Everybody, thank you for joining us on day two of the Mizuho Tech Conference. Joining me today, I'm Vijay Rakesh, Senior Semis A nalyst at Mizuho, and joining me today is David Goeckeler. David is Chairman and CEO of SanDisk Corporation. With over 25 years of transformational leadership across semis, global networking, and enterprise software, David has established himself as one of the industry's most visionary executives. David is also instrumental in driving all the innovation, growing industries ranging from semiconductor to global networking to enterprise software. He's recognized as a transformational leader with an exceptional track record of driving highly profitable core business at scale. SanDisk is a global flash semiconductor storage and advanced memory innovator, has extensive operations throughout the US and Asia with almost 10,000 employees worldwide. The brand is recognized globally. With that, please welcome David Goeckeler. With that, let's get started.
Thank you. That was very generous.
Okay.
It's great to be here.
Well.
Before I start, let me just say that our safe harbor is on our website. Please refer to it.
Yes.
All the statements will be subject to that safe harbor.
Absolutely. Thanks for that, David. I'll start off. At the Analyst Day, when we met, I know one of the things you said is the reason you came to SanDisk is there's a lot of things happening in NAND. It's much more interesting, and you see a lot of changes coming to NAND. I was wondering if we can start right there, maybe give us a big picture overview of what you see as some of the big structural changes happening in the NAND industry. What are the drivers there? How do you see those kind of playing out?
Yeah, I think I'm really optimistic about the NAND industry. First of all, I love the hard drive industry too. Nothing but positive to say about that, but that's not why I'm here. Those guys will do a great job. NAND industry, $65 billion industry going to $100 billion, let's say by the end of the decade. Very big market. That's like 101 kind of stuff. Be in big markets. When you're in big markets, you have big levers. You start moving those levers. You can drive a lot of profitability. Number two, built-in growth driver. We can debate about what the growth rate is. I think in the past, we stuck too long to the 30% growth rate kind of thing. I think the growth rate is really low mid-teens kind of number. That's fine. That's great growth at this size. We got that.
You got big market, built-in growth driver. Check, check. That's all very good. Now, the R&D side of it. Can we supply this market? Let me tell you, Moore's Law is alive and well in the NAND industry. We know how to scale NAND. We have a roadmap for NAND for decades of being able to supply the market. We're not going to run out of R&D. We're not going to run out of innovation. That's a great dynamic of the market. Who are our customers? Our customers are like the most enviable technology companies in the history of the world. They can't run their franchises and build the great products they build without our product. Another great thing. For SanDisk in particular, let's get down to us, we have all of that. We have this global consumer franchise as well.
There's only one company that can sell a SanDisk product. It's SanDisk. That brand is known globally for reliability, performance, really great brand. We have this huge dynamic range as a customer. We can sell to an individual, wherever they happen to be in the world, whether it's an e-tail platform, brick and mortar, they walk in, there are our products on the shelf. We sell to the largest customers in the world, the NAND that's required to build these awesome products that they build. All that is just a fantastic place to spend time. I have a belief that the way the industry is going to be managed and the way a business needs to be managed is fundamentally changing. I think that's happening before our eyes. It doesn't happen in just one quarter. It doesn't even happen in just one year.
That change is going to cause this industry, I think the financial dynamics, to be very, very different. We can dive into that as well.
Got it. Absolutely. I'm sure we'll talk a lot about all that. I just want to take one a little bit deeper dive into what's happening on the AI server side. Obviously, that's a huge market on the data center side, as you mentioned. You're seeing increasing storage demand for LLMs, larger inference workloads. One of the things that you talked about at your Analyst Day was this high bandwidth concept of high bandwidth flash and how that is kind of augmenting the HBM memory roadmap. Maybe you can talk to what's happening there, how you see that roadmap evolving. You talked about massive capacities, like 4,000 GB of high bandwidth flash versus today, we put somewhere like 192 GB of HBM DRAM. Maybe you can talk to how you see that evolving. Obviously, those can support massive LLM models on them.
Let's walk into that a little bit. Let's start at a little higher level. First of all, the AI server market. Just in general, the data center market in NAND. This is one of the things I love about the NAND market. I talked earlier, I got this built-in growth driver. People wake up every day and they have new use cases for NAND. Now we have this big new use case for NAND called AI model training, and then we're going to get to AI inference. That's going to cause, big picture, that's going to cause all the data that's stored to become warmer. As data becomes warmer, it moves into enterprise SSD and it moves on to NAND. That's this huge structural tailwind we have in the business.
If we look at just how NAND is, before we get into high bandwidth flash, if we just look at how NAND is consumed in the data center, enterprise SSD market, very attractive market, it's kind of bifurcated now into two kind of major sub-markets. One is this very high performance enterprise SSDs that are connected to the GPUs that are feeding data in. Those tend to be lower capacity, let's say 16 TB, very high performance. We've launched a new product there over the last year. It's doing very well. You got this very high capacity part of the market, 60 TB going to 128, going to 256, going to 512. That's everybody building these very large AI data lakes to train models. We have a new product coming out in that.
We talked about this in, I think, this very room, except it was bigger during our investor day three months ago, this Stargate product. We've got a new product in that part of the market. We're very optimistic about that. It's BiCS8, which is very high performance due to wafer bonding. We have a 2 Tb die. We have a clean sheet ASIC we're bringing to market. So we got all new product coming there. That makes us very optimistic about our ability to have enterprise SSD to be a bigger share of the market going forward. Okay, we got all that going for us in NAND. Then we looked at the AI architecture on top of that, and we started to look at, is there a way we can use NAND in a different part of the architecture? Because it's kind of profound change.
AI is one of the biggest trends we've seen in a very long time, maybe ever, in the technology space. I like it when people say it's going to be bigger than the internet. I know a little bit about that, having run the franchise for the global internet. And that was a big deal. If it's bigger and better than that, which I think is probably true, then we've really got something on our hands here. We looked at this stack, and the models are getting bigger. They're getting more sophisticated. Where are the bottlenecks in being able to build bigger models, being able to drive more inference? It's this ability to just have enough storage to hold the model. The models are very, very large. That's HBM market today, which is fantastic. It's doing great, has a roadmap.
What we looked at is, is there a way we can use NAND in that market to get better cost, better scaling dynamics, especially, and be able to store more in the same footprint? That is where we came up with this idea of high bandwidth flash. Because, as I said earlier, when I was going through the NAND market, we check all these boxes. We have got the scaling thing down. We can bring more NAND to market. We get more bits per wafer. We have still got this Moore's Law alive and well. We have got BiCS8 coming out, which is more bits than BiCS6, which is more bits than BiCS5. And guess what? We have a BiCS10 and a BiCS11. We have got that scaling dynamic figured out.
How can we apply this to this great AI market, which already has a big tailwind for NAND in a different use case? That is what high bandwidth flash is all about. We are still very optimistic about that. We are talking to a lot of partners, customers about how we could use that. I am not ready to announce the dates for it yet, or I am not ready to announce much more about it. It is in that phase. We are very optimistic that we can apply NAND to a new use case in this market, especially as we get to inference. There is a training side of it where we think we have a play, but especially in inference, you cannot get enough. How are you going to do this on small devices if you do not have a NAND-like device as the models get very, very large? It is not just one model.
You have lots of models that you're running on one device. We think there's a big play for that. We're very optimistic about what we talked about on this very stage three months ago. I think people were maybe a little surprised. I think now others have come out and talked about the same thing a little bit more. I guess what I'll say about this, we're still very optimistic. All the scaling characteristics you talked about and the ability to continue to bring more capacity are still there, very focused on it, on the R&D side of it, and talking to partners. We'll have more to say about it as we move through this audit.
Fantastic. It's interesting you mentioned Stargate because we got Rene, CEO Rene from ARM, presenting right after this. And we had OpenAI yesterday.
I just want to say we picked Stargate like three years ago. Not that we're in a competition. Everybody can use the name, but it was an internal project name. I'm sure by the time it gets to market, it will have a different real name. The internal project name, we had to start ASIC a long time ago. We have been working on that for a while. We are going to stick with the name, though.
Okay, fantastic. Are you seeing interest from the CSPs in this massive terabyte storage to kind of have access, much higher access speeds, I would think, on an SSD?
I think absolutely. I think that the demands on being able to deliver higher capacity, higher performance storage is kind of built into the architecture at this point. How can I get more capacity? This is what's driving QLC. This is what's driving we feel very good about BiCS8 I talked about before. We continue to see a pull on, can we build bigger, faster enterprise SSDs?
Got it. I think what you traditionally see or conventionally see on SSDs is a much higher read speed, but a much slower write speed. Even as you see some of the CSPs like Meta come out and say, hey, we want to put QLC SSDs, the pushback you hear from investors is, or even from the hard disk drive guys, is, hey, they do not have the read speed. The write speed is much slower on SSD, so you will not see that adoption picking up. What is your pushback to that?
My pushback to that is that interface speeds change over time. These kinds of things, in the technology space, when somebody starts saying, oh, it's not fast enough or something like that, it's like, okay, well, that's an argument that's going to go away over time because as technology progresses forward, things get fast. That's not an argument to say something is not going to be adopted. That may say it's not going to be adopted right now. If it's the right architectural fit, if it's the right value proposition, these kinds of things change over time. As I look at our Stargate roadmap, back to that term, and I look at the interface speeds we're going to have on that, you're going to get this equivalency with hard drives on the right side of it as well.
Eventually, you're going to be able to surpass that. You're going to get increased density. Again, it's not a question of—I'm not talking about a conversation around replacement. That's a different issue. What I'm talking about is data gets warmer. Can you put it on these high density enterprise SSDs and get the performance you need on both sides, the write and the read? The answer to that question is definitely yes versus a hard drive. That doesn't mean that SSDs are going to replace cold storage. It means that as data gets warmer, it's going to move on to enterprise SSD and be more of a tailwind for NAND growth in the data center.
Got it. No, that's interesting because I think some of your high-end 128 TB SSDs, they have pretty significant read speed, like 14,500 Mb per second or something, very high read speeds on them.
It gets higher. Not that I'm going to announce a new product here, but if you have a Stargate 1, you're probably going to have a Stargate 2 someday. You're going to have a Stargate 3 someday because you have PCIe Gen 5, then you have Gen 6, and sometime you'll have another one. Everything gets faster over time. That is the story of semiconductors, that I can continue to shrink. It's not like a mechanical device where you have other limitations. Things get faster over time. That opens up more of the market. That is a very good dynamic. Back to your very first question, why am I here spending my own personal capital running a NAND business? It's because I have a very deep belief that these things are going to change and the market's going to continue to grow.
I think that the way you manage the business of just infinitely supplying on the supply side, I can build a new node, therefore I should do it. That is completely changing now before our eyes. That is not the way you manage the business anymore. The business is not just going to be managed in a way that I just am trying to fuel TAM growth all the time by driving costs down. It is that I am going to more precisely match supply and demand to support profitability to be able to continue to grow this market going forward.
Absolutely. The high bandwidth flash is something that almost seems disruptive in the industry.
That is a disruptive topic yet. Yeah, but even if you take that out of it, take that out of it and just say, that is like a great thing. Just the overall rest of the market, the other $65 billion going to $100 billion of the market, that dynamics of how that is managed is changing because the idea in the past was, when I start to get into a down cycle, what I need to do is bring out a new node. Why am I bringing out a new node? It is because it lowers my costs, and that is going to support pricing. There is enough elasticity in the market to absorb the additional demand that the new node provides from the supply side. That calculus has now changed. The last downturn showed us that that calculus of how you manage the business does not really work anymore.
Why has that changed? It's because the ratio of new bits to cost downs is very different than it used to be in the past. You used to get more cost down, fewer bits. Now you get more bits, less cost down. The idea that I just bring out a new node to change the economics of the market doesn't work anymore. You have to now work a little harder, if you will, to get supply matched with demand and use some of the other knobs you have to make sure those are matched up. I think this is what we're seeing in real time. We're seeing nodal transitions slow down. Just because we can do it doesn't mean we should from an economic point of view. That makes complete sense.
You're seeing underutilization when you're still at a profitable level because I want to support pricing. I'm not just going to wait for this low point of the cycle and bring out a new node to save me. That's not going to work anymore. I need to support profitability through underutilization earlier in the process. We're living that right now. I think these dynamics of the way you manage the business, that's what's changing. That's what's going to support better economics that's going to allow us to invest in all of this future growth.
To that point, what I take is, as you look out, what you're seeing is supply slowing down because the cost downs are also slowing down, but that improves the profitability of the industry. There is obviously a drive to increase capacity on some of these storage drives, and that also helps you on the demand side.
Yeah, I think what we're seeing is that demand is demand. We try to estimate demand. We think demand is good. I like to say we're more technology dependent and technology enabled than we've ever been before. Maybe it's better to say it the other way around. We're more technology enabled and technology dependent than ever before. I don't know anybody that's going to give up their device, whatever it happens to be, your PC, your tablet, your phone, all those kinds. Nobody's giving those up. In fact, most people are just consuming more of them. I don't know how many tablets you have. I can say I have more than one.
I stopped counting.
Yeah, I probably do not want to admit how many I have. I do not upgrade it because it does not work. I do not upgrade it because it wore out. I upgrade it because the applications I want to run on it are now so sophisticated that I need better capacity, I need better capability on that device, or I am not as productive anymore, so I buy a new one. AI is just going to accelerate that. If I do not have AI in the enterprise, my enterprise is going to start to atrophy. I cannot have that, so I better enable all of my employees to have an AI PC. I think these dynamics are what is playing out. That is going to drive demand for us. Now what we need to do is get supply matched to that.
I think when we look at the market, we do a lot of modeling of what that demand's going to be. We look at what our technology roadmap is going to be. We look at technology roadmaps across the whole industry, and we then kind of get those two things aligned. That is why you're seeing, again, nodal transitions are the main way you increase that supply. You are seeing those slow down. They are very productive. New nodes, lots of bits. You are seeing them slow down. You are seeing these micro adjustments of underutilization to make sure you keep this supply and demand in balance. I think that is the way to manage the franchise. I think that leads to much better through cycle profitability going forward.
Got it. Very helpful. I just want to hit on one other new product that you guys talked about, which is, I would not say new product, but the 3D Matrix Memory, I guess.
Yes.
It obviously drives significant capacity expansion versus DRAM at much lower cost. It is hitting all the right, checking off all the right boxes, I guess. Can you talk to how you intend to bring that to market? What is the value proposition on 3D Matrix versus what you do today, conventional NAND?
I think the way you started there is the right way to think about it. We have a very, I think at SanDisk, one of the things that we really, one of the really precious assets of the company is our R&D team. We have a really, really good R&D team. There is a lot of seminal NAND IP.
The founders of NAND, so.
Yeah. He actually invented NAND. We worked with him for 23 years. But we have a lot of seminal IP in NAND that other people that want to be in the business license from us. So we have a really, really kind of storied, really good R&D capability. And you're constantly looking at these markets saying, where is there opportunity? The high bandwidth flash, we just talked about that. That is a market where we're able to take the device we have, which is basically a BiCS8 device. We don't need to build a new device. We just need to use it in a different way. That is something that could happen much faster. Again, we'll talk more about that later this summer. The other side of it is, well, we're always thinking about where are the bottlenecks in the market? DRAM is obviously a place.
It's a nice market. It's another $100 billion TAM. The financial dynamics of it are attractive. It doesn't have as much of the scaling properties anymore that NAND has. They're kind of much further along in their roadmap of when did kind of, if you want to call it Moore's law for memory, but when did their scaling dynamics start to slow down? Like a decade ago. Is there a way we can look at building a new device that has the same characteristics of performance, all those other kinds of things, but has different characteristics of scaling so that we can bring the same dynamic that I can bring more scale and keep driving the cost side of it as well? That's what we call kind of Project Neo or 3D Matrix Memory. That's what that is.
That's a project that's been in that R&D side for like seven or eight years. We're now building. We partnered with IMEC at the beginning of 2024. We have a pilot line running in Belgium. We're building wafers every single day. We have engineers in there building wafers. We're iterating on getting yields up, understanding the properties of the device. This thing is pretty far along. We're not ready to build high volume manufacturing just yet, but we're going through that process. It's building a new device, which takes longer and is a more fundamental problem to solve. We're years and years into that. We're optimistic about it. We're not at the finish line yet. If we were at the finish line, we'd be talking about it.
We're far enough along where we feel confident in talking about it that this is a new device we think we can build that can play into that market, especially as that market changes to CXL attached memory devices. You have this whole new architecture emerging in the standard server market and the AI market that allows you to insert a new device into that market in a new way. That's like a pretty significant disruption.
Memory leak, yeah.
Yeah. So it's a new device. It's promising. We're going to continue to work on it. We'll continue to talk about it more as we make progress on it. Again, that's not something we just woke up and thought of a couple of weeks ago. Oh, we have this analyst day. We better come up with something. No, this was something that people have been working on for seven, eight years already. We already have a pilot line that costs hundreds and hundreds of millions of dollars that IMEC was gracious enough to partner with us on to put that in place so that we can work through this. We're optimistic about that technology. It's in the R&D phase.
We thought as we became a public standalone company as SanDisk that it's something we owed it to the people that are going to invest in the company to talk a little bit about. We don't want to talk a lot about it because then we give away maybe all the intellectual. All the intellectual property is tied up, but you don't want to say everything about what that device looks like and what the material science is behind it and all that kind of stuff. We think it's a fair thing that people are at least aware that we're working on it and there's some promise there. You're fishing in deep waters, if you will. You're not looking at playing into some small new market that we have to go create.
This is a huge market that everybody in it would love an alternative technology that has better scaling characteristics.
When do you expect to see some of these technologies hit the market, whether it's flash or 3D?
See, I know you're going to ask me.
We had to go there.
I have more to say about that in the future. I'm not going to put a date on it just yet.
Obviously, you guys are very well represented on the consumer side and the mobile side. I'll make this last question, and we'll take some from the audience. Data center is where you're slightly underrepresented, I would say. I think you guys have mentioned it's almost a 1 zettabyte exabyte market today, and 30% of that is data center. What's the roadmap to kind of get fully represented in that market, I guess?
Yeah. I mean, look, it's a market where we have work to do. We want to get our fair share of that market. We want to be able to mix into that market and be able to go up when the getting is good. Sometimes that market's great. Most of the time it's great. Sometimes it's not so great. What we want is diversity across portfolio. We've got a great consumer franchise. I hope I don't need to say a whole lot about that, but that consumer franchise is on a through cycle basis, more profitable and less cyclical than the rest of the portfolio. That's a good ballast for the portfolio to have. We're going to continue to invest in that brand. We think there's value in the SanDisk brand.
Again, back at our investor day, we just hired a new leader into that business that comes from the consumer packaged goods business. When she joined the company, I remember talking to her, and she says, "Oh," she says, "I don't understand NAND." I'm like, "I got plenty of people that understand NAND." That's not an issue. There's plenty of people at SanDisk that understand NAND. What we need is more consumer-focused. How do you run this franchise? Somebody that's in the CPG business would do it. We are getting that now. I think there's more to come there. Client, we have like 25% share or something like that. Like one of every four devices in the world has our NAND. That's a great spot to be. We have a great gaming portfolio. We just launched a new gaming product that just got wildly positive reviews.
People were actually asking, "How did they even do this? The performance is so good." That is a good harbinger of what is going to come in the rest of the portfolio. Enterprise SSD is the area where we got opportunity. When you run a portfolio, it is good to have opportunity in markets that are high growth, very profitable markets. That is a good place to be as opposed to defending it. We will get into that point where we are defending it. How are we going to go attack that? We are going to build new products. I talked about it before. This market is bifurcating into two markets. One kind of high performance, high bandwidth, high interface speed, closer to the GPU market. That is this product we launched last year. It is going very well. We are in more qualifications.
Every quarter that goes by, there's more people interested in it, so we're ramping that. Remember, enterprise SSDs take a long time to qualify. Especially at very large customers, you're talking like a year, maybe, to get something qualified. We got back to where we started, this whole Stargate program, where we've got BiCS8, 2 Tb die, new ASIC, all that's coming to market in that 128, 256, 512, and even the architectural support up to a PT enterprise SSD. That story is going to start to play out as we move through FY2026. We are super, super excited about it.
Great. I think that brings us to the top of the hour, but we'll take some questions. Are there any questions in the audience? Okay.
All right. Thanks for having us.
Yeah, thank you.
You did a great job.
Appreciate the time.
Fantastic.
Yep.
All right. Thanks, folks.