---
title: 'Is the Bubble Over? Some Proof...'
source: 'https://youtube.com/watch?v=R6CTLeHIJBY'
video_id: 'R6CTLeHIJBY'
date: 2026-10-31
duration_sec: 1562
channel: 'Meet Kevin'
---

# Is the Bubble Over? Some Proof...

> Source: [Is the Bubble Over? Some Proof...](https://youtube.com/watch?v=R6CTLeHIJBY)

## Summary

The video analyzes SK Hynix's earnings call, highlighting their bullish outlook on AI-driven demand for memory chips. It discusses long-term agreements (LTAs) that cover over 50% of capacity for 2027-2028, custom chip trends, and the company's strategy to mitigate the boom-bust cycle. The presenter also evaluates SK Hynix's valuation, noting a low P/E ratio that suggests potential undervaluation.

### Key Points

- **Earnings Call Overview** [00:16] — SK Hynix's earnings call is described as 'wild' and one of the most bullish for memory plays.
- **Long-Term Agreements** [01:54] — LTAs are used to avoid the historical boom-bust cycle by securing long-term demand, with over 50% of 2027-2028 capacity already committed.
- **LTA Details** [02:40] — Customers are requesting longer-term collaborations, and SK Hynix has LTAs with 14 customers including major cloud providers.
- **Custom Chips** [03:06] — There is a trend towards custom memory chips, similar to ASICs, which could improve efficiency but also bears risks if needs change.
- **Capex and Supply** [05:28] — SK Hynix plans to increase capex significantly through 2027, but argues that more investment doesn't necessarily mean more supply due to advanced packaging requiring more wafers.
- **Free Cash Flow** [08:48] — The company's free cash flow is exploding despite rising capex, showing strong operational performance.
- **AI Demand Evidence** [10:39] — AI investments are leading to greater adoption, with hyperscalers reporting 30-80% YoY growth in cloud revenue from AI.
- **Sold Out Through 2029** [18:24] — SK Hynix is sold out through 2029 for DRAM and HBM, with customer volume requirements more than double current supply.
- **Valuation** [24:10] — The company trades at a 6 P/E ratio and 0.3 PEG, indicating the market is pricing in cyclical risk, but the outlook is positive.

## Transcript

results boosted hopes that profits can be sustained as AI [music] developers build data centers which drives demand for Marraming. Joining us, Bloomberg >> the SKH Highix earnings were crazy. Let's go through a breakdown of not just
the raw numbers. Let's go past that. Let's go into what they're saying about the entire AI buildout in the earnings call and do a quick summary of all of because I think this is absolutely wild.
know is they do these things called LTAs. They're long-term agreements. And the whole reason they do this is because back in 2023 and many times in many cycles before that, I think 2014 was another one. Uh the.com bubble was
another one. There these cycles where memory basically goes through a memory basically goes through a shortage. prices of memory go skyhigh and then all the memory manufacturers are like you know what we should do we
should build more and then all of a sudden the underlying desire to you know buy more memory chips fades or slows or the growth rate slows now all of a sudden you have a glut and you have overp production over capacity prices
fall off a cliff that's what everybody's worried about with the memory stocks and honestly I've been a little bit of a poopy dupy bear since this summer on at this point. I mean, it's kind of going up like GameStop at 400 bucks and
and you know, it's like Leopold came in and turned off the buy button. You know what I'm saying? Like the momentum just evaporated. And so, uh, we we we were not very excited about these plays going even into the LEO fold because we saw
this as sort of a short-term trade and you better have a trailing stop. Many times called out, have a trailing stop. Now, what do we hear now? Because what we're hearing now is actually really optimistic for these memory plays. What
these 5-year contracts with their customers. And right now over 50% of all of SKH Heinik's capacity, the production that they can generate for 2027 and 2028 is going to long-term agreements. Now,
That's great because it gives them some stability. But what happens when those you could be left holding the bag, right? But that's actually where things are getting fascinating here. Look at this. Customers are requesting
longerterm collaborations based on our product roadmap and supply capabilities, indicating that long-term agreements are evolving to strengthen customer relationships and business continuity. To date, we have long-term agreements
with 14 customers, including pretty much all of the major cloud service providers, and we're proactively in discussions about extending those want to extend contract periods? Doesn't that create more risk for a company like
Google or Microsoft or Amazon or Anthropic or whatever or Nvidia or AMD? Sure. But take a look at what they're also doing. They are building now custom
memory chips. So we already know that Marll and Broadcom work with companies like Anthropic and OpenAI to make custom application specific integrated circuits. Those are called AS6. You remember when those were all the rage
for mining Bitcoin? They do one thing really well. That stands in contrast with an Nvidia GPO GPU which is like a workhorse. What do you want me to do today, boy? I got you. You want me to do that? I got you. Okay. The A6 like man,
I push this button all day long, bro. obviously push the button and so if needs change those A6 can lose value very rapidly. It's one of the reasons people will pay a premium for an Nvidia GPU or some of the AMD products. Custom
interesting because this is now suggesting hey how about these memory companies come in and actually customize what individual customers want for their
product. Now, what we're seeing a lot of is spec changes at companies like Nvidia where we're saying, "Hey, do we really want so much memory or should we despec necessarily need all the memory." What you don't want is stranded memory. You
don't want to oversize a chip with so much memory and you can't even operate all of what you're trying to run through a chip because the bandwidth is your one flow, what do you need all the RAM
for? What do you need all the high bandwidth memory for? So that's where longer term deals if you want. We'll even customize the chips for you. These even customize the chips for you. These things are very smart. And the bull case
in investing in these companies is a they've gotten cheap, b is they're know, capital to shareholders. We just saw at the beginning here 45 billion soon and they promise to return even more in the future. But what's really
fascinating is that these companies are trying to kill the boom and bust cycle for memory. And that's where reading this is either really good if you're a bull or if you're interested in what's going on with the overall artificial
intelligence space and frankly what's happening with the AI buildout. This is a key read. So let's go through some of the highlights for you. So we expect to increase capital expenditures uh throughout 2027 significantly compared
throughout 2027 significantly compared to 2026 to expand the share of leading edge processes and ramp up our first fab in Jungen. Okay. Generally when investors see wait you guys are going to increase capacity even more you're going
to you know spend more money on capex. When people see that, they get nervous. They get nervous that, oh, you're definitely going to lead to an overupp. So, they actually got asked specifically about over supply. And uh what they said
was this. They said, "Thank you for your question. It is true that company is investments to accommodate strong demand, but more investment and more fabs does not necessarily mean more supply. with a rising share of high
memory which keep in mind that takes about two to five times as many wafers as DRAM. Okay, just to be clear here like RAM is like your overall umbrella. everybody wants right now. So just
umbrella product and then you could divide that into like DDR5 uh which could be like a server product or you could stack DRAM and turn it into high bandwidth memory. The point is, the more you're using these advanced uh chip
packages, the less or the fewer chips you're out potentially better quality or more capable chips. And so that's what just because we might be able to produce more and more different more facilities,
we might be creating more complex chips, outputting them with finer technology. uh and therefore that actually restrains our potential total bid output. Now that's obviously their bullish point of view. It's their company trying to pitch
their own products and their own stock. Obviously I generally see earnings calls as a sales pitch uh for their own stock. But this is a really good comparison or argument for the bulls. It's basically saying, "Hey guys, like just because
we're going to manufacture more doesn't doesn't mean you know we're going to over supply." Obviously, at some point, if the AI trade slows down, you know, you're probably going to be in a recession. Uh, but, you know, until that
point, so far, the AI boom seems like it is crushing it. Now, I want to show you because some of these numbers and a chart really, really interesting. A very brief reminder, though. Reinvest one, the super intelligence and one terminal
coming out in this Q4. Uh, does have a coupon code expiring on Friday, October 30th. Join before the price goes up and make sure you lock in your lifetime access to the alpha membership, the portfolio transparency trade alerts,
reinvest stocks terminal, the homes terminal, the reinvest research terminal, all nine courses, the course member live streams, the agentic AI modules. You can learn all about it over at reinvest.com. We'd love to see you
there. Get that lifetime access before Friday. We have a price increase. Okay, let's take a look at this chart because this is fascinating. So this is the cash this is fascinating. So this is the cash machine of SKH Highix. So basically
their capex is going up. That's the little purple box here. Okay. The cap their capex how much they're spending and investing going up bigly. I mean you're seeing a massive increase in capex from last year. However, their
capex from last year. However, their free cash flow is also exploding. All the full bar chart here is their operating cash. The green is their free cash flow. So even after their capex, they still have massive free cash flow.
we want to give a lot of money back via dividends or whatever to shareholders." Now, obviously, that's a sales pitch to get people to buy the stock. Duh. That's wrong with that. It's just always worth taking things with a grain of salt. So,
here they talk about returning 50% of free cash flow to shareholders, but also ramping capex. The chart I just showed helps a lot. So, what's going on with AI demand? Because there's been talk about pacing the AI buildout. Is this talk
about pacing the AI buildout going to, you know, crush the economy? Where do we down AI investments is the only thing holding up GDP. In fairness, should do a quick check on GDP. GDP has been doing
really freaking well. Okay. Uh a lot of that obviously driven by these you know expenditures and these investments that we're seeing by some of the richest companies in the world. Atlanta Fed real GDP estimated to be at 3.3% right now.
Uh and that's actually come down from as high as 6% earlier in this quarter. Those are the estimates for the third quarter. They come in very noisy at the kind of settle out a little bit towards the end of the quarter because they
collect more data. Uh so it's kind of like a real time response to what regarding the pacing of AI, there are two main reasons why they think artificial intelligence momentum will continue. I I think this is so
fascinating that you could literally open up an earnings call and see exactly what the industry insiders are telling you is happening. Take a listen to this. First, we are seeing customers AI investments actually leading to greater
investments actually leading to greater adoption of AI services and they are producing business results with major hypers scale customers reporting year-on-year growth of 30 to 80% of their cloud revenue. So growth of 38 30
to 80% on cloud revenue a substantial portion of which SKH believes is because of the expansion of their AI business and their AI products. You know what I like to do sometimes is I'll sort of just see and and Google has been so left
if I can pull it up right now, but sometimes I'll just go into um Google admin and I like to go see like, hey, what are what are they charging for, you know, upgrades or like what, you know, what what can you add to to users or
seats or whatever. Uh, and so what you could do is you could actually kind of go into the back end and see how they're trying to upsell people licenses, right? So if I go jump in right now, yeah, AI expanded access here. I don't think that
any of this is pretty proprietary here. I'll just make sense. This is what I like to look at. I like to keep an eye on this. So I look at this, for example, uh, and I'll see, okay, so we don't have any more Google workspace enterprise
licenses. That's fine. But then I look at this AI expanded access. I could go add to this email address AI expanded access for 24 bucks a month. And then what they'll do with this is they'll pitch you, oh hey, you know, now you
could use a better version of the chatbot on uh you know, Google chat or chatbot on uh you know, Google chat or Google Gmail or whatever or we'll let you have more usage blah blah blah whatever. uh and they're really trying
to integrate AI throughout a bunch of different components of the business. Like here's here's how Google does it for example. They'll come in and say expanded access will get you uh uh you know more queries per month. Build and
them. Build more canvases. Generate more videos, more avatars, uh whatever. You could get more thinking responses. You could get uh you know more argon responses. Look at this. up to 200 of their Argon. Now, that's Gemini 4. So,
don't get Argon. It's kind of like their way of saying, "Hey, you want to pay for Fable? Oh, you're going to have to add it to to, you know, your workspace seats." And it's so easy to check the box and add it. And when an industry
does it per seat, that's where companies like Google print freaking money is know, we're going to enable this for everyone. It's like GitHub, you know? that security feature cuz everybody's worried about cyber security." Great.
That's going to cost you 10 bucks more per month per seat. You know, you got thousands of people. It gets really expensive really fast. So anyway, like I happening. Have you or a loved one child
been affected by social media? Morgan Morgan can fight to help you. If you or your child spent years growing up on social media and experienced serious mental health struggles, you should hear this. Recently, a jury ordered Meta and
YouTube to pay millions for their role in designing and promoting addictive platforms. They knew their apps could contribute to anxiety, depression, and body image issues, but did not share those risks while prioritizing profits.
The recent historic verdict underscores the consequences of those decisions. Now Morgan and Morgan is stepping up to hold these platforms accountable for the harm they caused. Click the link in my bio to take a short quiz and you may be
entitled to a potential recovery of over $1,000. It's quite easy to do. Go to the link in the description, fill out the form for yourself or a loved one. It's quick, easy, and you could recover a potential
easy, and you could recover a potential recovery of over $1,000. enterprise AI solutions as well. While other cloud order backlogs have grown
two to fivefold yearon year, cloud order backlogs growing two to fivefold year on year is incredible. Cash flow will only improve they say across AI like this is
improve they say across AI like this is so damn bullish AI it is insane. It is it I almost want to be bearish because of how bullish it is because I'm generally a contrarian. I'm the kind of guy that's gonna be at a party, okay?
guy that's gonna be at a party, okay? We'll be at a party with a diet coke. Hey, you want to buy real estate? And then 20 people be like, "Real estate? F that. We hate real estate. Fine. I'm going to buy more real estate." You
the same thing next time there's a stock market crash. You know, everybody's going to be like, "Ew, stocks suck." I'm going to be like, "Hey guys, we should buy stocks. Ew, leave." Like, I don't know. I just always seem to want to do
the opposite. Uh and so I feel bad almost that I am aligned and in agreement with this bullishness because that is not part of jaded Kevin. But this this makes sense. This is what we're seeing. And I believe there has
been a massive inflection in the last 6 months and how good AI is. And I think people don't even notice it yet. They how good it's gotten. It's scary. AI and will ease the current market concerns on the investment burden. In
multi-sight development, smaller or midsize data centers become uh to bring power and data centers available more rapidly. This is basically saying, look, Olirri facility that everybody's going to go protest outside of, what if we
have fewer protests and if two of them get shut down by city permitting issues, actually do the same thing at reinvest applications in hopes of getting three of them because, you know, California is
when it comes to building. So, you just you just go in knowing that, you know, uh and that's okay. You know, you underwrite. You always underwrite as if you are not going to be able to build. And as long as your deal makes
sense and pencils out, as long as assuming you're not going to get what you want from the city, if you get what you want from the city, it's pure bonus. you want from the city, it's pure bonus. I love that. But anyway, so uh a single
company adjusting its investment in the past could directly result in overall demand volatility. However, today because AI is rapidly broadening to sovereign AI, on-prem demand, non- tech companies investing in their own AI, all
of a sudden you no longer have this single company uh uh problem. So in other words, in the past and and I've changed I changed this yesterday. Okay, remember how I always say there are these two canaries. Okay, canary one is
the labor market. Unchanged. Labor market's a big problem. But the second canary I've historically said is anthropic and open AI. But yesterday what I did is I actually changed that. I said no, I think that's wrong now. I
don't think they matter as much as they used to because you have competition Amazon, from SpaceX, from Grock, you know, whatever. And what's actually happening is you are seeing more diverse AI expenditures that insulate
AI expenditures that insulate uh uh the entire AI boom uh from volatility from one of the companies slowing down or one of the companies having problems with financing or whatever. Like it's not just Oracle,
it's everyone, right? It's no longer just one company that moves the whole market. So I actually agree with that. Okay. Then they mention here that they're sold out through the rest of the decade. That DRAM and high bandwidth
volume for the next three years through 2029 is effectively sold out. That their goal is to improve longer term demand visibility that could lead to stable revenue and earnings growth. Yep, that's the whole point of these long-term uh
agreements. We talked about the over production fee. We talked about the custom chips. Uh they do mention that prices for HBM4 are expected to increase and that long-term high bandwidth memory demand outlook remains unchanged.
Memory or demand remains so strong that customers volume requirements are more than double what we can supply today. That's freaking insane. Uh then over here, this is where they talk about nextgen chips. This is where things get
a little bit geeky. So, I'm not going to go really crazy on this, but basically, right now, they use a technology called 6F. Okay, this has to do with uh the the
measurement of the chip technology that they do. And the smaller they go, the more they could put memory cells on chips and potentially uh the the more the problem with that is they say there's a risk to going to 4F right now
could have reliability issues, you could have competitiveness issues like cost competitiveness issues. So what we're going to do instead is we're going to focus on what we do really well, which is smart for a business. We're going to
focus on the 6F structure. We're going to maximize our potential income from the 6F structure. That's what we're going to focus on. And we'll scale when kick the can down the road a little bit on that. Keep in mind that this company
on that. Keep in mind that this company is both a DRAM play and a NAND play, personally have this thesis that DRAM is going to explode in supply, which, you that might be true, but we might net have fewer products out because we're
adding more DRAM to high bandwidth memory or whatever. Uh, like we're increasing stack sizes potentially or whatever. But they're also a NAND play. 21% of their income comes from NAND. Now, this gets
confusing to people. So NAND, the way to think about it is NAND is just this is all storage. Okay? When you hear NAND, when you hear SSD, when you hear NVME, just think storage. Okay? Storage is different from RAM. DRAM, high bandwidth
memory are under the umbrella. DDR5, right? Those are all under that the right? Those are all under that the memory umbrella. NAND, SSD, NVME, storage. NAND is the actual memory cell
the finished product. That's what you buy. The NVME is what plugs in through your PCI port and the communication right between those blah blah blah. The point is this company actually does both. I am bullish on NAND because I
talked about this even before I read this. I'm like dude I'm finding that these computers where I run agents on if they don't have a good SSD they get bogged down. They shut down. they basically blow up if you're trying
Because I think what's happening with uh agents today is they're offloading uh more of this memory to your SSD and not really telling you about it. It's happening in the back end and it's crushing some of the weaker computers
and uh it's actually making the case for like, oh yeah, I'm just going to go buy Apple Neos and run chat bots on them. you need better computers, you know,
like for example, I got an M4 Max right here uh laptop and I could run agents on this guy all day long, but those other guys is just going to start smoking. But talk about how this is going to happen more and more where uh basically you're
more and more where uh basically you're going to see nan demand increase as AI inference demand increases. So, you could really store uh some of that cache off of memory because you can't keep it in memory
your memory and then you're going to be memory locked. So, this I have to say this was a lot and we're going to do a valuation on them in just a moment. This was probably one of the most bullish earnings calls that I've ever read.
earnings calls, okay? I put myself to sleep reading Ernie's calls. Sometimes call. [laughter] Um but uh but yeah, this this was incredible. So SKH 10 out of 10 for giving us insight into this boom is
nowhere near done. Now what I do want to do is grab the SKH valuation. Now these are ADRs, so I have to be a little careful with uh how we do these. Uh but we're going to pull the valuation here. So,
uh, let's see. A skhy.
we're looking at is their forecast earnings per share and then their expected growth rate. So right now earnings per share are expected to come in and let's go ahead and change this all to US dollars. Remember this is a
foreign uh company. So everything's going to be in Korean Juan. So we're going to be in Korean Juan. So we're looking for EPS adjusted of $270.
ratio is 10:1. So you need to divide EPS by 10. This is where it gets complicated. You need to fact check me on this, but you need to divide EPS projections by 10. So that means one to get to an ADR 163 divided
by uh sorry 27 yeah divided by 27 basically that's the stock price right now divided by 27.1 we'll call it six. This is trading for a six PE ratio right now. This is insane
going to roll over and fall off a cliff. And honestly, the analysts are pricing that in too. There's a big negative year in 2029 that they're forecasting, but analyst expectations are just looney tunes right now. But anyway, 2027, we
expect 35% growth, 22% growth, minus 18% growth, and then 28 or sorry, 38% growth. Let me do that math one more time to make sure I got it right. 3522 time to make sure I got it right. 3522 - 18 + 38.95. Yep. Divided by 4.
- 18 + 38.95. Yep. Divided by 4. Expecting 19% growth on average for EPS. better than that. They're trading for a 6 PE ratio divided by 19. Dude, this is 6 PE ratio divided by 19. Dude, this is trading for a.3 peg. This has been so
discounted for cyclicality. It's insane. Now, I want to be clear. I don't own any memory or storage plays, but I have to say your boy is salivating. The only
that is you checking out reinvest one.com. We did have a great options trade uh that we closed out on a sold put today, which that was up about 40%. that. I'd love to see you there. And uh you can uh once once you join, you get
video, check this one out. I think you're going to love it. told us here? I feel like nobody else knows about this. >> Congratulations, man. You have done so
you. >> Kevin Pra there, financial analyst and YouTuber. Meet Kevin. Always great to get your take.