Great. Welcome back, everybody. For those of you who do not know me, I am Joe Moore. I run U.S. Semiconductor Research for Morgan Stanley. Very happy to have today the executive team of Sandisk, Dave Goeckeler, CEO, and Luis Visoso, CFO. Guys, thanks a lot for making the trip here.
Hey, Joe. It is great to be here. Really good.
Really transformative-
Hey, Joe. Before we start, sorry.
Oh, yeah. Sorry.
Got a special thing to read.
Very important. We will be making forward-looking statements in today's discussions based on management's current assumptions and expectations, including with respect to our technology and product portfolio, our business plans and performance, market trends and opportunities, and our future financial results. These forward-looking statements are subject to risk and uncertainties. We assume no obligation to update these statements. Please refer to our annual report and Form 10-K and other filings with the SEC for more information on the risk and uncertainties that could cause actual results to differ materially from expectations. We will also be making references to non-GAAP financials, and reconciliations of GAAP and non-GAAP financials can be found on our website. Thank you, Joe.
Thank you. Sorry, I should have left you in time for that.
Okay.
Really a transformative kind of 12-month period for your industry and for your company. Maybe we could just start with, this time last year, we saw data center NAND starting to really take off. Pricing was really good.
Yep.
Demand was really strong. You guys were kind of modestly participating in that, and benefited this time last year from a tightening NAND market, for sure. One of the things that struck me from your last earnings call, that you had 18x growth, if I am not mistaken, in data center, and it has gone from single-digit percentage of revenue to closing in on half. What led to that? Obviously, the market has been good, but how is Sandisk's participation also so much stronger?
Yeah. Joe, I mean, it has been. I mean, to say transformational last 12 months may be a bit of an understatement. I do not know a better word. That is a good word. The market is changing in an incredibly dynamic way. In many ways, I think even what is happening in the market right now, a lot of the change that has happened over the last 12 months is just starting to show up, and we can talk some about that, about how the market is stratifying. Different parts of the market are growing at different rates. For Sandisk, like we have been traditionally, I think if I think about Sandisk, there is kind of maybe three big arcs of the company, and clearly a very, very strong consumer company, consumer brand, intellectual property, just enormous respect for the people that built this company and built the Sandisk brand.
We sell in 350,000 points of presence around the world. We have a big share position as a consumer company. That has always been an important part of the business for reasons we talked about back in August, our investor day. It is less cyclical, all these kinds of things. It is kind of now, we have this huge growth in data center and change in the market. So it is not as visible as it used to be, but it is still a big part of the portfolio and the strategy. I think maybe the Western Digital era of Sandisk was characterized by what I think is the rise of client. That is really not a big surprise. The play that was going on in those years was in the PC, in the device, replacing the hard drive with the client SSD.
What better company to understand the use case, the requirements, customer relationships, than Sandisk-Western Digital combination? I think when you look at the end of that era, we had a very large client business, like 25% share when we had, what, 13%, 14% bit share. So very over-represented in clients. So we were kind of this client, consumer client company. Now, in that era, when I came in, we spent a lot of focus on really building out the enterprise portfolio. To build a product takes a long time. We have to build an ASIC. We've got to get a lot of things lined up. There's TLC product, QLC product, and all that work was going on behind the scenes for many years.
Really, the emergence of our portfolio and the emergence of the data center growth were really highly correlated, and we benefited from that tremendously. We literally had the right product at the right time. We always had good customer relationships with the data center as well, and partly out of the HDD business, which is a data center business. So, you have the right product at the right time. You recognize the market change early, and it was this time last year, it was really right around August, September, October, we could really sense the market was changing in a major way. That leads to, as you just said, 1,700% year-over-year growth in our data center portfolio. It went from a small part of the business to now a large part of the business and continues to grow and will continue to grow.
I joke with our sales leader that when you turn in 1,700%, maybe it's going to be hard next year to do that again, but we'll see. Really, data center will continue to be a growing part of the portfolio. We'll still have the client business. It will still be there. Spectacular, great companies. That market's going through a lot of change, which we can talk about. But also, we're seeing the emergence of other segments that are in that edge business that are growing quite well. I mean, automobiles, we talked a little bit on our earnings call. When we look at exabyte growth in that business, it's twice what the market growth is over the next several years. Physical AI, all these kinds of things we have customers talking to us about. So we're still very optimistic about that edge business.
I think it's going to be a different business. Again, it's not a business that's dominated by selling NAND to the consumer. But our business is clearly changing in a fundamental way, and you're going to continue to see the data center portfolio continue to grow.
Well, I think one of the things that makes our job difficult and has created angst for shareholders at times is you do have this dichotomy that there's demand destruction in some portions of the client business. For me, the true north has been the data center fulfillment rates, which are still very low. People aren't getting anywhere near what they need. Shortages of HDDs are helping you guys. Shortages of DRAM are helping you guys in different ways because you use NAND to supplement those things. To me, the true north is that things are really good and are likely to stay that way. But we know that there are challenges in PCs in particular, and that leads to, from time to time, okay, there's module inventory buildup here and here because demand's not that good.
Right.
How do you guys think about that? I don't imagine you spend a lot of time worrying about it, but just how do you make sure that you're adjusting properly, and how quickly, if there is a weakness in PC, how quickly can you adapt and move that to data center?
Actually, we pay very, very close attention to this. It's a very dynamic business that moves very quickly, and I think one of the great things about Sandisk, because we're a very agile company, we can move very quickly, we can pivot very quickly. You're right. I think the reality of what's happened over the last 12 months is NAND has become a very valuable technology. I've always think it's been a very valuable technology. I can tell you, coming into the company as a lifetime technologist six or seven years ago, and talking to the engineers, understanding the manufacturing process, just the enormous intellectual property and the amount of effort that goes to this and the skills of the people doing it's extremely difficult domain to work in.
It just was the markets it was selling into, the business model was essentially taking that product, packaging it up in a device, again, a spectacular device that's very well designed, but then selling that to the consumer and monetizing that volume of NAND. So, that created a kind of back pressure of certain business practices. Highly negotiated, every penny that was paid to the supplier was a penny of profit that the consumer, that the demand side didn't have. I think the demand side had the ability to shape demand a little bit. How do I advertise? What's my mix? All these kinds of issues. Well, now we have the market is predominantly a data center market, and I think that the last 12 months we're coming off of, 2026 will be more than triple-digit growth in data center exabytes.
Data center will exit the year as more than half the market. These fundamental changes, I think we've been very good at recognizing them and taking advantage of them and trying to change the business model of how we work with our customer. But again, back on the PC side of it and the smartphone side of it, there's clearly parts of the market that if it's dependent on basically very inexpensive NAND, that's going to be very challenged, right? We're watching that very closely. We see it in the business, and we're adjusting appropriately. Clearly, we're pivoting more towards data center, and you can see us pivot our portfolio very quickly. But NAND is a very complicated business on the inside. We have multiple nodes. We have multiple products.
Not every node can go to every product, and so you're constantly trying to fit those together. Luis is kind of the tip of the spear of doing all that. I mean, you want to talk a little bit about how you think about that, Luis?
Yeah. The way we think about it, Joe, we have nodes and dies, right? We can move supply one way or another, but not 100%. You have older nodes, which can mostly serve PC and mobile phones. There are other nodes, the newer nodes, that can go anywhere, right? So, what we normally do is we obviously want to maximize the value for new nodes, which are in very high demand. If they are older nodes where we're not seeing the demand we need, we can adjust production lower into those nodes so that we're not producing bits that we cannot sell at attractive prices, right? So, you can still maximize all your production to the newer nodes.
You maximize your presence with data center and the attractive parts of PCs and mobile, or the edge in general, but you really reduce your production in nodes that are not selling at the rate you want.
Great. Thank you. If you think about the data center opportunity, I feel like investors feel like it's maybe sort of fragile, right? That these margins are not infinitely sustainable. On the other hand, you have customers putting down multiple billions of dollars of deposits for 2029, 2030, 2031 memory, so they think it's durable. I guess maybe one of the questions we get a lot is, what happens if there are incremental issues in land, power, shell? Looking at what happened with the midterms, but also looking at what's happening internationally, new BIS controls on certain regions. If we run out of space, what does that mean for you guys? What does it mean for the LTAs? Do you hear any anxiety from your customers about that?
That doesn't tend to be a conversation we have with our customers, right? I mean, these companies, think about the companies we're working with. We've purposely pivoted our business. We have eight NBM customers. We've disclosed that three of those are U.S. hyperscalers, right? So those are naturally very large businesses. They've been doing this a very long time. They have by definition, a global footprint, that they manage, have been doing for quite some time. And even a company like Amazon has been I'm not saying that's one of the three, but even a company like that has been doing it with distribution for how long, right? Very long time. And we went through the period of wonder, where are you going to build all these warehouses? I think it's certainly something to monitor.
I think as new technologies come out, you tend to have issues with this or concerns. I mean, when we were rolling out 5G networks, it was, "Where are we going to put cell towers?" And there was all that kind of stuff. And so, it's not surprising we go through this, but I think our customers are very, very good at managing this, right? And they understand what their demand is. And by definition, we're working with customers that, as Luis would say, are winners. We think they're going to be long-term winners. They have very large, at-scale businesses. They know how to manage these issues within their own enterprise without making it our issue. What we predominantly see from them is them coming to us. They want to strike deals. They're interested in certainty of supply. Everybody is interested in paying the lowest price, right?
But it's less this consumer kind of use case negotiation as opposed to, "I have this incredibly growing business. I need to make sure I have supply." We supply it, and this transformation we've talked about in the market, we've recognized that and used that as a way to transform the way we transact with our customers. And they're willing to go there with us, sign contracts, put financial guarantees aside, give us certainty of economics, not just give us certainty of demand, and we'll talk about the price later. That's the consumer market. This is, we'll talk about demand, and we'll talk about the economics of it all at the same time, drive that partnership. And, as we talked about on our earnings call, which we continue to see, is the usual case is those customers coming back and wanting more.
And that's what you would expect when you have a partnership with somebody else. We want them to come to us when they need something.
Yeah. Thank you. That's great. AI has already been transformational, but you've talked about a lot of growth ahead. Can you talk about how customers are using NAND today and how you see that evolving in the data center?
Yeah. This has been, I think if you go back a couple years ago, AI was all about training. I mean, we got to train models. Again, the spectacular thing of the cloud, there's never been enough compute in one place where we can do what we can do now. All of a sudden, now we can train these huge models. Clearly, model training was all about ultimate performance, HBM, DRAM. But as you start to move out into inference, and now you have the models built, and you have to scale inference to the entire world. Again, go back to these companies that I've said this before, we're very lucky as a society to have these companies that have global footprints.
And we can get this new technology that is just incredible, and we can basically distribute it to almost everybody in the world very, very quickly. As you go through that exercise, it's very important to do that in an economic way, the most economic way possible, right? Just in general, that's going to drive you, from a memory architecture point of view, that's going to drive you to NAND, because it's available. I mean, what's the entire global DRAM production? My team tells me, I'm not an expert on DRAM, my team told me, like, less than 100 exabytes.
Right? I mean, NAND is, you're talking about a zettabyte market. It's available. It's highly scalable. As you're looking at that architecture and how do I scale this, because if I'm going to scale it needs to be the most economic answer. That's going to drive you to NAND, and we continue to see that. KV cache, staging, model post-training, all of these kinds of issues. How do we use more NAND, and how do we optimally use NAND? Because it's available, it's cheaper, and that helps us scale faster and helps them make more money. To the point, Joe, where this is becoming quite fascinating over the last several months, is customers are coming back and looking at attributes in NAND that we never used to talk about. NAND was always, "Deliver me density," right? TLC, QLC, more layers, more density.
Now it's, "We need durability. If you can deliver me durability, I can replace DRAM in this part of the architecture." I'm not saying replace all of DRAM. DRAM's a spectacular technology. But on the margins, if you can find cases where you say, "Hey, I use highly durable NAND, SLC, more than an order of magnitude more durability than TLC," then that's good enough. With DRAM being 20x - 40 times more expensive than NAND, that's an obvious economic trade-off for a system designer. We're actually seeing NAND. You go back a year or two ago, and people talked about it's all about density, commodity pricing, all these kinds of issues. You look at where the market is today, NAND is, okay, you have to deliver density. We know how to do that. That's not an issue. Now it's performance, power efficiency, durability.
It's like becoming this multi-attribute technology that is being fit into the architecture in more and more ways, and that just continues to drive more demand for our product.
Yeah. I think one of the points of contention that I hear is all of these things are in short supply at the same time. To me, there's a reason for that. When you lower the memory content of the rack, which NVIDIA did, it means we need a lot of really fast local storage, because if there's something that's not in memory, I need it to be very close by. You push one thing down, you push another one up, and so you end up with everything being in short supply at kind of the same time, because we're all making do with 70% rates of fulfillment on what we actually need. Is that fair? Do you disagree?
I think that's partly fair. But I think that the motivating factor is a little bit different than scarcity, right? Scarcity is part of it. I think it's taking it from you have to move from the design of the component to the design of the system. When you're starting any technology project in the early days, you really kind of overwhelm the architecture with everything you need. I don't care about power, I don't care about cost, I don't care. Just give me everything I need, because I don't know what use case I'm trying to solve, and I don't want to be constrained. Right? That's great. But at some point, if you're going to scale that technology to the world, right, that's a big deal. I think that's where I start with AI. I start with any technology.
A lot of technology you can make work on a small scale, you know? When people show me a demo of something, I'm very skeptical. Okay, you built a demo. That's nice. Then you build something that's like a proof of concept, or then you build something that can scale to, you know, whatever, hundreds, thousands, millions of users. But the technology where you're going to scale to billions of people, and everybody on Earth is going to use it, that is something. When you do that, you have to be, as a system designer, relentlessly focused on cost. You have to be relentlessly focused on the use case you're trying to solve. I can't just deploy this technology that's just going to do everything, right?
I need to figure out what my user's going to do and then optimize around that, because if I don't do that and I deploy a technology that's too expensive, and my competitor does it more economic, then that's not a very good business strategy. I think what's happening with AI, because the use case is so compelling. I see it in my own company. I mean, the number of ways we're using the technology is incredible. I mean, just literally hundreds of different use cases across our how to lower time to market on our product development, right? Taking months and months out of the usual cycle of building something that takes years. Yield improvement on our wafers.
We can now take 20 years of accumulated expertise on how to debug NAND problems, put them in a model, and when we see a new problem, almost instantaneously figure out what the answer is and make the change. So, the use case is so compelling, and the distribution of technology is happening so quickly. There's this relentless view of, "How do I build it in the most optimized way?" I think that's the primary issue, and I think the scarcity part is the secondary issue, a part of that.
Helpful. Thank you. Maybe along these lines of NAND usage in the data center, High Bandwidth Flash was something that you guys started talking about, and it felt kind of, okay, this is an interesting thing that maybe someday will happen. I went to Open Compute a month ago, and SK hynix spent literally 1/3 of their memory presentation talking about just this. It seems like we're moving towards an idea, towards something that's going to get implemented here pretty quickly. Can you talk about the technology and what it means for you guys?
Yeah. I have to give our team just an incredible amount of credit for this. One of the things that people used to ask me, I would get this question all the time when we were doing the spin-out of Sandisk, "What are we going to be most surprised by?" The answer to me was always easy. It's going to be innovation, because the team is just incredibly innovative. The team that's, you know, the Sandisk memory team has been there for decades. I think they had this insight many years ago, which was, as we get to inference and we have to go through the process I just described, I'm going to have to deliver something in a more cost-effective way. Models are going to get bigger.
This idea that I could scale inference with the current technology, that there was a better way to do it. That if we could develop a new memory architecture that essentially, again, brought the density of NAND. NAND, whenever you think of density, you check the box. We got that solved, and we got many generations of the future where we can deliver more. But we needed to, can we increase the bandwidth? Because, inference, you have this huge model, you're pipelining it into a CPU, and you're running queries against it. So, it's kind of a deterministic process. Could we design a NAND die where we significantly increase the bandwidth? Turns out, if you go to a NAND engineer that's been working on NAND for 20 or 25 years, and they've been told their whole life, "Build me more density," they completely optimize the technology for density.
But if you go back and you say, "Well, no, no. Okay, we got, figure we got enough density. Design me a die for more bandwidth." "Oh, you want me to design a die for more bandwidth? I can do that." So they start working on it, and turns out they did come up with a die where the bandwidth, we're not trying to replace DRAM. What we're trying to do is get the attribute of our technology good enough to solve the use case. That's the essence of product management. That's the essence of system design. Only apply what you need and not anything more, because the more is just cost that you can't recover. Then the third part of it was durability. You got to make it more durable.
We did a bunch of work on that, and it turns out we had this conviction that we could build a product that had the bandwidth that was appropriate for inference, had better durability, and had 8x-16x the density. That turned out to be an idea that the industry resonated with as they started moving towards scaling inference, or bringing inference down to a device where real estate is precious. We are going through that process. We are optimistic about it. We have now got far enough along where we have got enough input from people that would use this technology to tape out a die, and that is in the fab right now. It will come back sometime next year. We are building a controller that would go with it.
We will put that in customers' hands, and then we will really start to understand how they would use it, what the scale would be, exactly what the use cases are, because it is not plug compatible. We are not saying, "You just take this out and put HBF in." It is a systems play. As I said before, you got to change the rest of the system a little to accommodate our technology, but if you do that, you are going to get the performance you need, and you are going to get 8x-16x the density, which is very attractive.
If that happens, is there LTA to support that? Or it seems like this is more emerging. It seems like it is an incremental upside demand. Can you meet that demand if it is there?
I think we have ways to meet demand, as we talked about in our investor day. We have a lot of productivity in our technology. That will be something we will talk about at the time. But Luis can talk about when we have already had this case where customers come back and they want-
Exactly.
More upside to what the NBM we just signed. How does that go?
Yeah, I mean, we signed these NBMs, and to what Dave was saying earlier, customers are coming back to us and saying, "Hey, I underestimated the amount of NAND required for inference." I think that's a pretty good statement. While HBF is not included in our contracts, the relationship between our customers and us is just so much tighter from a technical, from a commercial point of view. All these discussions come a lot more natural because we have these QBRs which go a lot deeper than they have ever gone. A deeper relationship, deeper partnership, not just on today, but also in the future.
Great. Well, I want to talk more about the NBMs, but maybe first just to finish on the technology, and not to nerd out too much on this, but the single-bit per cell NAND is something that Lam Research has talked about, and it's important for generalist NANDs because it does soak up a lot of supply if we move that way. It's a lot fewer bits per wafer if you move to a smaller bit per cell count. Are you seeing demand for that? It seems like for you guys it's still relatively small. Maybe speaks to the durability importance that you talked about. Can that be something that's impactful down the road?
Yeah, I mean, this is why I talked about it on our earnings call, because it's kind of like, when I look down the road of what's super interesting that's going on in the business, this rises to the top very quickly. Because it's the customers coming back to us and saying, "Hey, can we talk to your NAND engineers? Because we want to understand the durability of your product better." It's exactly these use cases you asked earlier. For example, I'm going to more agentic workloads. Well, the CPU has what we call a scratch pad where it's keeping track of its work as it goes on for however long. You're going to write that a lot more times, so you're going to need more durability.
The natural thing would be, "Well, let me put DRAM there." That is probably over-spec'd, and there is not that much of it in the world. But maybe we could use SLC NAND. I think on our side of the business, people are like, "SLC NAND is back? Really?" Right? That is something like, if you are a density per, you have been told density is the most important thing your whole life, and now people are telling you, "Well, wait a minute. We want to come back, and maybe we can use this attribute of your technology that where the density," what that system thing I said before, "it is good enough for what I need to do, and it is very available, and it is plentiful on a relative basis." So I am going to put that in. It is like, great.
I would say we are at the point where customers are giving us insight into the way they want to do it. They are curious about what the dynamics of it are. Explain to me exactly what the durability is. By the way, I mean, we are not going to get into the super detailed right now, but the durability across all the different suppliers is very different, right? This is not a, "commodity," right? This gets down to the electrical properties of your cell, how good they are. It is a very detailed technical conversation, but we think we are in a very, very strong position here. What we are seeing is this kind of detailed input for our customers where we can see, hey, there are use cases here that are going to start, over the next couple of years, could start to drive maybe tens of exabytes of demand.
But that is, to your point, there is every. Let me say one more thing before I get to that. You can take any of our products and you can run them in SLC. We do not need to build a new node, a new die, any of that. You just configure the product. If you were planning on 20% supply growth a couple years from now, and then you had 1% of that or 1.5% of that end up being SLC demand, you would basically bring that supply down to maybe 16% or something like that. The impact is quite significant, and that is why it is something, to me, I find it super interesting. Number one, because it shows the differentiation of technology. It is completely orthogonal to this idea is, oh, NAND is just a commodity, right?
The other side of it is, it is going to change the demand dynamics potentially significantly, even in relatively small use cases.
Great. Well, given everything that we talked about for the last 10 minutes or so about the growing importance of NAND in the data center, it kind of brings you to the NBM long-term agreement conversation. Can you just give us an overview of how you thought about structuring those? In 2025, Joe would have been very surprised that we would be negative, that we're capped at 85% gross margin in 2031 or something. Just how do you think about the trade-offs between free-floating and contractual? It always comes back to people don't believe the enforceability of these deals. Can you give us your views there?
Yeah. We're trying to solve many problems. Think about two primarily, right? We want to make sure our business is attractive financially, and we want to make sure it's durable, and we want to achieve both at all times. That means you cannot maximize one over the other, right? Because they sometimes go in opposite directions. So our customers, as Dave said, came to us and said, "I need supply certainty because I'm building these data centers and I want to make sure that when the data center is ready, these racks are filled with NAND." We wanted demand certainty. So the two come together, and since the numbers were so big, we needed to start to think about a different construct which served their needs and served our needs. So we started these conversations.
We have eight NBMs, new business models, where our customers agree on a certain amount of time they're willing to commit to a certain amount of bits. Those bits are fixed. It's not that we're going to negotiate how many bits they take. We know exactly how many bits they're going to take for the full term, by year, by quarter, and most cases, by month, by SKU. So it's a very detailed plan that we know exactly how we're going to execute. Once you do that, yes, there is a conversation around price. The first portion of those NBMs, in most cases, include a fixed portion of fixed pricing where we know exactly what the price is and we get to an agreement to our customers that meets our needs and their needs, and then we establish a floor and a ceiling.
That's not true for every contract, but just in general, that's kind of the construct. The floor pricing for those contracts, we said in aggregate, will deliver at least 80% gross margin for every single year. So we feel good about the construct. We have, as I said earlier, built very close relationship with our customers, both commercially and technically. So far, the only thing that we've seen is customers come back. Customers come back and say things like, "We underestimated the demand, the NAND needs for inference," or, "We need more. Our lab folks were actually right, and we need more NAND." So that's very encouraging. We are very choiceful, very selective in which customers we sign these deals.
We only have eight, and that's on purpose because we want to build these very close, very deep relationships versus having relationships with 20 customers, which just cannot be as deep and as close as what we want.
Philosophically, how do you guys think about this? I mean, it seems like it gives you a lot of visibility into your business. Seems like there's a lot of positives from it, but you're also preserving some that is free-floating. Just how much of the business do you want to have under these NBMs over time?
I think that's something that Luis and I talk about a lot, and it's something we've been doing more work on in the last, I would say, three months because we're at 2/3 for FY 2028, so there's less left, right? At the same time, the demand is higher. I would say you go back to nine months ago, as Luis said, it was customers coming to us and saying, "Hey, we've qualified your product. We love your product. It's going to be an important part of our infrastructure for years to come, and here's how much demand we need." The numbers were pretty eye-popping, and so that led to these kind of contractual structure that we came up with. But that was just a few customers, but now it's become, I guess I could say, the predominant way customers approach us, right?
There's people now that, now the inbounds come, right? Inbounds come with, "Hey, I'll offer you a three-year contract. I need this supply certainty. Here's what my demand is." We have a lot more choices, and so we're trying to fit that together. I think we'll know more over the next couple quarters as to where we want to settle in for the amount that we commit. I wouldn't be surprised if it's higher than where we are now. But there's always going to be some that is not under this contractual structure, and there's a number of reasons for that. First of all, we have our own consumer business. That's like our house business, right? That's going to get supplied, and we're not going to have a contract with ourself.
Then we have some customers, very attractive business, but they just don't consume enough to put all of this machinery around it. There's long relationships there, it's fine, right? Those are good businesses. There's a couple of customers, quite frankly, that we've had contracts with on a year-by-year basis. There's never been an issue with those contracts. Even in the depths of the 2023 downturn, customers honored their contracts. These are not public, obviously, because we don't talk about individual customers, but we have use cases already of customers that honored their contract even though they could have bought in a different way at an incredibly cheaper price. But that's the way business works. You sign a contract, you live up to it. Then there's another set that I think about where NAND is like an evergreen business.
There's always somebody coming up with a new way to use our technology, and you want to have some supply where you can engage in those markets and make sure that you're seeing what's coming. But the predominant thing we're going to do is have, as Luis said, very, very substantial, really important customers that already have very large businesses. What did we say at earnings? 95% of our customers are investment grade.
Yeah, and they generated like $150 billion in cash the previous quarter, right? So very good customers, very established, very successful.
We run a big business. We produce a lot of product. Part of the NAND business, you've been a part of this business for a very, very long time. When you're producing an enormous amount of product every quarter and you're selling it on a quarter-by-quarter basis, that's a kind of a hardscrabble business. It's a constant negotiation all the time. But because of the rise of data center, and now it's going to be significantly more than 50% of the aggregate demand of the whole industry, you just have these massive customers that are going to drive incredible amounts of demand all by themselves. So to believe that the market is going to continue the way it worked in the past, when on the demand side, the customers are just completely different. They have different financial profiles. They're different scale.
They have different capacities in how they procure. Their businesses are not dependent on how much NAND they sell. If they do not have enough NAND, their business does not work. It is just incredibly different. I think you started out by talking about some of the markets that are under stress. I think this is a market in transition where now you are full on seeing all of this change happen. It is a little confusing in parts of it, but I think a lot of the old heuristics of the way you would look at the market and you would interpret a signal out of the market and then apply it to the whole business, that is very tricky right now. Very tricky. Because every contract we sign, or that anybody signs, by definition, changes the market.
This kind of highly transactional market continues to just get much smaller and much smaller. I think it will always be there, but the idea that you could look at how this market is behaving and then say, "That is the way the entire market is going to behave," that is not the way the business works from the inside.
Very helpful. I will open up to the audience after this, but just on the supply side, you have talked about all of these demand drivers. Your CapEx has been incredibly disciplined, I would say, given everything that is happening, both you and Kioxia, your partner. The rest of the world kind of has clean room shortages and is incrementally allocating more of that clean room space to DRAM. I love the way this is playing out, but are you guys tempted to say, "We are going to spend a lot more because the demand is there and we are the ones that can create the supply to meet it?"
We are not. I mean, we have been, I think, hopefully remarkably consistent. We launched the company in February of 2025, and Luis and I got on stage and we said, "We think the market will support mid to high teens bit growth over a long period of time." We promptly got a $7 billion valuation for our company. Everybody basically told us, "You guys are crazy," right? "That is not true." We said, "Hey, we think by the end of 2025, pricing is going to be up." The consensus view was, "That is not going to happen. Price is going to be down." Then we got to January, and it was obvious now that something had changed.
Every time I would do a session like this, not with you, with your peers maybe, I do not want to indict anybody, but the question was always, "When are you going to raise supply?" "Are you going to raise supply? Are you going to raise supply? Are you going to raise supply?" Then we got to like two months ago, and it is like the market is oversupplied. Are you going to cut supply? It is like, you can kind of play this game all over the place of like, if you look at the market today, you cannot make a decision. You have to look at where the market is going to be two years, three years, four years from now, because that is when any capital decision we would make is going to-
Have an impact.
Have an impact. We continue to believe that over long periods of time, we are talking mid to high teens bit growth. Is it going to be a little bit higher some years? Yes. Is it going to be a little bit lower some years? Yes. Luis said it earlier. What we are going to do is, if we do see gaps in demand, and you are seeing gaps in demand, like in the PC market right now, from what we just said, changing dramatically, we have the ability to go in and adjust supply on a dynamic basis, and we will take those opportunities when they come. But we are very happy with a business that meets our long-term model, 80% gross margin, 70% operating margin, 50% free cash flow margin, and grows mid to high teens. I am very happy with that business.
Pretty good business for you.
It's a pretty good business. I mean, it's a pretty good business, right? I think it's a good business. I mean, we're buying it. The two of us are buying it. Every day, we're buying.
Pretty high percentage of it.
Yeah. It turns out.
Great. Let me see if we have questions. No, so I can go longer, but if we do have anything. Maybe, one question that always comes up, particularly in Asia, is China NAND. Maybe just talk about Chinese competition. How do you think about it? How are you seeing it play out? Are you seeing competitive impact? And maybe, what I think is sort of a silly notion that there's a bunch of NAND supply in China that have just American companies qualified everything. Can you talk about that dynamic?
It's a big warehouse just waiting to sell on the cheap.
Meanwhile, NAND in China is actually more expensive than the rest of the world.
Yeah. I think this dynamic is. First of all, I'll make maybe one observation about this. NAND is, and true for DRAM too, just memory in general, there's no ecosystem in this business to enable new entrants. That's very important. If you look at the logic business, and those are all friends of mine, it's great businesses, but there's a whole ecosystem there for people to enter that business, including all the customers. All the customers. Any customer of scale is building their own CPU, TPU, whatever your variation, because of this dynamic I said earlier. They're driving a systems play. When I drive a systems play at scale, I need to drive the cost out, and I need to build only what I want. I don't want what everybody wants. I only want my own thing, so I'll just build it myself.
We've seen this, these very large companies have done this extremely well. I used to do this when I was at Cisco. We built our own ASICs because they were just for networking. I didn't want a general-purpose CPU. Here we have a Chinese peer. They were founded in 2015 and have been investing for the last 12 years. That's kind of the new entrants in the NAND business. That's my first view on this business. I think the predominant view on this business in the past is, "Oh, there's too many players and it's too competitive." If you roll forward to where we are right now, there's really not very many people can do it, and there's not going to be any new ones, because it's extremely difficult.
Even if you basically have incredible access to capital, see it in 10 years if you want to start a business. That's a good place to be. China is clearly a formidable company. YMTC is a formidable company. I think my base case is they're going to continue to grow and consume the China market. I think that would be a natural thing to conclude. I think after that, it becomes a different question about how it's going to impact the rest of the world. It's less a technology question, more of a policy kind of question. But I think the most important thing about this is to understand the rest of the world is growing. I think when you think about competition in any business, and you think about it as a static zero-sum game, somebody wins, somebody loses.
I take share from you, this kind of dynamic. That's not the dynamic we have here. The whole rest of the world is still growing mid to high teens. I think this is the most spectacular thing about the NAND business, why I love this business, because the hardest thing in most technology business is to grow, especially when you get big. Right, Luis? You've run a bunch of technology businesses, right?
Totally. That's the biggest challenge.
You start going down the M&A route. I need to find adjacent TAM. I need to expand my portfolio. It's just like, it's very difficult. This business, you were just telling me maybe I should grow faster, right? I mean, growth is not really the problem we have. When you look at the playing field of all the people in it, I think that's one of the things you have to start from. There's a lot of growth. This is not a zero-sum game. I don't think this is going to play out the way maybe most people think it's going to play out, as if it's this idea that there's going to be one winner at the end. I don't think it's going to work like that at all. It hasn't happened.
Maybe we can close, we just have a couple minutes left, with the JV structure and Kioxia, because I know there's a lot of interest here in Kioxia as well. This JV has worked really well since inception, maybe with the exception of like the couple years before you came in. You very clearly immediately recognized the value of the JV, enhanced those relationships. It's working very smoothly. Can you just talk a little bit about that? I feel like we all think someday that this JV gets put together, but assuming status quo, that it runs as is, just your view of the importance of the JV and how well that structure is working.
When I came into the business, and there had been some-
Friction.
Ripples, friction in the relationship. Anything that goes on for 25+ years, it does that because it's valuable, and both sides understand it. It delivers real value on an ongoing basis, and that is 100% true. It was not hard when I came into the business in 2020. It doesn't take a whole lot of time inside the business to realize this is an incredibly important strategic asset, and it needs to be invested in. Not invested in from a capital point of view, invested in from a relationship point of view, all different kinds of levels. I found a very willing partner. This has gone on for a very long time. It's very productive. It's extremely important. Why is it important? I think most people see it as a manufacturing relationship, and it is a manufacturing relationship, very, very big.
The most important part of it, in my mind, is the R&D collaboration. We both develop the same BiCS roadmap together, and that means, look, in an at-scale technology business, you can afford to invest R&D essentially commensurate with your market share. Together, we have the largest market share. That means we can invest more. When you do that for decades, across 19 generations of technology, you have this sustained accumulated value, and that's how we get our technology is 30% more capital efficient than the industry average, because we've been doing it a long time with a lot more people. That's how we get to things like CBA years ahead of other people. That's anything negative about anybody else in this industry. Anybody in this industry is really good at what they do, by definition.
But it is just the reality, if you can invest more, you should have a better product if you are doing your job well. So, we end up with very good base technology, and then obviously we have all the benefits of economies of scale of manufacturing. And that is what makes it a great relationship, and that is what makes it enduring. And I think both sides, I can only speak for my side, but obviously I am very close to the team, I think both partners see it as a very valuable relationship.
100%. Great. Well, Dave, Luis, thank you so much for your time.
Thank you, Joe.
Thanks, Joe. Appreciate it.
Appreciate it.