Chip Stocks Are On Fire — Will It Last?

At some point, will you see an air pocket? I mean, presumably you will. I mean, this is what always happens eventually. Only thing I can say is it's not now. It's certainly not this year. It really does not look like it's next year. 2028, I don't know. You know, we'll have supply starting to come online in a bigger way in '28, and then it will be a question of demand.

Again, there's no incremental capacity. Supply is tight and demand is very strong. That's not slowing down. You know, you probably don't have to worry at all until supply starts to come on in a big way, and then you'll see, you know, the tide go up. We'll see if everybody's naked or not.

You bet. And I mean, semis have been the primary beneficiary here. And as you said, the SOX index is up almost 100%. It actually may even be 100% after today's close year-to-date. And AI has just gotten so big it's dragging everything along in the space. You could have almost owned anything. You would have been just fine. It's interesting though, some of the divergences we've seen this way. So the traditional blind AI winners that you would think, the Nvidias and the Broadcoms of the world that are doing, you know, the actual AI accelerators, they've actually had the worst performance. You know, they're up, but they're up not anywhere near as much as the sector. And the reason is people have been playing the so-called bottlenecks. Again, as AI has grown, like sort of one area of the space at a time followed by the next, you've been sort of hitting the limits of what they can supply, and then prices go up. And you know, semi investors love to play bottlenecks, and so the stocks have gone up. And you know, we went from the accelerators to the memory to the semicap to the optical to the networking to the power semis to the CPUs. Now people are playing discretes and other things. It's really been kind of remarkable overall though. You could have owned almost anything in the space year-to-date, you would have been sitting pretty.

That much.

I mean, because, you know, it really has been about earnings. And frankly, even with the space up 100% year-to-date, if you sort of decompile it into the drivers, you know, probably 70% plus of that performance has actually been earnings growth. So like, multiples in the space are up, and I won't say that like the space overall is expensive, but it's not egregious. And by far and away, the earnings so far year-to-date have grown much more than the valuations have. And some of these bottlenecks, we've just seen massive revisions. I mean, take memory for example. I don't cover the memory space, by the way, it's a colleague of mine, but to talk about the industry rather than the stocks, I mean, we've just seen some phenomenal positive revisions in the earnings powers. You know, memory prices have just gone through the roof as supply has gotten really tight as AI has grown, and it's just driving massive revisions in the earnings. And we're seeing that among a number of the bottlenecks. I think for some of the other ones, there's the hope or the strong belief that numbers as strong as they are right now are just too low given where AI demand is growing. And again, you can look at semicap or maybe some of the optical names or, you know, some of the other ones that are playing out right now. You can sort of look out at where people are forecasting demand to be, and you can see where the numbers are sitting right now. And it's clear that one of those things is wrong. Like either the demand is not going to be there, or numbers broadly probably still need to come up, and in many cases across the board. And that's why you've seen some of these other names respond. And so I don't know that it's necessarily unjustified. Again, I wonder why some of the other compute names haven't performed as well because again, numbers there I think are going up too, but I think maybe they're more well-known or more heavily anticipated that you will see that kind of performance. And again, there's only so much money to go around, right? Investors have to invest in something. And I think there's been a belief among the more traditional guys that, you know, they're safer. You can use them as a source of funds in some sense to play some of these more esoteric names because, you know, like numbers at least for the big guys are going to go up anyways, that there's less risk from doing that. And so we've just seen some of the funds shift into other areas, I think.

You could look at it probably there's some elements of truth from all of those things. Look, semis I think are cyclical and they always have been and they probably always will be, but there's a variety of different types of cycles and different durations. And like you could argue that this cycle has duration. And let's take the memory space for example, and that's probably one of the areas where we've seen the biggest price increases as well, which is driving all this. But the reason is demand, as you said, demand is very strong and supply is very tight. And you know, this time is different is always sort of a dangerous statement, but there are some things that are different. Take the DRAM space for example. Like DRAM is like in your PC. It's like the system memory that's running things as you're working. And AI uses a specific type of DRAM memory. They call it HBM, or it stands for high bandwidth memory. And it sits on the AI chips that get sold. But to make say a gigabyte of high bandwidth memory takes four times as much capacity versus making a gigabyte of like the standard what would be called DDR5 DRAM that would go into your PC or smartphone. So you're in a scenario where you could be adding, you know, wafer capacity to make DRAM for AI, and actually even though you're adding a lot of wafers, you're not necessarily adding as many bits as you ordinarily would because of this differential in how many wafers. I should call it a trade ratio. Right now it's like 4:1 between AI memory and traditional memory for DRAM. And so because of those dynamics, I mean the other thing I should say is it takes time to add supply. And one of the issues right now is we're short on what's called clean rooms. So to add supply, like there's a whole industry, it's called semiconductor capital equipment. They make the tools that make semiconductors. But before I can sell those tools and add capacity, I need to have somewhere to put them. You need what are called clean rooms. These are the buildings, right? That the factories that make up the factories. And we're short clean rooms right now. So they have to build the clean rooms first, and then you can add the tools and add the capacity, and it just takes time. I mean, like this year for total semiconductor manufacturing equipment will probably do $145 billion, which is up 20% year-over-year. So it's a strong year, but as strong as it is, it's a constrained year. The clean rooms don't come online until, you know, well into next year and beyond. In the meantime, demand is still growing, and so it's entirely plausible that this cycle, this upcycle, could last quite a while.

So far again, I would say as long as AI demand continues, and I would say if AI demand does not continue, we're probably screwed. So we'll put that aside. All the signs right now seem to be pointing to continued strong demand, at least for now.

I'll say the same thing I've said since this started. At some point, will you see an air pocket? I mean, presumably you will. I mean, this is what always happens eventually. Only thing I can say is it's not now. It's certainly not this year. It really does not look like it's next year. 2028, I don't know. You know, we'll have supply starting to come online in a bigger way in '28, and then it will be a question of demand.

Again, there's no incremental capacity. Supply is tight and demand is very strong. That's not slowing down. You probably don't have to worry at all until supply starts to come on in a big way, and then you'll see, you know, the tide go up. We'll see if everybody's naked or not, right? This actually happened, you know, it wasn't that long ago, a year, two years ago, when the actual accelerators, the GPUs were in very tight supply. And by the way, I would say semi investors always worry because there's a phenomenon that happens in many cases when supply is tight and customers can't get what they want on the time frame that they want, is that they order more. It's called double ordering, right? And so the question is always like when supply is tight and demand is strong and you're adding capacity, is that demand you're adding for real or is it phantom? This happened during COVID, you know, we had big shortages and lead times stretched out, and you know, they added a bunch of supply, and as it turns out, like it wasn't needed, and it took three or four years to actually work off that oversupply. On the other hand, you go back a year or two when the GPUs and the accelerators were in tight supply, and that supply came online in a bigger way, and actually demand, as the supply came, demand got stronger rather than weaker, that demand was real. So that will be the question. We probably won't know the answer for a couple of years. In the meantime, for the next, you know, certainly for this year, like I said, and almost certainly for next year, supply is going to remain tight, and in that environment, you know, people will act like the demand is going to be sustainable. So far though, everything we've seen just points to everybody's short compute. That seems to be the case. And again, it's not like people are reserving compute and just sitting on it, like they're reserving the compute and using it. The compute is getting used. We can have a discussion on what's the ROI and return on that. I think that's a valid discussion that we can have. But the compute is getting used, and in fact they want far more, they want to be able to utilize far more compute right now than is currently available.