Dan Niles: People buying Micron and SK Hynix are Missing THIS major thing coming

Yeah, they just seem to keep coming, don't they? So it'll be interesting to see how this goes ahead and plays out given the interesting times we're in.

Well, I mean, I think it speaks to the fact that you have multiple things going on. Obviously, memory — or at least obviously to me, memory is a bottleneck. I think most people agree with that. The question is how long does it last? And I think SK Hynix coming into the US market just speaks to the amount of capital needed to build out the infrastructure to support AI. So I always say it's never the data, it's how people react to the data, you know. So do you look at the fact that SK Hynix is raising high-20s billions of dollars as a good thing because it shows that, oh my god, the demand is still so strong they need to raise this money? Or do you view this as a bad thing because, oh my god, they're raising this money to expand? So I think for right now, you know, the stuff that's going on with memory — and obviously you've seen violent swings in the Korean stock market and the semiconductor index over the last month or so, you know, every day. And I think the Korean stock market has had what, three trading halts, I think. Um,

Recently. So, you know, if you're asking longerish term, I just look at a very big picture and say you've had agentic AI come out on January 30th. That requires, you know — and I'm using obviously the Open Claw formalization on January 30th as sort of the starting point — and then that requires 10 to 100 times more tokens for those types of requests versus just chat-based AI. You've also got the Vera Rubin platform coming out that requires at least three and a half times more memory in the minimum configuration of that relative to the Blackwell generation. And so from a longer term perspective, I see memory still being pretty strong at least for another year as we go through this AI buildout. So, you know, it's a complicated answer, but that's kind of how I'm thinking through this right now. And in general, I'm bullish on memory over the longer term. In the short term, I think there's going to be a lot more — you know, in terms of stock prices being driven by what is mattering more: the token minimization, as you see companies trying to control their AI spending bills, or kind of the architectural shift underneath?

Well, what I'm trying to point out is that I see a lot of arguments that people are making these days of, 'Oh, you know, don't worry about China at all because they're trailing edge and it really doesn't make a difference.' And what I tried to show with that chart is — and I, you know, I started in the industry, I'm an electrical engineer by training, and you know, worked at Digital Equipment Corporation in the 1980s — and if you go back to that period of time, you know, Japan, when they started in semiconductors, they were also trailing edge, and you can see that the US had 90% market share. Most of that was Intel, for those viewers who don't know the history of the market. They had almost all of that share, and then Japan started to ramp up, and Intel almost filed for bankruptcy in the mid-1980s, and they shut down producing memory and gone into this business called microprocessors. And then I was, you know, on Wall Street in the '90s, and I remember looking and seeing how the Koreans were getting into the market, and they were also trailing edge. But you can see by what's happened with the share, they went ahead and then they put the Japanese pretty much out of business. Now what you're seeing is China ramping up, and people are like, 'Oh, don't worry about the Chinese. They're several generations behind.' But you have to remember China, from an end consumption basis, they buy 20 to 30% of all PCs made, all smartphones made, etc. And so it'd be crazy for people not to think — especially when you know that Apple is out there trying to qualify CXMT, and by the way, they're going public next week on Thursday and they'll be raising close to $4 billion or so — it's crazy not to think that they're going to keep ramping that up, especially with the support of the Chinese government, because when you're absorbing that much of the end market for electronics, you're probably going to want to get self-sufficient on that. And so longer term, they're going to be a problem. And especially because obviously China is a massive country; it's going to be state-supported, and that's going to cause an issue. And so that's more of a — the way I think about it is that affects your terminal multiple and how you should be thinking about it. And I just wanted to bring that up because I see a lot of, you know, people talking about, 'Oh, it's different this time,' and I'm like, 'It's not different this time. It's the exact same thing.' It's obviously not going to matter near term, but from a longer term perspective, China is definitely going to matter.

Sure. So yeah, I mean, it trades at a discount to Micron, and so you can say, okay, it's trading at, you know, 20% to 30% cheaper depending on how you want to look at it. You know, the way I think about it is they are the leader in high bandwidth memory, right? Samsung obviously has number one market share globally for memory, but SK Hynix really got HBM up and running earlier, and then Samsung sort of was playing catch-up to SK Hynix and Micron. So to me, the discount makes it actually on the margin probably more attractive than Micron is to some degree. They're all the same trade. Now, Samsung obviously has other things in there like cell phone division, etc., which kind of complicates this. And so when you go kind of pure play, it's really SK Hynix and Micron. And so I can see that valuation gap closing over time. Now, whether that's because Micron comes down or SK Hynix goes up in the near term, we'll have to see how that plays out. But from a stock price perspective, if you're a US investor, you really only had one choice up until today — or sorry, two if you want to throw in SanDisk, obviously. So you had two choices. Now you have a third, and they're a really good company. And so I can certainly see money coming out of SanDisk and Micron and people putting some of that into SK Hynix. So you have three names in your portfolio as opposed to just two. For me, if a space is doing well, I typically try to say, okay, I don't want as much company-specific risk if I have a big picture theme, which in this case is memory is going to continue to be important. I'd rather have diversified exposure to the space unless there's some really super compelling reason not to.

I think it's next Thursday, CXMT goes public, and that, if nothing else, will get people thinking, 'Huh, China's in DRAM and there's something public and they're raising money.' Because I don't think — because CXMT is private right now, right? And YMTC, which is, you know, the NAND memory producer, they're also private. But CXMT coming public may actually have people who have longer memories going, 'Hmm, I remember the Japanese in the 1980s. I remember the Koreans in the 2000s. You know, what is the right terminal multiple?' Because that's what everybody's debating. And so I think that's going to be the very — I'm highly curious to see how the markets take that. Obviously, they're listing in China, but them being public and you being able to see what their stock price is doing, I think will be very interesting, because I think that's the long-term threat to all the players, whether it's in Korea, Japan, or here.

Yeah, I mean, there's only one other major one, right? It's Yangtze Memory in China. And so I think, yeah, given the amount of capital needed — I mean, this is a very capital-intensive business, right? And so while the money's available, you should go and get it, because as you've seen, you've now got the biggest companies in the world, whether it's Amazon or Google with that $85 billion offering. Amazon just did a bond offering just a couple of days ago. You know, the AI buildout is so expensive, and the money right now seems infinite, right? Like there's enough money to fund everything, but money is not infinite. And so at a certain point, it's going to get more problematic the longer you wait to raise the capital, because the capital's already been allocated. So it's going to be a fascinating year as this plays out.

Well, here's the interesting part, right? So if you go back to — so you have Open Claw come out on January 30th. In March, you have people at Meta talking about their leaderboards for token maxing, and everybody is cranking up and focused on, 'Hey, how do we get our employees to use more tokens?' Then you have Uber come out in April and say, 'Hey, we blew through our entire AI budget in four months.' And then you fast forward to June, and you have the Coinbase CEO come out and put a post out on June 26th saying, 'Yeah, we've gone ahead and done five different things, but one of them is we now route queries to the cheapest models where we can.' And we've managed to cut — even though we're generating more tokens, we've cut our revenues — or the AI bill spend, which ultimately turns into revenues for the next set of companies — we've cut that AI bill by almost 50%.

So then you go, 'Oh my god.' Now, for those who don't kind of understand, if you go ahead and ask Anthropic what 3 plus 3 is, that's stupid, right? But go back, you know, six months ago, that's probably what people were doing, right? You can route that query to a very, very, very simple model and pay a tenth of the price of sending it to Anthropic. And so what people are doing now is saying, all right, for the really simple queries like 'summarize my emails,' you don't need Anthropic for that; you can send that to a really cheap open source model. Now if you're trying to, you know, figure out some nuclear physicist stuff, then yeah, okay, then I might want to use the Anthropic model for that. And so that's a quick way — very simple example, but a quick way to get your AI bills down. So that's what I'm really watching to see what happens when these companies guide for the September quarter. The June quarters are going to be just astronomically good. The question is what happens for guidance for September, and now the interesting development over just the last couple of days is obviously Meta coming out and saying, hey, we're going to sell Metamuse Spark at one-fourth the cost of the Anthropic models or the ChatGPT models, and then Grok also came out and had a big step up based on their work with Cursor in terms of their model performance. So yeah, now you've got even more low-cost alternatives. And so that revenue generation — it's going to get very interesting what that looks like for the next 3 to six months. And that's why back in June, I posted about the fact that I was worried about a speed bump coming, because you're switching from 'let me spend all the money I can on tokens' to 'oh my god, I'm blowing through my entire budget in four months; I need to cut my spend back.' And so what does that do to the next three months in terms of the revenues being generated from AI as people go through that token minimization process?

Interesting. Yeah, I mean, it's going to be fascinating just to see the interplay between that, Micron, SanDisk as the day wears on, and, you know, how is the money moving and flowing, because obviously they just raised, I think it was like 28 billion or so — I don't know what the final number ended up being — but somewhere. And if you're a portfolio manager, what you probably did was say, okay, you know, I own SanDisk and Micron, let me sell a little bit of that and put it in SK Hynix. But, you know, there's a lot of other stuff going on. Obviously Meta today is having a huge move based on what they've talked about with regards to their presence in AI, and then, you know, what does that mean for capex? And so this is going to be really super interesting between now and then, and also next week when you have CXMT come public in China.