저자: AndrewPaulo · 외국인
요약 보기
마이크론·SK하이닉스는 선행PER 약 6.4배로 S&P500(20~21배) 대비 극단적 저평가. 하이퍼스케일러(MSFT·META·AMZN) 실적이 AI 데이터센터 capex 지속을 확인해 메모리 사이클 종료 우려를 불식시킬 촉매라고 주장. 칩 수요는 GPU당 메모리 요구량 18개월마다 2~3배 증가, 장기 공급계약 약 50% 확보 근거로 제시. 최근 800% 상승 후 조정 중이며 가격이 '고무줄처럼' 급등할 것으로 전망.
다듬은 글
Hey guys, check this out. We have Hyperscalers presenting earnings this week as well as SKH Highix. And this could be a massive catalyst for memory chips like SKHix and Micron. So, we're going to talk about exactly why. Please like and subscribe. First of all, if you guys don't want to stick around for the entire video, bottom line is, you know, the sector can only get so much good news until the price of these stocks finally go up to reasonable levels.
And I've talked about it in various videos. Guys, look at Micron. It's at a forward PE of 6.42. The S&P trades at about an average of a forward PE of about 20 to 21. So, this is extremely undervalued. So, we're going to talk about exactly why, you know, what the market is scared of and why these earnings coming up this week could actually be a massive catalyst finally proving to people convincing investors that the cyclicality of the memory sector is done for at least the next 3 4 5 years which could eventually lead to a massive increase in the price of these stocks and I do think it will rubber band right so we've seen the price of SKHix and Micron go up rapidly.
I mean, we've seen, you know, 800% returns in the last year. The market's kind of taking a breather. People are taking profits. People are skeptical as to whether or not, you know, this is sustainable. And we've had this recent dip. But I do think that, you know, the market all at once will essentially accept the fact that, you know, this AI buildout is here to stay.
It's going to last many years. And when they do realize that, I think the price of both of these stocks is going to rubber band really quickly to the upside, guys. Also, let me know how you guys are doing in the market this week. My portfolio, it's down slightly for the week, down 1.16%. It's been a rough couple of weeks.
Still did slightly outperform the S&P 500 and still, you know, slightly under 1.2 million there. So, we're at 1.174. Let me know how you guys are doing in the market this week. But I want to jump into it, guys. And really quickly, if you guys want to see what's in my portfolio, you can join the Patreon link is in the the description of this video.
You can see what's in the portfolio and I show you the method that I use to analyze stocks and just highle how I'm building wealth through stocks in this market as well. So, let's jump into it, guys. Hyperscalers this week, we have Microsoft uh presenting earnings on Wednesday, July 29th. Meta as well on the same day and then Amazon on July 30th.
Now obviously these are massive customers of both Micron and SKH Highix which have led to massive demand for the chips and massive pricing power for SKH Highix and Micron as well. Now we all know what's going to happen, right? We've had good news for the last couple of earnings quarters and we expect more of the same, right?
So, you know, let's move beyond that cuz we know, you know, all of these companies essentially are going to say, "Hey, you know, we're continuing to spend on capex. We're continuing to spend hundreds of billions this year." But what could they say and what could, you know, they project further into the future that could calm the nerves of investors and finally bring this forward PE up, you know, to the point where SKH Highix and Micron could massively explode to a double or triple really, really quickly.
Also want to note, Apple is going to have earnings on July 30th as well. Now, although they're not, you know, buying HBM's high bandwidth memory like these hyper other hyperscalers, they do buy memory that goes into iPhones, you know, MacBooks. You guys have seen the headlines where, you know, they had to increase the price of their computers and I think their iPads as well because memory is becoming so expensive.
And it's estimated that memory is about 30 to 35% of computer prices just in general, right? So memory is becoming a massive part not just of AI buildout but of you know consumer electronics as well. So you know if Apple comes out and says something like hey you know we're increasing the memory in our phones you know for ondevice AI compute that's something we expect to see eventually.
We're not sure if it's going to be this quarter but that's going to lead to more demand as well. And I want to bottom line you guys before I jump into the details here. A lot of the, you know, one of the things people aren't talking about is the fact that, you know, the demand for memory chips isn't just a linear, you know, graph going up with, you know, the buildout of AI data centers.
It's actually been going up exponentially because these GPUs that are used in these data centers, they become more and more complex and they demand more and more memory per individual GPU. And it's been estimated that over the last 18 months and every 18 months or so, the requirement for memory per GPU has been doubling to tripling every 18 months and it's expected to do so for the foreseeable future.
So what does that mean? Right? Even as hyperscalers continue to build out these data centers, we expect those to be mostly built out in the next 3 to four years and kind of stabilize. On top of that, these chips are only expected to last three to six years, which means once they build these out, they're going to continue to cycle through them to update these data centers.
And on top of that, as these GPUs become more complex, more, you know, compute inensive, it's estimated that every 18 months, they're going to need double to triple the memory capacity per individual GPU. So you have, you know, basically a perfect storm for demand to continue to skyrocket and pricing power to continue to skyrocket as well.
Now, we have concerns that from the consumer side, you know, all of this what they're calling chipflation. Essentially, the fact that, you know, these things are, you know, driving up the cost of consumer products so much that it's actually softening demand for computers and phones across the board.
And you know there's concerns that you know maybe that will then you know soften demand for memory chips as well but almost in a paradoxical way it's almost you know counterintuitive it's actually the opposite right because the chips in a large part going up in price is what's causing to a large extent you know this softening in demand for consumer uh devices.
It's you know it's not the other way around. It's not like you know the softening in the device sales is going to lead to lower prices in memory chips. It's actually the opposite. And this is one thing that a lot of people aren't talking about because this cyclicality of the memory chip industry has been shifting dramatically with this AI buildout.
Let's quickly talk about that. So, traditionally and historically, the far majority of revenue for, you know, companies like Micron and SKH Heinix was coming from consumer electronics. You know, companies buying the chips for phones, laptops, even things like cars. Now, that's completely flipped. Now, the the far majority of the demand is going to the AI data center buildout.
And as we spoke about, even once they're done building them out, they have to continually update these chips every, you know, 3 to 6 years. So, we expect this to be a sustained increase in demand. And the demand from consumer electronics is actually is actually a much smaller part of their overall revenue.
Now, they are diversified. So they have consumer electronics, they have you know uh vehicles are a growing sector when it comes to memory chips as basically a lot of things in in the economy are running on you know computer chips right as you know things become smarter things become more advanced we have consumer electronics we have vehicles we have the AI data center buildout but the point here is is traditionally you know as the economy went through cycles through booms through recessions you had you know demand for things like phones, computers, iPads, these things could go up and down cycllically, a lot more cyclical than things like infrastructure, right?
And people think of, you know, this d um AI data center buildout as cyclical as well. And to some extent it is, and they think, okay, well, you know, this rapid buildout cannot last forever. But essentially, like I've said a couple of times already in this video, once they build that out, they're going to continue to have to maintain that year after year.
And you know, a data center they built this year in four to 5 years, they're going to have to upgrade those chips. And then, you know, the data centers they built the following year, and so forth and so forth. So really, we do expect a sustained demand for these chips. And I want to quickly talk about that in this video.
I'm going to skip down in this presentation and outline with you guys exactly why we expect hyperscalers to sustain their massive capex ramps. And there's a few reason here. One, they have massive backlogs of unfilled demand. So hyperscalers aren't spending speculatively, they are racing to clear huge backlogs.
Microsoft for instance, I'm sorry about that guys. Let me share my screen here again. Microsoft for instance has reported 80 billion Azure backlog that is simply not fulfillable yet due to capacity restraints. So as they're building they actually have a lot of this already inked to be used right it's not like they have this massive surplus according to the contracts.
Uh it's also notable that CEOs like Zuckerberg have said the risk of underinvesting is greater than the risk of overinvesting. Right? So, a lot of CEOs have explicitly said they are willing to overspend and potentially take a hit on their cash flows in the short term to make sure they don't get left behind.
They're willing to take that risk. And guess what? You know, free cash flow going down for hyperscalers is revenue going up for chip companies like Micron and SKH Heinix. Now, front tier model scanning law model scaling laws and this is what I want to talk about guys. Each successive generation of Frontier models requires 10 to 100 times more compute power than the last generation.
Because of this, each GPU cluster needs exponentially more memory chips as well. And that's why not just with the data centers increasing, also the demand per GPU increasing as well, estimated to be every every 18 months at a double or triple like we talked about earlier in this video. We also have some massive hyperscalers already reporting that they're getting an ROI from this AI spend.
So, Google Google Cloud is up over 60% year-over-year and Azure is up over 40% year-over-year. And this AI capacity, these AI, you know, um, features and offerings that they're providing for their enterprise customers is driving more demand for their products. So, they're already seeing some ROI, guys.
Google Cloud, a massive cloud provider, growing 60% year-over-year, and they're attributing a huge percentage of that to their AI offerings. Right now, what could essentially cause a pause? And really quickly before I jump into that, I want to point out as well the fact that not only are they locked in with everything I've pointed out here, but also they have leases or purchases of real estate.
Sometimes they're leasing the real estate. They also have long-term energy contracts where they're prepaying or they're pre-signing. Hey, we are, you know, earmarking this energy because that is a massive bottleneck or it's going to be for these data centers as well. And these hyperscalers are already mapping out and signing multi-year, you know, 5, six, sevenyear deals for energy to provide this for the data centers as well.
So, if they decide, hey, you know, we're going to flip the switch off. We're going to stop building these data centers. We're going to stop buying chips, they're still on the hook to pay for all of this energy due to these contracts they've put in place. So, we have the CEO saying they're willing to take a hit on cash flow in the short term to make sure they don't get left behind.
So, they're going to continue to buy these memory chips into the future. They've signed massive multi-year deals with these memory providers, specifically Micron. And we expect SK, it's been reported in the next couple of weeks to report that they're going to be signing multi-year deers deals as well, locking in revenue at a at a clip at about 50% of what they can build.
And what I mean by that is, you know, SK and Micron, they're, you know, they're investing billions as well into building more fabs, more warehouses to or manufacturing plants to develop these chips. And over 50% of that future capacity has already been locked in through long-term contracts with hyperscalers like, you know, Google, Facebook, Amazon, etc.
Specifically with Micron, and we expect SKH Highix to announce that soon as well. Now a lot of people think well why is it only 50%. If demand is so high why isn't it you know 70 80 90%. The problem is you know and this is a risk in the short term as well you know Micron and SK have their own you know constraints as well you know building these fabs could take longer than expected you know receiving you know the um the infrastructure and the energy needed for this as well that that could get slowed down with permitting etc.
And they don't want to be on the hook to say hey we're supposed to you know give this to the customer but we're having delays in building these fabs. So 50% is a very healthy number. It gives them a cushion in case there's delays, you know, but these things are pretty much put in place right now. What could cause CAPX to pause now?
You know, these hyperscalers have the money. They've signaled that they're willing to spend the money. Uh some things that could happen is running out of power. That is possible. So one example is in Virginia, I believe it's Northern Virginia, there are permits laid out to build out these AI data centers.
and the amount of energy that would be needed to fulfill that demand is greater than what's currently existing on the grid for the entire state of Virginia, right? So, something has to be built out here and we don't know exactly how long that's going to take. That could be a bottleneck, right? That could slow things down.
These are risks we have to take into account. also just a shortage of the memory chips themselves, you know, and this is why, you know, Micron and SKH Highix, they've been getting a lot of flack because they're going to be spending billions on increasing demand. So, a lot of people think, well, if you're increasing demand, you could overbuild.
We're actually at a greater risk of the opposite, right? So, if Micron and SK don't continue to build out, they could lose market share, right? So, you know, they're in competition to build out to make sure they capture that mic that market share. If Micron doesn't build out, SK could, you know, capitalize on that and vice versa.
Uh, so that is a real risk and again, Micron having these long-term contracts really mitigates their risk to the downside when it comes to overbuilding, right? So, these are really some of the massive things in the pipeline, guys. We talked about all of this, you know, demand not just from hyperscalers building out, the fact that hyperscalers are expected to announce the continued spending in the coming quarters.
And one thing I want to point out here is, you know, a lot of people say, well, to calm these cyclical fears, why don't we hear hyperscalers say, hey, we're going to spend, you know, 300 billion in 2027, 2028, 2029, why don't they give specific numbers further than a year out? And essentially, guys, CFOs don't really like to announce things more than a year out because things could change, right?
there is a chance that you know and we have to acknowledge the risk you know software ROI for these hyperscalers could take a little longer than expected and if that does happen well first of all they can't really switch off the buildout of the data centers themselves because of everything we laid out they're pretty much locked in but they could slow down the upgrading of the older data centers right so if they have data centers that need to be upgraded in a couple of years and they're not really getting an ROI they could hold off on those and just continue to use you know the more recently upgraded data centers and then upgrade the other ones as demand picks up and as they're getting an ROI.
There's also things like bottlenecks, right? They could, you know, not have enough chips. They could not have enough energy. So, for them to really provide fixed numbers more than a year out, it's really difficult, right? So, we don't expect really to have numbers, you know, announced further out than a year.
But what we do expect is for them to announce specific numbers for the next 12 months and then to announce that they do expect year-over-year in 2027 and 2028 to spend more than they have been spending over the last couple year of years. And that is a bullish signal, right? And one thing I want to point out is they do expect analysts, you know, that are investigating this, they expect a continued growth until 2030.
And even through 2040, we still expect a 4 to 6% yearover-year growth. So currently, all of the data is pointing to continued growth. Yes. Will it flatten out? Of course it could flatten out, right? But really what we expect to see, and I'm going to show you guys on the iPad here, is we have this massive growth and then you might see a little cyclicality like this, but we continue to go up, right?
And a lot of people are thinking that we're going to have this massive crash back down to normal and then we'll come back here. That is not going to happen. That is basically impossible. But what we do expect is this massive growth like we showed here. Sorry, let me back out of there. So, we've had this massive growth.
We're basically right here currently, right? And what we expect is to actually continue to grow. Maybe not at this massive vertical line rate, but kind of like a like this, right? And then eventually in around 2030, a continued growth, right? So, that curve continues to bend but continues to go up. I'm sorry about that.
You're having trouble seeing that. But, we had this massive skyrocket. We're here. What we expect is this to bend initially but still go up and then around 2030 really bend again but again still continue to go up. We still we we expect this continued growth really until 2040 at least at least and then even then we expect it to continue to grow because at that point we expect the AI buildout or the AI data centers to really account for most of the data center spending and again they'll have to continue to upgrade these year after year.
So guys let me know your thoughts in the comments. Again, if you guys want to see what I'm buying in the million-dollar portfolio, click the link, join the Patreon there. We have that uh $15 tier. You can see what's in the portfolio, the method I'm using to analyze stocks and just build out my portfolio in general.
And if you want to get that higher $30 tier, we go deep into the method that I use to analyze stocks and build wealth, as well as a 90-day challenge we're going to be releasing in the next few months to really help you implement the method and show you how to build wealth in this market. So guys, thank you so much for watching. Again, let me know your thoughts in the comments, and I'll see you on the next
Comments 0