could this actually work to Nvidia's advantage?
This is another argument . It could be argued that Nvidia actually benefits from the decline in the stock prices of memory companies, as they will be able to keep GPU prices high while earning more by spending less on the cost of supplying memory. It is possible. Uh, that's normal. Now let's talk a little about Moore's Law.
Typically, chips are a source of deflation, as transistor density doubles every 2 years or so . So each generation of miniaturization produced chips that were faster, more numerous, and cheaper per unit. That's why semiconductor prices have been falling for decades. And they are a constant source of deflation.
However, the nature of high- bandwidth memory production, which requires such large quantities of DRAM stacked in stacks, is creating a structural shortage that will last longer than in previous cycles. Cyclicality will remain, but it will last longer. In the first half of the year, memory prices increased by 135% year-over-year, and if you look at the second quarter after the release of Claude, memory prices increased by about 400%.
I think this is a really important observation: the biggest increase in memory prices happened right after and, you could say, during the Anthropic Claude moment. This is important because it raises the question again: what is the catalyst that is supporting this growth? Let's see if we can find those price increases.
They are listed somewhere here in the Micron report. I will open them. Just for memory . We went from " oh, prices are up 110% year over year" or something like that to them being up 400% or something absurd like that. Let's look here. Oh, okay. Yes, yes, yes. Good. Well, this is an example of Micron.
Ready for this? Ahem. DRAM sales increased by 343%, primarily due to a 260% price increase, while average NAND prices increased by 310%. So, not quite 400%, but my point is that this is Q2, this is Micron's second quarter financial report . The second quarter was even more extreme than the first. Much more extreme.
These numbers, if you look at the first quarter, were in the 110% range—that's what they wrote. So, most of the price increases reported in the No More Research report occur in the second quarter, not the first, which is again important because it helps to identify the cause. Claude. OK. What else do we have here?
Oh, actually, I think I just finished my "Enough" material. This is my research on "Enough". So what are the main takeaways from this? So, what does that mean? Does this mean that the stock market is going to hell? No, this does not mean that the cycle is complete. There's a lot of crap going on in the market today .
This will create buying opportunities on the dip this week, given the approach of Jackson Hole. We will, of course, be broadcasting live. And of course, we'll discuss the main findings regarding memory, but a quick note: you can use the promo code J Hole. It is valid until August 27. This will be a significant milestone as we add more benefits for members who join by this date.
And then there will also be a significant price increase. So remember, your membership is tax deductible and you will receive our research, Alpha report, trading alerts, and access to all courses. This is very cool. So check it out on mekevin.com. But what is the main conclusion about memory? Where are we now?
The main conclusion regarding memory : the deficit will continue for some time. But at some point we need expectations to continue to stimulate the cycle to actually allow for that 40% Micron buyback. Otherwise, companies will ultimately act conservatively. They will think, "Well, it's not worth creating a bad impression by buying back shares at an inflated price." We'd better pay dividends.
Let's leave a little more capital for debt repayment, which is reasonable, or for capital expenditures. The problem is that all of these companies are building capacity based on expectations of future demand. In my opinion, they don't realize how every memory-related company is trying to invest money in expanding capacity to meet the demand we have today.
And that's how oversupply usually occurs, because they don't coordinate their development plans, even if someone thinks they do. This is not true. So, anyway , I think this is a useful introduction to what happens with the memory loop. If you're really optimistic about memory, I think it must be because you expect this part of the curve —this plateau—to last longer, or you see S- shaped curves that maybe I don't see.
Again, robotics, healthcare, AI— I think it's a little bit further out than the market expects, and that creates some downside risks for those areas. SanDisk is somehow included in this category, even though it is a company specializing in data storage. You can watch my full video about Samsung and SanDisk.
Search for it on YouTube. This will be Kevin's video about SanDisk. But it provides a complete breakdown of the memory marke.
Comments 0