Ray Wang: Why does I need a lot of memory at all? So you just explain, you know, it's important for inference. It's important for training, whatever. But like, why? Let's say you want to train a model, right? And then you need, you just need the pure tons of data. What you need to do is right. Tons of data there to train your model. And then for the inferencing part, like you need one data after another to kind of compute one after another, sort of a chunk of sort. Right. So to to do that process, how can you keep the previous data you process to keep in there and the process to another data that coming out after it? Right. So that's why you need a lot of memory,

Ray Wang: you know back to ChatGPT starts coming up. People just like hey I wasn't whether I like a very simple questions people right now is sort nastier. People are doing like hey can you write of, plenty more. How tensor can happen on people arrival example. They will give like 20 page report and they will wait like five minutes. Right. And so if you think about a process like they do all the calculation, all the researchers, and also give you the output tokens that, that's not, and so you get from your prompt less significant way longer than the prompts that you are given. Right. And that's also the longer response that you are getting when you go back to the GPT-3. And remember, usage, we know back to GPT-3 was the workload, you know, the monthly daily users. But right now it was I think the use of GPT was 800 million. Right. And we are not included. The users for the Gemini for called for AI. Right. So if you multiply loss, in fact, I think the memory the management inefficient and it's very, clear to me.

Joe Weisenthal: Hello and welcome to another episode of the Odd Lots podcast. I'm Joe Weisenthal and I'm Tracy Alloway.

Tracy Alloway: Tracy, you know what I'm worried about? I don't know if I'm worried about it, but it kind of feels like the direction things are going, which is that like all energy, industrial commodities, anything that we use for any purpose is just going to like feed the AI beast and us humans. We're just going to get left out in the cold. Okay? Nothing for you. We got to like we got to feed it all to the AI and and maybe in ten or 20 or 50 years, the AI is so powerful that this is AI. Why? Why do humans get any of this? We just, we should keep it all to ourselves.

Ray Wang: A legitimate concern, I would say. I mean, to some extent, we are already seeing this crowding out effect. Right? So energy prices in certain areas have been going up and most recently memory chip prices. So this has been all over earnings. Recently. You had companies like Apple saying that because of a crunch in memory chip supply, they might have to either raise their prices or like cut down on the amount of phones they use. Nintendo. Yeah. Like if you pull up a chart of Nintendo shares right now, they are just getting hammered and supposedly it is all because of this memory chip shortage.

Joe Weisenthal: By the way, do you remember, do you remember like buying your first PC?

Tracy Alloway: Yeah. In like you bought one before I did, I think. But mine was in the mid 90s.

Joe Weisenthal: Yeah, yeah. I think it's about when I go to remember how much memory those things had.

Tracy Alloway: Oh like nothing. Yeah. Like I looked it up or something like less than ten megabytes for a bunch of them. You know how much PCs have nowadays?

Joe Weisenthal: Actually, I have no idea.

Tracy Alloway: I think it's let's see, 16GB. Yeah. Is that crazy?

Joe Weisenthal: It is. It is really crazy. I remember, like, do you remember zip drives.

Tracy Alloway: Yeah. Oh my gosh. And or like and then there were all these special, you know obviously various memory peripherals that you could buy. And they came in sort of dedicated cartridges and stuff like that. I think they were called zip drive.

Joe Weisenthal: Yeah. I think right there. ZIP drive, I don't remember. Yeah, yeah.

Tracy Alloway: It was yeah. That's why additional memory storage etc.. It used to be this big part of the consumer process. I'm actually glad. You know, my son really wants a Nintendo, but if they can't produce them anymore because they can't get the memory, then I'll say, sorry, no video games.

Joe Weisenthal: That's so mean. You're going to have to sit him down and be like, I'm so sorry, son. You know, there's this thing called the supply shortage.

Tracy Alloway: Yeah, I'm going to use this as a learning opportunity to explain how global supply chains actually work.

Joe Weisenthal: Totally. You know what the other the one interesting thing is, if you look at spot DRAM prices, which we have on the terminal, you know, there are a lot this is, I think, maybe the interesting dynamic. There's a lot of eye related things on the terminal that have been surging for several years, most notably some of the big chip companies, Nvidia, etc. if you look at a chart of spot DRAM prices, the big surge wasn't until late last year, and then since then it's gone nuts. So for a long time even there's this AI story and I mean the AI industry like propagated people were really paying attention to that, that area. And then suddenly it's just gone like, you know, haywire.

Tracy Alloway: Haywire. Yeah. And so obviously huge supply demand imbalances. We need to understand what's going on because I think the, the stakes are, you know, it's getting so much as you mentioned, a lot of companies are losing out pretty big. Others are making a lot of money.

Joe Weisenthal: Yeah, we sort of need to get a handle on, what's going on with the memory? Let's do it.

Tracy Alloway: I'm really excited to say we really do have the perfect guest. We are going to be speaking with Ray Wang. He's an analyst at Semi Analysis and he's recently published a brand new report, on exactly this topic. Talk about a memory supercycle and so forth. So truly the perfect guest. As I mentioned, he's, comes to us from Korea. Ray, thank you so much for coming on.

Ray Wang: Hey, nice to be both. And thanks for having me.

Joe Weisenthal: So many questions. Really appreciate you. Joining us, let's just start with the basics. Like, what is it about AI or what is going on? What is the core idea such that demand for memory and various forms is absolutely seemed to be booming right now in a way that call it the, supply side of the equation, very off guard.

Ray Wang: Yeah. I think, you know, to really to get kind of, some imbalanced supply demand dynamics, really, we wanted to get like, you know, what happened a few years ago, right? Because, you know, your capacity is like your investment few years ago. And what's happening, you know, in 2022 and 2024, it's it's sort of, a down cycle. Right? Because you remember, during a Covid, all the companies are try to log in to buy tons of DRAM. Right. And you know, there's a lot of, purchases there. Right. And this sort of like up cycle, right. Plays like a very, very short life cycle because like when now Covid gradually getting better. Right. You actually don't need that much DRAM. And back then the company will be like, hey, you actually don't have a sustainable demand, right? So they don't want to over invest to expand their capacity. And when you go into the downside, your CapEx typically being very constrained, very like, you know, company from their perspective, they try to be conservative. So that leads to 2024 and 25 that you're in incremental wafer capacity, the goal. So DRAM it's actually quite limited.

Ray Wang: But what's happening on the demand is actually you know demand was accelerating so fast right. You know, just you know, one way to look at this is to get immediate earnings. Right. And if you, when you, when you on one side, you demand accelerating so fast. On the other hand the supply side, your, supply just couldn't catch up. And there's more nuances behind that, behind the supply. Right. Because before they are just supplying all the commodity. Ran like DDR, LPDDR, right. All the memory that goes to a PC goes through laptops, goes through your mobile, goes to Nintendo Switches. Right. But you know what's happening. You know, in this I it's like there's a new thing called HBM. And think about it's like a specialized memory for directly for this reader. And that memory is extremely DRAM wafer intensive. To give you a sense. Right. You know, on the same way, for a basis, you can produce three more bits, if you do commodity DRAM, but you can only produce one bit of HBM. Right. And that ratio will actually going higher when you go to HBM4 in the next few years. Right. So in that way, on the same wafer basis, you can you can only produce that much. HBM and right now what's happening is like, hey, this is super, super, profitable. And why not? We dedicate more wafer for HBM. But so here's the problem, right? You only have that much wafer and you need to try to do everything. So that's handing out a lot of the supply going into the commodity DRAM. Right. So we have we have sort of both things happening.

Ray Wang: And essentially comes down to like I think really in using start really kind of people aware, it's like probably Q2 2025 that people realize, hey, like, you know, we actually could have enough, like enough DRAM, both enough, commodity DRAM and also HBM, for the demand.

Joe Weisenthal: So you mentioned the, the C word, just that. Oh, wait, not that C word. Commodities. You mentioned commodities. And when I'm reading these articles about a memory chip shortage, people use commodity type language, right? So you hear supercycle a lot. You hear commoditized chips versus, you know, other types of chips. Are they customized? I don't think so. But anyway, how much does this particular industry actually resemble a commodities like industry. And then maybe explain a little bit more why historically it has been really, really cyclical.

Ray Wang: Yeah. So I think a couple of reasons. Right. You know, fundamentally the, the sort of the cost per bits for the year end was actually coming that coming down every single year significantly. Right. Even, you know, most recent year this isn't coming down. But like the the standard layer coming down for every single year is actually slowing down. Like, you know, over the past like ten, 20 years. Right. So the cost was key going down right. So the cost is hard to be kind of like you know, the most competitive part. Right.

Ray Wang: Another thing is the earth. So the, the understanding, the sort of the committee that that set up the standard for the event. Right. So well you are doing in a similar products. Right. You just having, you know, oh this is kind of a little better for efficiency. So the the performance was a little better right here and there. So like it's harmful, you know, a memory supplier to significantly differentiate their products compared to their competitor. Right. So essentially what's the what's the things that will in the end will be the market prices. Right. And then when you're selling the similar products and you are targeting similar audience. Right. That sort of like on a commodity, you know, sort of semiconductor commodity products to me. And I think those are the two men characteristics I'm seeing for DRAM. And I think that's a little, little different.

Joe Weisenthal: For each VM, explain this. So there's commodity DRAM and then there's HB. What it HBM is a type of DRAM. Is that what you're saying.

Ray Wang: Yes. Yes. So, so in the sense of like sort of yeah. Like walk us through like what the commodity version and why there is this thing called HBM high Bandwidth memory. What that's all about.

Ray Wang: Yeah. Yeah. So essentially the emergence of HBM is really go down to a kind of model layer because, you know, this is when you like, try to scale and try scaling your air model. Right. Your thoughts is if you're promising, increasing significantly every single year, right. Because your computers increase, lot faster, but your memory bandwidth was actually still off is increasing by some very limited, especially if you are using the memory from, kind of traditional commodity DDR. And then the bandwidth is actually limited. So from a memory supply, they are thinking, hey, like, how can we expand the bandwidth to supporting low memory. They're going to, to skill scale the AI, right? So let's call ourselves HBM.

Ray Wang: So the HBM. That's so there's that, multiply 8 or 12 or even, you know, future 16 there and dies together. Right. So that making you have like essentially having higher, bandwidth, have enough capacity to support AI development, over the past few years and in the next few years. And that's very different. Right? You're stacking all these dice together compared to, hey, I'm just we just have like one DRAM products as a chip. Right. And also that also comes out to a point that HBM manufacturing also in the packaging of both front end and back end is a lot more complex compared to commodity DRAM. And that's why I say commodity right is more like commoditized. So I think HBM is less so because it's a lot more complex. Right. And for supplier you can actually differentiate either from the front end or back end technology. And when you know it's better right. And gives you a better margin, better ability to compare, compete with your competitor. Right. You know, stuff like thermal, things like that is the thing that will really stand out. Right? And that's very different from a commodity DRAM.

Joe Weisenthal: getting more and more layers on it kind of reminds me of the razor wars where like, you know, Gillette would unveil a three blade thing, and then the next day someone has like 12 or whatever. So if I'm a company like Nintendo, how am I actually purchasing chips? Because I think this is going to be important when we talk about what's going on right now. But do I have like a forward contract, with a major supplier, or do I have, I don't know, a warehouse full of chips that I bought previously? How do they actually manage that?

Ray Wang: So there's a couple of ways they do it right. And then we'll never do it in in terms of like kind of they will do it different way like in different environments and also different vendors. Right. So it's hard to say like, hey, this is exactly, exactly the way they do it. But technically I think you can think about like, hey, this is a map of the chips and kind of purchasing the next four months and this is a month we're going to commit, right? And we're going to sign a contract Pi same whether this term will be four months or if for the LTA you can go like as long as like two a year or even two years. Right. So that's the way you think about it. But people, you'll be like, hey, we signed this contract for four months, right? And sorry, for three months for a quarter. So by the end of that quarter, you negotiated a new contract pricing for the next quarter. So that's typically the way it work.

Ray Wang: But, you know, the problem for right now is for consumers wanting this, like the pricing probably is sort of I will say secondary because if you couldn't secure value, you couldn't even make any money. Right. So number one is securing value whether that, you know, make sure that fits, to the demand outlook. You are saying to us whether we can negotiate the best pricing you can, but given the overall pricing environment. So a slow volatile was so asking so much, it's hard to like not taking some kind of margin impact because essentially, you know, you try to get a chips right and you cannot just, you know, get too much away from your vendors.

Joe Weisenthal: I want to get more into the demand destruction element, because I think this is very important to think about where the equilibrium is going. But could you some just to clarify, what is it about AI specifically? Is it in the training? Is it in the inference where specifically in the process of building out AI services, does this voracious demand for DRAM come from so also you say let's see what go to be.

Ray Wang: All of this I think is really where like right now and you know and training you need tons of HBM that just everyone knows right. You also need a lot of the CPU, DRAM and less LPDDR3 and a DDR to go to the server. You also need HBM, right? So that's the training part. You know, inferencing. You also need HBM, right. And you also need the CPU at the end of course, of loss, workload. Right. You know, the main one will be the HBM and even the inferencing. Right. Right now they are supporting to fill in the core. And, you know, I think 80% now, I think right now, like the most important thing will be to call. Right. And the call is super, super like memory bound and memory intensive. And the memory importance of there was just will only increase to my understanding, especially given the long run this window continue to expand. And the inference usage and adoptions continue to going up. Right. So I think those are the typical way I will think about those. Right.

Ray Wang: And the last thing is, I think a lot of people talk about these days about generative AI, right? I don't think I you want to power loss. You need a lot of CPU, you know, a lot of CPU server, right. So what's also what's included in the CPU server? There's a lot of theory in there. Right. And a lot of I don't think will also mean you need a lot of inference as well. Right. And that go back to my previous home. You need a lot of HBM and DRAM. Right. So I will say like kind of different. And I got HBM and it's kind of everywhere. And we had to talk about storage my storage. And so everywhere, you know in doing this process.

Joe Weisenthal: Why does I need a lot of memory at all? So you just explain, you know, it's important for inference. It's important for training, whatever. But like why.

Ray Wang: Yeah, I think I think this needs to like goes down to the, to the model layer. Right. Because essentially when you are trying to train, let's say you want to train a model, right? And then you need just pure tons of data. You need to draw straight tons of data there to train your model. And then for the inferencing part, like you need one data after another to kind of compute one after another, sort of a chain of sort. Right. So to to do that process, how can you keep the previous data you process to keep it in there and the process to another data that coming out after it? Right. So that's why you need a lot of memory.

Ray Wang: Then go back to my previous polling. Right. The colon context window basically means like when you're when the things that you are process is longer and longer, right. Like how can you how can you have enough memory to process those. Right. And you know, I'm going to go example, if you use you guys definitely remember it, right. You know back to ChatGPT coming out. People just like hey I wasn't whether I like a very simple questions people right now it's more nastier. People are doing like, hey, can you write a, plenty page for how tensor can happen on people, right. Example. They will give like a 20 page report and they will wait like five minutes. Right. And so if you think about a process like they do all the calculation, all the researchers, and also give you the output tokens that the, the sample, answer you get from your prompt, less significant way longer than the prompts that you are given. Right. And that's also the longer response that you are getting when you go back to the GPT-3.

Ray Wang: And remember, usage, we know back to GPT-3 was the would, you know, the monthly daily users. But right now it was I think the use of GPT was 800 million. Right. And we are not included. The users for the Gemini, for cloud, for AI. Right. So if you multiply loss effects, I think the memory the most inefficient. And it's very, clear to me.

Joe Weisenthal: Yeah, it's really interesting. I know some people were talking about this, but it really strikes me in my own usage of AI, just basically how much more use it as capabilities have grown. Like it was sort of cute in late 2022 is like, oh, get it to write a poem about this. And I was like, okay. And then I would go, you know, several weeks, sometimes without having anything to query. And then, you know, these days, like I'm using it all the time, like trying to do things with code or scrape data, etc.. So just overall token consumption has grown massively with increased capabilities.

Joe Weisenthal: All right. Let's talk about the demand destruction element. Any commodity is going to be the price is going to be a function of supply and demand. And eventually prices get higher enough such that certain forms of demand just are no longer economical. And it's like, you know what? We're just not going to make as many video games or cameras or whatever else uses, memory. Are we starting to see any of that yet such that certain, uses and consumption of. So Joe doesn't have to buy a switch for his son?

Ray Wang: Yeah. That's right. Yeah, I actually bought a switch recently. So on the price have I haven't see price spikes under my reseller, I don't know. We'll see how the average arbitrage there works. But yeah, I think I think we're already seeing a lot of impacts, actually. But you kind of, you kind of vary in different kinds on the line, right in piece. You are surely seeing that still not knowable. Answer. So so they are all having a price hike for kind of different Apollo lines, right.

Ray Wang: And, you know, on the kind of demand side, are you also seeing that, I think, for the Chinese smartphone market, right. A lot of the analysts of the research firm. Right, a lot of the company was saying, like, they are cutting off their smartphone. It for example, I think MediaTek recently definitely earnings. Right. They are saying they are cutting off the numbers on mobile too. I think it was talk for 10 to 15% of 2026 outlook. That's very significant right. Not to offer same this year will be this year supposed to be 100 million. But you will essentially goes to a 90 million. Right. That's very significant. So I think we are already seeing the app in kind of very different place. That's right. Apple will be a good example here because they are saying, hey, like, hey, we have all the price hike, right? But we actually managed pretty well. They are saying the real sort of meaningful marketing impact will actually show up in the second half of 2026, in stealth, like the first quarter of second quarter. I think are some of the things I'm seeing. So it's definitely already showing up in the market. Right.

Ray Wang: I think what you can expect this year will be two things. One is sort of demand disruption in PCs, right by the consumer warnings or latte spec, whether it's a memory display that's very straightforward, but also the camera, right. Camera is also very important, I'm sure, in some of the display on a camera as well. And, you know, also on the, whole what we call it delay luxury because you don't want to launch a new products that have higher price than initially thought, which will impact your new launch performance, which usually then you are new to have better ASP at a margin compared to a previous generation products. Right? I think those are the things that we going to see, in the coming months, in the coming quarters. I think we'll all impact and people will start feeling it.

Joe Weisenthal: Joe, do you think we're going to get people like stripping old tenders for memory chips? Like, remember how during the commodity supercycle in, like, the early 20 tens, I guess there were all those stories about people, like stealing electricity wires.

Tracy Alloway: Yeah. It happened again in 2021.

Joe Weisenthal: Yeah I think yeah, we might I might get some of that.

Joe Weisenthal: Well, actually, on that note, is there, is there like a short term fix for this problem either in terms of design, maybe you can make the products more efficient in some ways so that it needs less memory or, recycling old memory chips, I don't know.

Ray Wang: Yeah. I think, you know, upon the demand side is a lot difficult, right. On an OEM perspective. Right. There's only that much you can do, right? You can just spec. Right. But I don't think to address the fundamental issues. Right. You are just this micro upon us. And so you can ship. But that is a double edged sword right? When you don't start going upon us, that as you look back compared to some of your peers, who doesn't just spec right and make you less competitive, that impact your final sort of, quarterly or annual shipment, I think what need to be addressed, where what can be done. Let's all be on the supply side. Right.

Ray Wang: So supply side, we are mainly talk about couple of vendors, right. The Micron Samsung and for legacy ones. Right. You have Wimbledon. Yeah. And China like CXMT and JHICC right. Last people. So I think the most right way right now this year because this year that really challenges there's a clean room constraint. So clean room is basically you have limited space that you can put all this equipment. And so manufacturing chips in fab. Right. So you have clean room shortage in all three major memory makers. Right. So your incremental wafer capacity coming online for 2026 will actually be quite limited. So in that sense, how can you produce more bit more DRAM bits while having only limited incremental wafer capacity.

Ray Wang: And the only thing you can do, it's, not migration. So for example, so in theorem processors now you have the most advances. Right now it's one C right. Going down I have 1B1A and one x bar. But right. You know I think why now they are trying to do is migrate as much as they can to 1B and 1C, which are the same way for basis. They are produce, a lot more chip, lot more bits of compared to a legacy. Not right. So doing that way on the same way from pieces you can produce more bits. So that should produce more, supply even though your wafer capacity is constrained.

Ray Wang: But a challenge is two things. One, how fast you can read about this. Not migration, right? Not migration is typically difficult. Well, I will say difficult is it only takes time, right. For to go to the most advanced now. Right. You need EUV machine. Right. You need all different kind of manufacturing processes. Right. And not every fab can already prepare to doing that. Right. So you what takes time to remember all those production for the new events that also you are having the challenge to like hey okay I want to do now migration right. But how are we going to balance the capacity with HBM. Again? That's another issue. You want to kind of think about it, right? Because some of the HBM process not it's actually using of windings. They are using 1B. So let's actually do events processing. All right. So even you are including 1B by your listener. You want to do more HBM. So like the increase here is actually quite limited. So I think those are in a way can potentially fix the issue. But you know, you know, from our house view, right. Even we factor all those in. I think this year we're still going to have see question if we can shortage.

Joe Weisenthal: Yeah. One of the themes that comes up when we talk about any commodities supercycle is that, sure, higher prices will eventually elicit more production, more supply. But in the meantime, you know, there's a sweet spot for the existing producers, especially when you have just a small number of producers that can produce at scale. They can they enjoy massive profits. Right. And there sort of going to be reluctant to build out new fabs because that's money going out the door. It also means lower prices. So there's always the sort of Goldilocks period for them before the supply comes on line, where they're just raking in profit. What do we see is the, industry's impulse to invest. And I'm curious, like, you know, whether it's Chinese producers, are they in a position to undercut, potentially undercut the profits of the Korean makers or the Korean makers seeing themselves as having a long window where they can enjoy large profits before they have to invest? Like, what is the thinking on the supply side about those big capital outlays?

Ray Wang: Yeah, that's a very good strategic question. I think, you know, every manager will probably give you a somewhat different answer or in kind of on the high level, probably same, but I think in general. Right. Like, you know, you know, look at like historical cycle, it's like, you know, if we are seeing quite a sustainable demand in the next few years and then clearly the capacity couldn't catch on the demand, they are going to expand the capacity in some time. Right. Is probably not going to be like, hey, we now see in 2026 the capacity economy online like meaningfully is probably 2028. So they will I think, you know will still announce. Right. And we are seeing that from Micron right. Micron is announcing a new name Fab in Singapore right. They are expanding now. And they have two fab that currently is under construction in the US right. Also doing tons of migration and recently acquire the new fab from SMC which is a Taiwanese shoemaker. Right. And they all just move on and you are already seeing some sign of that.

Ray Wang: And another sign you are seeing it's the CapEx, right? Both the Hynix, Samsung and Micron, their direct CapEx has actually increased quite significant this year. And you see similar trend in 2007 given I think, you know, they're going to probably, you know, try to expand more capacity and invest more in both spend as well. Also equipment that goes to events, events, events, equipment. Right. That go by certain point I was talking about the non migration. If you want net migration you need to use more advanced equipment to produce more advanced chips.

Joe Weisenthal: Yeah. It really does remind me of the oil industry, right? Where like there's a bust in the airline commodity price and everyone starts talking about being disciplined. Yeah. Yeah CapEx spend. And so it takes a while to, ramp up.

Joe Weisenthal: I wanted to ask about, I guess, the balance between HBM and other types of DRAM. How are people or how are the actual chip makers thinking about that? And is there is there a chance that because HBM, my understanding is that it it has higher profit margins? Is there a chance that everyone just pours into HBM and kind of leaves the, the more basic stuff in the dust?

Ray Wang: Yeah. So actually, you know, it's very interesting. I think what always seems for here when you wrote about this, for our institutional clients that, you know, know the market. Well, typically we think HBM has higher margin, right? Which is which is definitely true. But what's happening, you know, we see all of this like, spot price control price going out so much. Right? The margin of the commodity line right now, it's actually higher than HBM. So here. So that created a real dilemma. That's crazy that you talk about it because when your margin is actually going higher, why would you make more HBM. Like why. Right.

Ray Wang: But I think if you really think about like a long term perspective on the memory, supply it right before the whole airplane, your demand is really driving from a market, right? PC, mobile, automotive, industrial, blah blah blah. Right? HBM right now is surge as a new growth driver for the company, which we believe will last for the next few years. Right. And for the memory supply, they're thinking the same. And if you're not, continue to push your capacity to push the technology. When you're lagging behind, it's hard to coming back and then go back. Go back to my previous point that this is a area that it's relatively easier to differentiate your products to get more market share, more meaningfully. Right. So I think remember, makers are very, very value HBM. Even though right now the margin of commodity right is high. It right. They will I think they will still invest quite a lot HBM and they will do their best to sort of balance the commodity market. But I think, you know, I think this year at least the number I'm seeing, it's still quite a significant shortage. Yeah.

Joe Weisenthal: How do the Chinese, producers compared to the, Korean producers. Is there a is there a gap?

Ray Wang: Oh, yeah, I think there is definitely the gap. I think, like, you know, if you look at the memory in general, I think is probably three years is fair to say. It's again, it's probably four years. I will say, you know, that this is Mary Lang, you know, timeliness is like, can I solve today? You know, who knows what's going to happen in the future? But I think, you know, there's definitely a gap. You know, when I think about the competition pressure from Chinese memory makers, we always want to, like, kind of separate the Chinese market and also the China market. Let's also in the memory supply was looking at it. Right. So Chinese market is probably about 25% of the global demand. So really the competition is happening in China because for the leading Chinese memory supplier like CXMT, for example, I believe like 90% or 95% of the revenue is coming from, China, Hong Kong, Hong Kong, Hong Kong. Right. So meaning, you know, most of the data is really company in Chinese market. Company with Micron and Hynix and Samsung. I want to say on the right, not on the events on the high end. Right. I think it will still dominated by industry maybe makers, but CXMT are getting two momentum right? Right now, probably getting some of the low end of Myanmar products. And I also benefited by Chinese comments policy in terms of the self-sufficiency drive to, you know, whether it's for a commodity, right. And also, right now they are pushing for developments that support Chinese AI hardware.

Joe Weisenthal: So if there's a shortage and it seems like the shortage is going to be with us for some time, how are the chip makers actually allocating what supply they have? Does it go to, you know, the company that I think is going to be really big and important in the future, like, I don't know, an OpenAI or something, or does it go to an existing customer that's been buying in volume from me for years?

Ray Wang: Yeah, I think there's no doubt me that, you know, some of the highest, highest here. A customer will get a volume for sure. Of course there are some more complex like pricing negotiation behind, but I will I will say like the volume will be it still will be the most important thing. I think that's probably what they're going to do. Right. And, you know, before I think about like a buy customer, I will think about kind of by sector, I will say the server DRAM and HBM together will be the top, top priority. Because, you know, if for the whole market, in theory, we are seeing actually HBM and server, the unit together is probably like more it's more than half of the direct market. Right. And that's important. And it's going so fast. Right. And it's unlike the mobile. Right. Mobile is kind of flat. The mobile demand is really driven by the increased, direct contents in a mobile, which is quite limited. Right. If you buy an iPhone over the past year, you know, like I actually can't imitate how they're doing content increase. So I think I think they are those are the top two priorities, for the memory company to focus on in the next two years. And we'll see what happened. You know, after like, you know, the second half of 2027 or 2028 if there is, some structural change, but I think serve a different entry and will be the top priorities.

Joe Weisenthal: You know, the other day I was very lazy, which is not just just that one day, just one day. And, I use Claude code. You know, I have a bunch of screenshots on the desktop of my computer, and I, I sort of make my desktop look like, imagine I can't see folders. So I like drag and drop them and put them in the recycling bin. And I was very lazy. And I put in to Claude code. I should clean up my desktop, clean up my desktop. And I was like, there's so insane. I'm using, you know, Anthropic computers, wherever they are on the other side of the country to clean up my own computer desktop. Why do I even have to do a good job?

Ray Wang: Yeah, of course you're doing great.

Joe Weisenthal: And I was like, why do I even have my own computer? At this point, I started wondering, you know, I'm like, why do I even have a computer could to related to this sort of DRAM question could we just get into a situation in which people don't really have their own compute on it, on their own devices, because more of it, like Anthropic brain is sort of, or OpenAI's brain is sort of controlling the things we use. Why not just have a very low spec, low spec computer or low spec phone or whatever, or, you know, low memory, whatever, because I'm already using their computers, etc.. And are we just going to see whether memory or elsewhere this sort of migration or all of the interesting stuff and all of the memory and all of the compute happens elsewhere, and my phone just becomes a sort of, you know, or my computer just becomes, a screen with an internet access.

Ray Wang: Yeah. I mean, I don't I don't know about the idea of, like, you know, not having the personal devices, like, I wasn't a device, but it's just sort of seems like it over time. Does it make sense for me to have all of this actual compute inside my house when I'm just like, I might as well let them know I was feeling to?

Ray Wang: Yeah, I say it depends on like, what's your emperor's purpose? Why? You know, for example, on laptops, for example, if you're doing like video editing, very, very heavy. Right. Why are we still allowed to I think probably still need quite a last year. They. Right. If you are just, you know, doing like you know document work like how Claude events are, how are you going to use it. If I'm using like super intensively I think you still need like a good year end especially you all like pulling all different API from different places. Right. So I will say you just it really depends on the purpose. I don't think like, you know, at least at least I haven't seen, like, you know, the API kind of structure or, sort of the render structure in the devices, from projecting AI.

Joe Weisenthal: So going back to the beginning of this conversation and the cyclicality of the industry, people have been throwing around the term supercycle again, very commodities ask, is this just a super cycle? You know, maybe bigger than a previous turns that we've seen throughout history? Or do you think something structural has shifted here, perhaps, given the rise of AI and the fact that Joe is, you know, using Claude to clean up his desktop?

Ray Wang: This is a actually very dangerous question to read. And you are basically asking, right, whether this time is different. Right? Right. You know, I think, well, you know, I think there are a couple things that actually run ins with the history right now migration, say capacity rhino things. But I will say, like, you know, there's kind of a kind of differencing I'm seeing. Right? Because we really see, sort of supercycle that there's a new demand driver coming online. It's not only constrained demand, this thing is also constrained supply right before because we, you know, for example, in 2020 10 to 2012 there's so supercycle driven by mobile right. So mobile is basically hey we have a new product coming online and, you know quite a lot less that's our capacity couldn't catch up. But right now, I mean it's like, hey, okay, we need to do HBM, but you I, it's actually constrained to a commodity, right? Capacity. So I think this is the key difference that we are seeing with this. Right.

Ray Wang: Permanently that demand serrations is also last quite long. Right. And if you're really starting counting, sort of I will say the aspiration in terms of demand probably, let's say second half of 2023, where really I start kind of emerging in people's attention, let's say like to now it's almost probably like the same three years. Right. So you have two weeks, two years. And we are looking at this cycle to last until let's say, you know, safely to say like second half of 27. Right. So we are looking at like what kind of four year cycle. That's quite rare in the history. The usually like in a cycle if you look at historically for the memory is probably like, 15 months. I, you know, probably longest like 18 months, from the start to its peak. But right now we are going to throw to an area that is like, hey, that email was going up. Not a price directly, but the demand at least was growing quite significantly over the past few years. Right. And the pricing impact is happening right now, starting from, probably Q2 2025.

Joe Weisenthal: Yeah. Well, what did when you say, okay, the cycle could end by 2027 or the second half of 2027. What are the ingredients for it to slow down? Is it the more production capacity? Is it a slowdown in total demand. Like what what do you see potentially happening in 2027 that bring supply and demand more into balance?

Ray Wang: Yeah. So I will say that you and I will say like it's a more tricky area to sort of estimate because, you know, based on a number, you know, I'm modeling, we are seeing here, it's actually we are seeing the the demand is actually going to actually the demand is actually going higher and the shortage is actually going to worse. And the part of that is because a media and server, they are different content is going to see acts compared to. Well, right. And then if we get a demand video service the drop into 100. Yeah I'm sorry looping else sorry nothing else are doing 300. Right. And if that's the case and the demand will continue to be that strong. All right. You're actually in a shortage to even be worse compared to 26. And your theory should go a little worse, right. Because the memory makers going to try to make more HBM wafer. Right. And that kind of cutting out more on commodity DRAM.

Ray Wang: Oh, I think the reason I say it's more tricky because two things, I mentioned about the migration, it's actually happened this year. It also going to happen in 29 seven. Right. So tons of migration going to coming online have been completed. So that will allow for additional bids for the memory supplier. Additionally, the wafer should have lot more coming online by the end of 26. And for, you know, throughout 2027. Right. So I think those are two variables that you want to factor in, for why I'm seeing at least I think like 20 to 27 you're and going to still be shortage.

Joe Weisenthal: Yeah. We need a strategic memory reserve. Yeah. Put a floor on prices and the cycle. Actually on that note do you see a lot of stockpiling in the market or panic buying right now where people are seeing these forecasts and they're freaking out and just buying whatever they can.

Tracy Alloway: How are you get a bullwhip effect. That's right.

Ray Wang: Yeah. It's hard to say no. Right. Like, one of the good I think in the signal you're seeing in timing Samsung and Micron they're doing inventory was dropping every single quarter since I think Q3 2025. Right. That's just a 1 or 2 signal that you are not just, you know, buying the, some of the products on the shelf. Right? And then upon us and coming online that quarter, you're also buying, now products that are in their inventory, right? Yeah. You're trying to get as much as you can and, you know, you know, given all the demand and how it's developed so fast, like by month, monthly basis, it's hard to say, like, hey, there's no, sort of preemptive purchasing behavior happening. Right? Especially when the hyperscalers and the media labs, they're competing with each other, intensively, securing capacity on the hardware is sort of the baseline, right? You you you want to get the best equipment that you can compete, right?

Joe Weisenthal: Yeah. So obviously, if I'm buying a Nintendo Switch or if I'm buying a PC or maybe some sort of consumer device, then the increased memory cost is a meaningful driver of that. And maybe, you know, the company we have to raise prices, maybe I won't buy it or whatever. And you might get this. What about like for the big, hyperscalers when we see these huge capital expenditure numbers from the likes of a Microsoft or an Amazon, do these prices changes move the dial at these levels in terms of do they affect anything when we're talking about these buyers?

Ray Wang: Yeah, I don't think you know, number one for sure. I think the the the CapEx is not really all the CapEx increase is not because of the different price increase, but the price increase, thus going to have some kinds of impacts to their purchases of memory. Right. Assuming the hyperscalers are the direct buyer of the memory, which, you know, is not necessarily the case in kind of different locations, right? So if they are assuming they are the direct buyers of your memory, for sure, I if you know, if your initial call was due to hundred million to buy a certain amount in memory right now, probably you need to do some discount there for sure. Especially this thing is that's probably going to the discount to be more and more in the coming quarter. Right? So why would the thinking now try and are hopefully they are trying to do. But I just still very difficult. It's I try to have a long term agreement. Right. To have a secure commit to commit large value for a year basis and to hopefully to have better pricing. But it's really hard to, you know, to achieve that because why no memory supply? I want to do that. Right. We've we can you know, negotiate the pricing, in a coding basis so we can actually get more money. Right. You know, especially given, current, pricing and supply demand environment we want.

Joe Weisenthal: Thank you so much for coming on online. It's really the perfect guest. Let's, stay in touch and maybe we'll, revisit it in 2027 to see if things have eased.

Ray Wang: Yeah, hopefully.

Joe Weisenthal: All right. Thank you.

Ray Wang: Thank you so much. Care. Take care.

Tracy Alloway: That was great, Tracy. That was fun. I do feel like there's a lot of moving pieces there, but just for me, I in general, the crowding out is such a big part of the story. When you saw those big capital expenditure plans for 2027, like these are real macro drivers. They're going to show up in the CPI, etc. like that. And because it's like, you know, maybe we'll get a lot of productivity gains in the future. But right now that pace of spending is so furious. It is like a fiscal boom.

Joe Weisenthal: Yeah. I mean still stimulus. This is what happens when some of the world's biggest companies and most cash rich companies decide that this is an existential threat, right? There's no upper limit on how much they're willing to spend in order to survive. And so you could just throw money at HBM chips, I guess. And then you get the chip makers going, well, actually, like we want to produce a bunch of HBM, like, forget about DRAM, all that stuff. I do think like, yeah, you know, when it comes to some of the data center and energy stuff, you know, it's a little ambiguous how much it's affecting energy prices right now. But already the politics of that is very fraught. And then you're going to start upsetting the gamer community because they can't get act. You know, Nvidia has talked about, a big mistake and then they're going to upset my son because he can't get a Nintendo Switch, etc.. People are just going to start feeling it more and more, the sort of visceral reality that various resources that they thought they could get abundantly. It's like, nope, we've switched this line over to the data centers, over to AI, and it's going to become more this crowding out effect is going to become more and more real to people directly.

Tracy Alloway: Which is so ironic because when you think about AI, you know, it's this thing that exists in the ether, right? But at the same time, it has this huge physical impact on the world. And it's it's kind of funny and I guess, like, not what a lot of people would have expected.

Joe Weisenthal: No, I suppose I suppose not. The DRAM people, if you anyone who has a terminal or just should just look up DRAM prices because this is

Tracy Alloway: I thought you were going to say anyone who has a terminal should start stripping out

Joe Weisenthal: everyone who has a terminal should just unscrew the bag and pull out the DRAM. No, but if you had, you people should just look at the chart because it's really here is this thing that's very sleepy. I mean, this was true commoditized tech. You know, this was the low end, in terms of what people were excited about in chips and so forth. And as we mentioned in the beginning, you know, costs generally were on this downward trend and so forth, because growth was modest and technology continues to improve. And then it literally just looks like an L really in like the last four months just gone completely nuts. And so yeah, kind of a fascinating spot to watch. And it's just interesting to the point that you were making how much it really is like a commodity supercycle.

Tracy Alloway: Yeah. Commodity cycle I well I guess in 2027 maybe we'll find out. We'll see if it balances out. Shall we leave it there?

Joe Weisenthal: Let's leave it there. This has been another episode of the Odd Lots podcast. I'm Tracy Alloway. You can follow me @tracyalloway And I’m Joe Weisenthal You can follow me @thestalwart Follow our guest Ray Wang. He's @rwang07. Follow our producers Carmen Rodriguez @carmenarmen, Dashiel Bennett @dashbot and Cale Brooks @calebrooks And if you want more Odd Lots content, then you can find that including out daily newsletter over @bloomberg.com/oddlots And you can chat about all of these topics 24-7 in our discord, discord.gg/oddlots And if you enjoyed this conversation then please like the video, leave a comment or better yet, subscribe! Thanks for watching.