Thank you very much for joining us today. Before we begin, please note that today's discussion will contain forward-looking statements based on management's current assumptions and expectations, which are subject to various risks and uncertainties. These forward-looking statements include expectations for our technology and product portfolio, our business plans and performance, market trends and opportunities, and our future financial results. We assume no obligation to update these statements. Please refer to the final information statement attached as an exhibit to our registration statement on Form 10 and our other filings with the SEC for more information on the risks and uncertainties that could cause actual results to differ materially from expectations. We will also make references to non-GAAP financial measures today. Reconciliations between the non-GAAP and comparable GAAP financial measures are included in written materials posted in the investor relations section of our website.
Please welcome SVP Global Corporate Marketing at SanDisk, Lynn Cox.
Good morning and welcome. A lot of hard work has gone into getting us to this moment right now. We are so happy to be here, and we are so happy that you're here as well to share in this moment with us. It's been a long journey. We are very excited for the moments getting us here, and we are very excited for our future moment. So thank you for being here. Now, today you're going to hear a lot about innovation, but we're actually going to start off with our corporate brand that we revealed back in December. We're very excited about it. I don't know if you guys noticed it when you came in. Not sure. We wanted to make sure that we put it everywhere so you had a chance to see it actualized at scale. So it's on the stairs. It's on the walls.
It's on some other walls. It's on all the walls, actually. It's also on our social media van touring around the city today. It's vertical. It's horizontal. So I truly hope that you have a chance to experience it and like what you see. The creative work behind this brand, our inspiration for it, was a concept that we call innovation in motion. You see, we believe that the world's innovation engine is constantly in forward motion, and it also stands on the shoulders of relentless forward thinking. So our brand is our nod to a deep understanding that in addition to SanDisk Innovation, there are millions and millions and millions of people in the world: adults, young adults, teens, even kids who every day are pushing their own personal innovation forward and who are actually bringing their aspirations to life with data.
So we're going to continue rolling out our new brand over the next couple of weeks. And today we thought that we would start off by letting you preview one of our new branding reels that we've not shared with the rest of the world yet. So our hope in this campaign is encouraging people to always remember to never give up conviction on your innovation and the change that you can make in the world. So thank you, and let's get started.
Mission: retrieve the package and upload the launch team.
Wherever you are, whatever you do, don't shoot for the moon. You've already been there. If you want to leave your mark, you need to make a new one. So if you want to capture change or plug in the progress, don't just unlock potential. Open new worlds. Store memories and create legacies. Don't just share knowledge. Build understanding. Reach for the sky and make it your runway. That's right. When you think you've gone far enough, push further. The sky is no longer the limit. It's the starting line. So set the bar and then pull yourself above it. Save today by solving for tomorrow. Upload excellence and engineer beyond it. Because greatness isn't just about what you've done. It's about where you're going. Don't stop.
SanDisk. Please welcome incoming Chief Executive Officer at SanDisk, David Goeckler.
All right. Thanks for joining us. It's great to see all of you here. Welcome to everybody on the webcast as well. As Lynn said, it's been a long journey to get here. Last 15 months of preparing to separate Western Digital into two dynamic, incredible companies. I'm also done with the questions of when is it going to happen and is it going to happen. I'm not going to call out any of you guys, but I know who's been asking all those questions, but we've been really dedicated to getting to this point, and as Lynn said, I can tell you everybody that you're going to see up here today is extremely excited to talk to you about what SanDisk is going to be all about. We have maybe some surprises for you, give you all the background you need, and let's get going. All right.
I'm going to start with just a view of the market. Look, there's no doubt that we are in strong and growing markets. So I came to Western Digital five years ago. I told all of you this back in 2022. About 10 years ago, I developed this conviction on the technology space. I've been in the technology world for a long time, maybe longer than I care to admit. And the world over the last 15 years has been building a platform for innovation. And that platform is very powerful cloud connected to ever more powerful devices tied together by high-speed networks. These three things tied together have completely changed the way technology is built and distributed in the world. And what's super interesting about this architecture, to me anyway, is that each part of it evolves independently and makes the whole thing stronger.
Cloud is now, I mean, the breadth of scale of the cloud is incredible. I mean, the most valuable, most accomplished technology companies in the history of the world are all about building the cloud. Obviously, the progress of the device ecosystem has been incredible over the last 15 years. And of course, networks have gotten very, very fast where the experience between these two, no matter where you are in the world, is seamless. And on top of that, architecture has just been wave after wave after wave of innovation. And I have an enormous amount of conviction. I had a lot of conviction five years ago that we didn't know what this was going to happen once this platform was there because all the innovators in the world come and they build their applications on top of it. They disrupt business models. They invent new things.
It was just going to go on for a very long time. And I think that's exactly what's happened. Now, if we look at NAND over this time period, it's been an incredible enabler of this. You go back all the way to 2000 and the move from film to digital, which was really kind of spawned the SanDisk brand, and then everything that happened to enable devices to where we are today with a billion and a half smartphones, 270 million PCs that have all moved from hard drives to SSDs. That transition is behind us. We've had content growth that is driving this market throughout this whole time. Wasn't that long ago, K hurram is going to have it on one of his slides. It was kind of amazing. You look back, it's not that long ago, the industry was shipping one exabyte of flash.
Now this ecosystem is almost 650 exabytes of flash. Around the mid-2014 or so, the cloud really started to get built out, really started to have another level of growth. Of course, an enormous amount of storage in the cloud. New use cases came out in the cloud that because the cloud was now mission-critical, everybody was basing their enterprise on it. The SaaS business model came into existence. All of a sudden, the need for faster storage in the cloud started emerging as well and yet another leg of growth on top of this. So this has been going on for 20, 25 years now. And we're about to enter, we are entering in the last couple of years, maybe the largest disruption of all, which is AI. How is AI going to start to impact all of this? Well, we're already starting to see it.
It's going to sweep across this entire architecture. It starts, of course, in the cloud with training of models, large language models. We've seen the impact over 24, 100% year-over-year increase in the demand for NAND in the cloud. But we're now waiting, quite frankly, about how fast this architecture is going to sweep across the rest of this. It's definitely going to go to the inference phase in the cloud, going to go to inference on the device, and this is going to drive increased growth across this entire architecture. We're all sitting here today. We talk about from quarter to quarter, where are we at in this transition? Maybe we thought the PC and smartphone wave was going to start a quarter or two sooner than we planned for.
In the end, it's not going to matter because this is going to sweep across us in a way that is so fundamental, and it's going to drive every single part of this architecture in a way that I don't even think we completely understand yet. So I think that puts us in an extremely interesting position as SanDisk emerging right as this architecture starts to drive an enormous additional wave of growth across this entire market. We're going to talk later in the day what we think about some of those growth drivers, kind of how we're responding to that, how we're positioning our portfolio across that, and what new opportunities this presents. All right. I've been thinking about this business for a long time. I've been running this business for five years as a franchise. What are its kind of attributes? How do I think about it?
What are the most important strategic elements of this business? And I kind of boiled it down to kind of four key things I'm going to talk about. Everything starts with innovation. All of you that have followed the company for a long time, this is not a secret. I talk about this all the time. If you're going to be a technology company, you better have really good technology. And SanDisk has a long heritage of very deep memory and storage semiconductor expertise. And you're going to see this across this entire presentation today. In fact, we're going to do some stuff at the end that may, one thing I'll give you a little advice. Don't leave early. Don't leave early today. There'll be some stuff throughout this whole thing. You're going to want to see the whole show because we're going to come back.
We're going to start with innovation. We're going to end with innovation. Decades of flash innovation, some of the most foundational IP in the NAND business. The whole MLC innovations, multiple bits per cell, all of these patents are a big part of the company. That R&D team that built all of that is still here. This is the core of this organization, is deep, deep, deep expertise and ability to drive markets in this space. On top of that, there's this extensive systems expertise. Not just the ability to build NAND, the ability to define how it should be built, but then the ability to take that NAND and package it in ways that makes it useful for all of us as consumers.
If you look at the whole consumer portfolio, all the ability to build all the controllers for all of that, everything we've done in the client space is all that systems expertise. And one of the things we're going to talk about quite a bit today is some new architectural work, some new products we've built to apply all that expertise to the enterprise SSD space. Something I know you're all very interested in. We're very interested in it. It's a part of the portfolio we've been working on for quite some time. And I think today you're going to see some really new innovations that we're going to bring to market over the next year or so. All of this adds up to over 11,000 patents, patents pending. Really, if you want to be in this market, you got to have access to this patent portfolio. All right.
Second, scale. We're a semiconductor company. Scale matters. It matters a lot, and we have scale where it matters, which is our joint venture with Kioxia. Joint venture started in 2000. Maybe it's in an extremely healthy position today. We just saw Kioxia go public. Congratulations to them. Ever since I came to this business, our partners talked about being public, and I'm just extremely happy for them that they are, but this joint venture is signed through 2034, so this has been one of the longest joint ventures in technology, maybe one of the most successful, and it's going to go on for quite some time. Why is the scale important? Two things. One, the BiCS innovation, the NAND roadmap, so when you run a technology business, you can essentially invest the number of engineers commensurate with your market share.
That's kind of how it works in a scale business, so if somebody else has twice as much market share as you do, they can usually invest twice as many people. So it's kind of hard to keep up, so with Kioxia, we have the largest market share in the industry, and we've been doing it for a very, very long time. Alper will talk about this later. 17 generations of NAND, so in this business, we can talk about all the products we want. We can talk about consumer products, client products, enterprise SSD. If your fundamental NAND is not great product, you can't make it up in the product you're going to build on top of it. The better foundational technology you have, the better position you're going to be in when you do build those products.
Now, one thing I want you guys to really pay attention to. If I could call out something to pay attention to, this whole idea of multidimensional scaling. Alper, who leads our technology business, is going to talk about this, about how we, the JV, thinks about scaling NAND. We've talked about the layers race endlessly over the last five years that I've been here. I'm sure people talked about it way before then. But I really want you to understand how we think about scaling the fundamental NAND because it's really, really important to understand it. Because it goes to this next thing, which is capital efficiency. If you build the NAND in a certain way, you can be very capital efficient. And that is extremely important in a capital-intensive business.
Alper will show you numbers that we believe over a long period of time, we're probably 30% more efficient in building additional output than the industry average. Now, the other side of it is manufacturing. We have scale manufacturing in Japan. Has anybody ever been to Yokkaichi? If you're in Japan, let me know. We'll get you a tour of Yokkaichi. It's an amazing place. You drive up to it on the highway. It's like up on a hill. There are seven buildings that are just unbelievably huge. I'm sure you can see the whole thing from space. It's one of the largest NAND manufacturing sites in the world, 500,000 wafers a month. It's incredible. So this scale manufacturing, Kioxia is outstanding at driving these facilities forward. We've opened the second one in Kitakami. We have the first fab going. We're building the second fab.
So we really have a good position. Now, one thing that's important to understand, Kioxia runs these fabs, so that's their part of the JV. But SanDisk has 1,000 employees in Japan doing R&D, process manufacturing, all the kinds of things we need to make these things run. And all of that process IP, all of what it takes to actually drive those facilities is jointly owned through the JV. But this joint venture is incredibly important to give us a scale where it matters in the business. All right. The next thing I think about, and I've talked about this a lot, agility. It's not a surprise to any of you. We're in a pretty fast market. Things change pretty quickly in this market. This is not a market where you can check out for a couple of weeks and see what happens.
You got to pay attention all the time. So I'm a big believer, whether it's the company you run or you're a part of or your own personal life, you have got to move faster than the environment you're in. If you don't move faster than the environment you're in, you just get sucked along and you go wherever the environment's going to go. You have no ability to steer, for example. So this ability to move at the speed of the market or move faster than the speed of the market is really important. And we've talked about this a lot. We reorganized the company to do this. We built the company going forward to do this. But there are some things behind the scenes that are important that allow this to happen. First of all, a broad portfolio, increased optionality.
Some quarters, consumer business is the best business in the market. Some quarters, it's the client business. Some quarters, it's gaming. Some quarters, it's mobile. So how do we continue to have the most optionality for where we're going to get our supply to get the best economic return? That's a big part of it. And again, we're going to talk about building out the enterprise SSD pillar of the portfolio, and then we've got a very, very broad portfolio. Another part of it that maybe you don't understand or have seen much, and we don't talk about a lot, is although we do front-end manufacturing with Kioxia, we have our own captive backends. We have one in Malaysia and one in Shanghai. And this is really, really important to have a captive backend where we can dynamically mix what's happening in our portfolio.
Now, if you're in Malaysia, again, let me know. We'll get you a tour. But since I can't take you all to Malaysia, what we're going to do is bring Malaysia here and give you a chance to see what these facilities look like.
I'm not actually a spy. I'm an actor. And this, this isn't a set. This is a factory so advanced, it could be mistaken for science fiction, but they don't make spy craft. This is where one of the world's most advanced, pervasive, and essential technologies comes to life. Smartphones, electric cars, supercomputers, and even the AI systems pushing the boundaries of what's possible are all powered by the game-changing technology made here. This is one of the most sophisticated facilities on the planet. And today, I get to meet with the visionary leading its operations, Dato' Bok, also known as KL. KL, nice to meet you.
Nice to meet you. This place is incredible. I feel like I've walked into a sci-fi movie. There's all these robot arms, massive screen walls, and blinking machines. There's a precision that's in the air with everything moving in perfect sync. Is there any chance you could show me around? Sure. My pleasure. Let's go. This is what we call the Intelligent Command Center. Wow. This is a lot to take in. But starting with the basics, what are you making here in this facility? We are producing 1.8 million single die equivalent components, which we transform into 170,000 drives per day. And this facility runs 24 hours every single day. How do you make that possible? So basically, enabled by automation and analytics frameworks that we have put in place. The automation and analytics basically connect all the equipment and provide production 24 by 7 without interruption.
What do each of these screens correspond to? What am I looking at? What you see here is kind of at a glance the connectivity data of the production floor where you can remotely control or access to the machine in the production floor. Example, if there's any alarm on the machine, the operator will be able to assist remotely from here to resolve the problem. Or at the same time, we enable the artificial intelligence capability that will be able to resolve the problem in the line, and if that is not able, the last result is that you will activate the smartwatch and connect to the technician, and the technician will be able to access inline to resolve the problem. Now, there's robots all over the place. What purpose do the robots serve here?
The robots basically replace the human mundane tasks, able to kind of dispatch from one process to the next process. It's a human and machine collaboration that replace the needs for physical human. It's like the best of both worlds. Yes. They are intelligently integrated to be able to communicate with process, flow, with equipment, and also material dispatching. I've heard you use the term vertically integrated. How does that apply to SanDisk? And what exactly does that mean? In SanDisk, vertically integrated means wafer in, SSD drive, and box out to the customer. So what I mean by that is that we receive the wafer, we transform it into component, assembly, and test. And then from there, we transform it into SSD drive, adds a finished good, and ship to customer. What unique approaches would you say that SanDisk brings into the space?
One of the unique approaches SanDisk brings to the space is what we call lights-out factory. In lights-out factory approach means you can run operation literally with lights out, with minimum human intervention. And with this lights-out factory, of course, it also brings values of sustainability where you don't have to have the lights on to run operation. KL, thank you so much for taking the time to show me around. This has been incredible. Welcome. My pleasure. It's clear after this tour, this isn't just a glimpse into the future. This is the future. A place where human imagination and machine precision unite to bring out the best of us. Where science, technology, and imagination don't just shape the world we live in. They lay the foundation for the world we're building. Progress takes movement, momentum, and the courage to keep pushing forward.
As much as I'd love to stay and soak in the brilliance of this place, the future doesn't wait. Neither do I.
All right. Like I said, if you're in Penang, let me know. These factories have won every single award you can possibly win: World Economic Forum, Lighthouse Awards for Automation, Sustainability, some of the first factories, if not the first factories in all of Asia to win those. KL just does an awesome job. Now, what he said there, 1.8 million single die equivalents per day. We have two of these. Wafers come in on the fourth floor, and on the first floor, somewhere between five to 15 days later, boxes go out to customers. Whole thing integrated in one building. So within five to 15 days cycle time, we can change the mix of our portfolio.
So again, this whole idea of how do we get scale on the front end, agility on the back, put the right management between it with the right portfolio, then we think it's a recipe for great success in this market. All right. The final thing I'm going to touch on is resilience. So this is a cyclical business. We're going to talk about how we're going to do some things to reduce cyclicality. We're going to try and focus on supply discipline a little bit more than maybe we have in the past. But a big part of the resilience of this business, and I think you saw it in the latest downturns, we have this multi-billion dollar branded portfolio, consumer portfolio. This business, there's only one company. I like to say there's only one company that can sell a SanDisk product, SanDisk.
And so we have reach across the whole world, sell hundreds of millions of different products to customers every day, sell anywhere it's legal to sell, every e-tail platform, retail platform. This business is less cyclical. It's better through cycle profitability. And I think we have opportunity to expand this business, think about brands a little bit more. We're going to talk about that again later today. And one thing that kind of opened my eyes a little bit is over the last three or four years, we created this brand around black with gaming. And we put it out there, didn't really put a ton of advertising behind it or anything, but all of a sudden, the black brand is like one of the most profitable parts of the portfolio. And gaming ended up being, depending on the quarter, up to 10% of the mix.
So we think we have latent opportunity here to expand this business. We're going to talk about how we're going to do that today. But this gives us a level of resilience and insulates us in ways that maybe other people in the industry don't have access to. IP licensing revenue, I talked about before that patent portfolio. Probably need access to that patent portfolio if you want to be in this business. IP licensing revenue is not cyclical. The checks show up every quarter no matter what's happening in the cycle. And maybe one of the things you'll see from all the R&D we're doing, we'll try and continue to drive that lever. The final thing I'll talk about is something with this split. This separation is something that allows us to set the balance sheet where we think we need to be for each of these two businesses.
It's a great opportunity for us. The board has thought very, very deeply about this. We've talked about this over the last year and a half. We're going to set these balance sheets, what's appropriate for the assets that we're putting into the market. We believe we have line of sight to a cash positive balance sheet. Luis will talk about that. We set ourselves up to start that way. We want to get to a cash positive position to run this business on an ongoing basis. And Luis will talk about what we're going to do after that. So again, I think we have a level of resilience in this business that really gives us a lot of confidence in what we're going to do over the next several years and going forward. All right. Let me move on. You don't get to create a company very often.
This is kind of a really interesting thing of what you can do. It's a lot of fun, especially when it's scale, like a company that has billions of dollars worth of revenue. So I want to introduce you to the board of this new company. So we really built a board for this industry. So Kimberly and Matt are going to move over and actually sit on both boards of Western Digital. Great continuity of the business. Kimberly is, of course, the audit committee chair and Matt, longtime chair of Western Digital. So they'll be key members of the board. Coming over from Western Digital to the SanDisk board will be Tom Caulfield, who I think all of you know is a CEO and now going to executive chairman of GlobalFoundries. Congratulations, Tom, on that transition. Brings us an enormous amount of semiconductor expertise into the business.
Miyuki Suzuki, who is a well-known Japanese executive that I've worked with in my past. Of course, Japan is a very important market to us for a lot of reasons. To have her expertise and global perspective being based in Tokyo is a huge advantage for the company. Now, to that, we've added four people, deep in expertise they're going to bring to us. Ellyn Shook, who leads leadership in human resources for Accenture. Again, massive scale business, global business perspective. Necip Sayiner, who was longtime CEO. A lot of you probably know him in the semiconductor industry, Silicon Labs, Intersil. Spent some time at Renesas in Japan as well. Very, very deep semiconductor expertise. Rick Cassidy.
I think everybody in the United States that has anything to do with semiconductors knows Rick Cassidy because he ran TSMC Americas for a very, very long time and just has an enormous level of expertise in the semiconductor industry, and if you're talking about producing wafers every single day in this industry between Tom and Rick, I think they have an enormous, enormous amount of expertise in how to do that, and of course, the longtime CFO of AMD, Devinder Kumar, so I think we just have an awesome board. I think we're really looking forward to working with this group of people to drive this business forward. I think this board is built for the industry we're in, and we're just really excited about that. Let me introduce you to the management team. Look, I've thought a lot about this business over the last year.
It's an interesting process when you go through separating the business. And you got to think about this is a new company. This isn't just a continuation of what we were doing before. And you got to think about it as a new company. And I'm a big believer when you put together teams, it's about getting the mix correct. People that know the business and have been around a very, very long time with people that come in from the outside and have new perspectives and new experiences. And how do you put that together in a way that gives you the possibility for kind of non-linear outcomes? So this group is kind of like that. So I'm looking forward, of course, to being the CEO. It's going to be a lot of fun. I've been able to convince a former colleague and good friend of mine, Luis Visoso.
Luis and I spent time at Cisco. He supported me when I ran the portfolio there from a finance perspective, and then he got poached away from me by this guy, Jeff Bezos. I don't know why he ever would leave me for Jeff Bezos, but he came back, but Luis has an unbelievable background for this business. Technology CFO for the last, I don't know, five, eight, 10 years, places like Palo Alto Networks, Unity, Amazon.com, AWS, and Cisco. And before that, he spent 23 years at Procter & Gamble, so I talked about that consumer business, so it's an ability to mix in somebody that's got the finance background, deep technology background, deep consumer background, incredible mix. Christine joined me about four years ago to lead our People Solutions. She's moving into the new company.
Jerry Cagle, longtime Western Digital executive, Chief Revenue Officer, has relationships with about, well, not about every single person that buys NAND in the world. Jerry sees them on a regular basis and will bring a huge amount of stability and continuity there. Bernard Shek, who's in the room today, will be the Chief Legal Officer. Bernard, now, I'm sure when the Form 10 came out, everybody read the Form 10, right? You were all waiting for the Form 10. There's like 1,000 pages of JV agreements in there. This guy understands those agreements better than anybody in the world. He's been part of SanDisk for a very long time, has just been an incredible member of the team, and we're super happy stepping up as Chief Legal Officer for SanDisk.
Don Angspatt, who again joined me about five years ago when I came over, is going to lead operations. Alper, longtime semiconductor memory and storage innovator. Goes all the way back to the SanDisk days, even before that. You're going to hear from him not once, but twice today. So he brings an enormous amount of expertise into this business. I talked about innovation at the core of this business. He's the guy that has all those people that build NAND and all the other semiconductor technologies we're working on. Khurram Ismail, again, somebody that's been in the SanDisk business for decades, has all the systems expertise of how to build all of these products that we talked about come out of KL's factory at a rate of hundreds of thousands per day. So Khurram brings all that expertise. You'll hear from him.
He's got the portfolio strategy and all the engineering. You saw Lynn earlier today. And Janet has joined us just recently as an expert in consumer products. And we'll talk to her later today as well. So we think an awesome team to lead this business, an incredible board. We're very excited about it. Let me tell you what's going to happen the rest of the day. Ivan Donaldson, probably everybody here knows him. He's been in the flash business for 20 years. We're very lucky to attract him to lead investor relations. He's going to talk about the industry. Alper is going to come up and talk about our NAND flash technology leadership. I'm going to reinforce again, pay attention to what he says about multidimensional scaling. I don't know if there'll be a quiz later, but I might ask some of you later.
Khurram's going to talk about the product strategy, product leadership. Again, we're going to talk a lot about enterprise SSDs there and announce some new technology we're bringing to market. Janet's going to join me for a fireside chat on the consumer business. At that point, we'll take a break. Luis will come back and kind of pull it all together into what the model and the value creation opportunity is. And then as a special encore at the end, Alper's going to come back and do something we don't do very often, which is we're going to talk about some of the R&D projects that are in the organization, not quite ready for prime time, but are super promising. All right. Thanks again for being here with us. We look forward to an awesome day. Of course, at the end, we're going to save plenty of time for Q&A.
I'm sure people have lots of questions. You'll have access to all of us for Q&A. But at this point, let me hand it off to Ivan so we can get this moving.
Hey, everyone. It's nice to see quite a few familiar faces out there. So really happy to be here today and talk about SanDisk and talk about the NAND industry, which is near and dear to my heart. As David mentioned, I've been in this business for a little over 20 years. And it's been dynamic, to say the least. I guess you might put it that way. But no, a lot of fun. And I'm just extremely excited to join the team here. I think it's a pivotal point for the industry and for the company. And I think we're really well positioned.
And I'm really excited about the industry too overall, which I'm going to talk about here today. I think you've seen, again, there's been certainly some challenges in this business, which I'll get into some details of the how and the why. But we think we're undergoing a structural change. And fundamentally, a lot of that comes down to how we manage supply to align to demand more closely and really think about generating the value on our IP and technology, which is fundamentally required to enable the innovation of our customers that we're all very familiar with. So what we really need to focus on and what we are focused on is generating higher returns as a business. Less focused on just shipping bits. Less focused on build it and they will come.
More focused on, again, how we generate a sustainable return on investment for us and for our shareholders, so a couple of key points to that. Overall demand, we have a huge market for NAND flash. It's pervasive now. This is no longer the days of we needed to enable NAND to penetrate brand new markets. There's a little bit of that left, but we span multiple segments that are growing in a very healthy manner, which I'll get into some more details, but we've got line of sight to $100 billion TAM, and demand is not the problem. It's a massive market, so growth is there in front of us, but we need to focus on slower node migrations. That's happening already. The need to just add layers for the sake of supply growth is very capital intensive, as most of you are aware.
I'll show some of the data behind that. But what you're seeing is a slower move to just add layers, slower node migrations overall. That means too we're elongating the demand period for our trailing edge NAND production, which, of course, enables us to continue to generate margins and free cash flow on those nodes for longer periods.
Like I mentioned, we're moving away from, again, this strategy of build it and they will come. Let's just build a bunch of incremental wafers with adding layer counts. Part of the other thing too that's happening is we see competition for capital, right? I mean, everyone knows what's happening in DRAM and HBM. Most of those suppliers, all those suppliers are in the NAND business as well.
So you see really a starving of capital to some extent in the NAND business, which, of course, is good for us. So let's dig into demand in a little bit more detail. So as I mentioned before, we see really a diverse demand environment for our products going forward. One of the things that gets talked about a lot is clearly data center, right? And data center is definitely the fastest growing market. But the other markets are growing double digits too if you just look at the data here. In fact, if you strip out data center, we're still in this sort of 17% range on a CAGR basis for demand excluding data center. So clearly we're focused on data center and growth there. We're going to talk about that in detail. But we have this diverse product portfolio that addresses all of these markets over time.
Just a good example, though, specific to data center. I mean, again, you see reports here and there of people thinking, "Well, geez, maybe this is slowing down. What's happening?" We're seeing prices come down short term, which I'll talk about. But just this earnings season, just something like $215 billion being spent by just the three largest hyperscale providers this year on data center CapEx. That's up like 45% year over year. As I said before, we don't have a demand problem in this business. So let's talk about supply. So this is a look at historical CapEx and CapEx to sales for the industry. And you can see the trends. And what this aligns to, this big peak is essentially the 3D transition.
And leading up to and through that 3D transition, that led to this massive CapEx for both greenfield fabs and conversion of 2D to 3D technology. Now, again, it generated a lot of growth, enabled a lot of markets, but ultimately also destroyed a lot of value if you look into and out of cycles. But we think this paradigm is absolutely shifting. So you see this CapEx to sales ratio, which peaked at around in the 50s, is now getting back down into the 30s, which is more sustainable. And we think this is where the industry is going to generally maintain. We think definitely you see an overall lower total rate, total CapEx spend, and a lower rate of CapEx spend too. As I mentioned before, the layer race is over. We just don't see that continuing.
Yes, you're still going to see layers added from a 3D structure perspective. But this need to just go as fast as humanly possible, we see that ending. So now, look, we need to generate the right technology and the right innovation for our customers to deliver their innovation to the market that we all know about. So it's performance, it's product capability, and it's not just, again, bits. On a related note, looking at cost per bit. So this, again, is a business where we talk a lot about our ability to reduce cost per bit. And again, a lot of that was so that we could enable these new markets. Well, this is starting to change, this paradigm again shift, where the 3D transition, again, as I mentioned, generated a lot of just pure bit growth, but the cost per bit reductions have come down.
If you look at the node-to-node annual cost per bit reductions, again, you're going from like the mid-20s to more like the low teens now. And part of what this fundamentally means is we are less focused on just delivering cost per bit reductions that get passed along to the market every quarter, every year. And there's just this expectation that that's the way this business gets run. This rate of cost reductions needs to be leveraged to grow our margins more sustainably, to generate higher margins sustainably into and out of cycles. And that enables us to deliver the innovation, again, that our customers need. NAND provides the systems and the capability that everything, all the mega trends that we talk about related to AI, just to use one example. So we believe we deserve a higher rate of return. And part of that is leveraging our cost reductions.
Okay, so I'll talk now about just more of the short term, what's going on here supply demand-wise. So everyone's seen, obviously, pricing has been pretty challenging in Q4 and into Q1. Some of that's just timing of capacity coming back online after it was reduced previously, some of its inventory movement with customers. But the bottom line is we see this pricing headwind through Q1, as we've talked about recently. However, a couple of things are happening. We really already had a pretty low supply growth year in the NAND industry, fairly benign CapEx. So the setup certainly wasn't bad. It was quite good, quite healthy.
But what we see now because of this price pressure is we're seeing action being taken by us and by the industry to reduce utilization, slow the rate of supply growth, and try to get back supply to match demand more closely as we move through the year. This is really significant. Again, if you follow this industry, I mean, typically utilization cuts happen when you're bleeding cash, when your margins are negative, when you're selling at or close to cash cost, which we're not even close to now. So seeing this happen now, I think, again, is evidence of this structural change that I keep talking about. We see the industry, again, taking these actions to right-size supply to better align to demand. And we think that this is a healthy move for everyone, and really at the end of the day for our customers too.
Now you can see if you look at the range here, the undersupply in the second half of the year really depends on how long these utilization rates cut. But if we just look at our baseline forecast, this assumes by Q2 we are into undersupply. And by Q3, Q4, that undersupply should be at least 5%. 5% is kind of a magic number historically. If you can be in that range, you have some more sustainable price increases that tend to last longer, again, more sustainable. We think the setup is pretty good there. We're hopeful that we get back into balance and move forward from there. In summary here, again, we think the industry is structurally improving. It's a very exciting time to be in the business.
We are focused on maximizing returns for our company and for our shareholders, and I'm just really excited to be part of the journey, so look forward to working with all you guys. Thank you.
Turn it over to Alper.
Good morning and welcome. I'm so excited to be here with you today, and I will talk about some of the work our team does on NAND flash technology. As David mentioned in his opening remarks, at SanDisk, we do have a long legacy of innovation. Since the inception of our joint venture with Kioxia, nearly a quarter century ago, we have delivered 17 generations of industry-leading NAND nodes. Throughout this time, we were able to reduce the bit cost of the technology by more than 1,000x, and through these generations, SanDisk teams have made significant contributions to the storage technology with many industry firsts.
A notable one being the multi-level storage technologies, or MLC technologies, where we store multiple logical bits in every physical device. First with MLC, two bits per cell, and then with three, and finally with QLC, we are storing four bits in every physical device, driving the cost significantly down further. Now, one other thing you're going to notice in this slide is what an incredible strong scaling we are enjoying with NAND technologies today. You may hear other technologies like logic or DRAM really suffering from the end of Moore's Law or Moore's Law being dead already for them, but with NAND, we keep going very strongly. Actually, we have seen Moore's Law slow down for us more than a decade ago. When we were on the 2D NAND technology, the physics of building devices smaller and smaller became very, very difficult.
We got around that problem by inventing 3D NAND and stacking devices on top of each other vertically. As a result of which, today, I can tell you that 3D NAND is the most scalable semiconductor technology in the world. I'm going to make this point over and over to you today, how important scaling is in semiconductors, especially when you're facing data growth that is exponential. Having a highly scalable technology gives you an incredible innovation platform, and not having one exposes you to disruption. Speaking of innovation and scaling, here I'm showing BiCS8, our latest generation of 3D NAND that we introduced last year. When you look at the generation-over-generation improvements that we were able to deliver with BiCS8, you can appreciate what I mean with having a robust scaling path. These are truly incredible numbers that we are very, very proud of.
Behind some of these numbers is a key technology that we call CBA, or CMOS bonded array. I'm going to touch on the CBA technology in more detail in the coming slides. With scaling at the center stage, I want to talk about how we think about scaling in the technology world and how we sort of balance out different vectors in scaling. I'm going to touch on four unique vectors by which we do scaling in technology. As I'm introducing them to you, I'm going to grade them according to a figure of merit that is measuring the cost reduction percentage over bit growth percentage. This is a critical parameter that we measure to make sure that we are not overproducing. In this case, our figure of merit is going to vary between zero and one.
We want it to be as high as possible or as close to one as possible. Okay, let's start. The first vector of scaling is lateral scaling. What I mean by lateral scaling is how many bits can we pack into every millimeter square of every memory layer. As you can see, the figure of merit here is an optimum close to one. The reason for that is squeezing more bits actually does not add significant processing cost. Scaling of cost and bit growth kind of follow each other pretty linearly. Now, as much as this is great, this is also technically probably the most difficult scaling vector. It requires significant innovation and new techniques to be able to squeeze more and more bits together. As I will show you later, we lead the industry in lateral scaling. Next, I'm going to talk about logical scaling.
I kind of touched on that being with MLC, SLC, QLC before. The figure of merit here is 0.76, which is what we got going from 3-bit per cell TLC to QLC generation. Now, this is a pretty high figure of merit, and the key of success in logical scaling is actually being able to deliver the performance and reliability required by demanding storage applications. We do this by carefully optimizing process and device parameters, as well as deploying signal processing and error correction algorithms at a system level. This very organic and intimate relationship between system and device design really makes this possible and allows us to deliver industry-leading solutions, and my friend Khurram is going to talk about some of the details of how we work together to deliver industry-leading QLC solutions shortly. The third vector is vertical scaling. You probably hear a lot about this.
This is adding more layers. I think you heard many times that adding more layers is actually pretty cost-ineffective, and you can see our figure of merit is a very low 0.24, and the reason being is, again, you add layers, you have to buy a lot of expensive equipment, your CapEx goes up, your cost goes up, and you're not getting significant returns out of this vector. Technically speaking, however, adding vertical layer or scaling is not all that difficult. If you're willing to increase your cost, if you're willing to spend money, it's reasonably straightforward to keep adding layers, which is what you see in the industry a lot, and finally, I'll talk about architectural scaling. In architectural scaling, what we mean is how do we lay out our support circuits with respect to the memory arrays? Are they next to the array?
We tuck them underneath, or we implement them on a different wafer altogether and then bond the wafers together. During my conversation about the CBA technology, I will show you how we are leading the industry in architectural scaling. Not only that, but also combining with additional innovations and inventions that we are delivering, we use architectural scaling to further our lateral scaling vector as well. So in quick summary, at SanDisk, we pursue all these four vectors together. But our focus is on the most productive vectors: lateral scaling, logical scaling, and architectural scaling, in all of which we lead the industry. When it comes to vertical scaling and adding layers, of course, we will continue adding layers. But our priority is going to be delivering cost and performance leadership pursuing the other vectors.
Now, this graph, I think, offers a good demonstration of our technology choices or scaling choices with respect to our competitors. Now, what I'm showing you here is on my y-axis, I have a number of layers or vertical scaling, and on my x-axis, I'm showing bit density per layer or my lateral scaling. The contour lines, the dashed lines that run across, are sort of device densities. You achieve a specific device density by adjusting your two vectors, and you can find a combination that delivers a desired result. Now, what this graph shows is that at SanDisk, compared to our competitors, we achieve similar densities by first going in the lateral direction and then adding layers, and this is allowing us to maintain CapEx efficiency and cost leadership. In technology design, what are our guiding principles?
So the question, what features matter, gets different answers in different market segments. In AI, capacity. In compute, performance. At the edge and in the data center, power matters most. Overarching these markets, capital intensity is something we watch very carefully across the board because that drives our capital intensity and hence our cost. So we watch all of these. So in the next few slides, I want to show our competitive positions in these vectors. So I will start off with our capital intensity. I think a very good way to measure your capital intensity as delivered by technology innovation is looking at how much CapEx do you have to spend to deliver a certain increase in your output. And in this case, we have taken the third-party data from TrendForce, looking at over the past. I think this is going back all the way to 2018.
We're showing data on a rolling three-year average. Every year, how much did we have to spend on CapEx versus the industry? And what you're going to see is that our capital efficiency is significantly better consistently compared to the industry average. Another way of stating this data, and if you do the math, you will see that we have been at least 30% more efficient. We spent at least 30% less than the industry average to deliver the same output, which gives an incredible capital advantage as well as cost advantage for our business. Next, I want to talk about capacity. This here is showing our BiCS8 2 Tb TLC die with the CBA technology. It is the world's highest capacity memory die in production. I am so excited about this that I run around with one in my pocket all the time.
So here is 2 trillion bits, 2 trillion bits at my fingertips. This is what we're able to deliver with technology. So very shortly, my friend Khurram is going to come out and show you what his team has been delivering based on this technology. And I think he's going to amaze you. Talking about power and performance, what we did here, we looked at that QLC performance out of BiCS8 and compared it to our competitors in terms of, again, I/O speed, read latency, and write power efficiency, which are the most relevant metrics. So you can see that our BiCS8 QLC die is delivering significant performance and power advantage over competitors' existing product lines, current generation. We went further and said, how is BiCS8 going to behave or compete against next-generation devices from our competitors?
Of course, these devices are not out there, but we looked at their pre-announcements, papers, etc. Here is what seems to be in the pipeline coming over the next few years from our competitors. You can still see that BiCS8 is going to be super competitive against the next-generation devices from our competitors, which gives us a lot of comfort. Now, at the heart of these huge performance improvements that we have and the advantage we are enjoying is a technology we call CBA. It's a wafer bonding technology. With CBA, first introduced at BiCS8, by the way, we build our arrays and the circuits controlling them, the CMOS, in two different wafers. We take these two wafers, and then we flip one to face each other, and then we press them and bond them together and finish them off at the end.
When we are done, that wafer essentially is pretty much indistinguishable from any regular wafer you would have. You couldn't tell here that I'm actually holding two wafers bonded together. So this is what a CBA wafer looks like. Now, CBA gives us multiple advantages, and we truly love this technology and are building on it significantly. So the first advantage is when you look at the array circuits as well as the CMOS circuits, they really prefer to have different process parameters while they're being built, such as the temperatures, for example. If you're building them on a single wafer, you're going to have to find sort of a compromised midpoint to sort of find a satisfaction point. But when you build these wafers separately, you can optimize them separately and get the best of both worlds: higher performing cells and higher performing circuits.
This is how we achieve superior performance. The other advantage of CBA is by building the two wafers separately and in parallel, simultaneously, you're reducing your wafer cycle time in the fab. This gives us significant operational advantages, such as reduced WIP inventory levels. We have also combined CBA with additional innovation to drive further lateral scaling in our chips. And we're actually able to reduce our die size further using these techniques. And finally, CBA is enabling us new flexibilities with our product roadmaps. In traditional NAND roadmaps, we got both performance and power efficiency, as well as more layers and more bits, of course, by spending more CapEx in every generation.
If you identified a new opportunity, a new market segment where you wanted to go with specific new designs, maybe higher performance level, you simply had to wait until the next generation and spend the money to get there. Now, with CBA, what we can do is we can actually pull in the performance and power efficiency of the next generation by just pulling the CMOS wafers in and combining them with the current generation array technology without spending much of a CapEx at all in a very cost-effective way. I can go to new markets. I can complement my product portfolio and chase opportunities. This innovation flexibility is enabled by CBA as well. You're going to see very soon we're talking about a BiCS9 node, which is going to be essentially combining the BiCS8 arrays with our next-generation BiCS CMOS.
So we're going to accelerate the performance transitions very quickly. Okay. Finally, as a competitive advantage, I want to touch on test technology. So you have seen the video that David showed from our Malaysia factories. Indeed, having a vertically integrated backend facility and operations gives us a very intimate relationship between our product design and test and manufacturing. If you remember how our product moves from wafer to components to maybe SSDs at the end, at every stage of this production line, we do extensive amounts of testing for these products. And that can add quite a bit of cost to our products.
Now, through this very organic relationship we have between the two organizations, we were able to really excel at AI/ML improvements in our test methodologies, maximize our bit consumption, and also be able to develop our internal test platforms, all of which help to contain the test costs. So I wanted to show you what, over a 10-year period, our cost contribution to our product cost would have been if we did not do these techniques. Essentially, because our capacities are increasing year over year significantly, even if I account for the performance improvements, our cost contribution from test would have more than tripled over the past 10 years. Through these advanced techniques that we are deploying, we actually reduced the test cost contribution by 30%. So instead of going up 300%, we came down by 30%. So this is another significant competitive advantage we have in our cost structure.
Now, on my final slide, as a technologist, I always want to show something new, and I'd be remiss if I didn't talk about a new technology to you, and I want to give you a sneak preview. This wafer that I was holding a little earlier happens to be our next-generation BiCS technology. We are going to announce it next week at the International Solid State Circuits Conference in San Francisco, and it's going to have over 300 layers, and when you look at the improvements we are expecting out of this technology, you can see that we are taking the incredible performance levels of BiCS8 to even higher levels, so very proud of this and looking forward to talking to you about this technology with even more details after the conference. W ith that, I thank you all very much.
And I would like to introduce my friend Khurram Ismail, who heads our engineering and product development. Thank you.
Thanks, Alper. Good morning. My name is Khurram Ismail. And I'm here to talk about portfolio strategy and product leadership. I've been in the industry for 25 years, all in flash. And hearing Alper talk about the technology transitions that have happened over the last few decades reminded me what an amazing run we've had. Now, I feel like my career has flashed here, as I've been very fortunate to work on a lot of these products as the growth of this sector happened. But I started my career with a 4 Mb flash device. I worked on a 4 Mb flash. And to hear Alper introduce 4 Tb QLC flash is really a testament to the adaptability, scalability, and resiliency of flash technology.
NAND flash continues to serve as a springboard for innovation and interesting use cases that have powered many applications, as can be seen over the last three decades. Started from removable. Really, SanDisk was a pioneer in creating that market that powered the first phase of the growth, and then a lot of questions started happening. Can flash be adopted into mobile and embedded and client applications? I actually remember when we were trying to enable the flash as a primary boot device in mobile, a lot of chipset vendors told us that this will never happen. NAND flash will never replace NOR as a primary boot and storage device in mobile phones, and look what has happened. We have captured that market. Client flash has done quite well, and now to the next phase of the growth, which is the AI, and I'm here to talk about that today.
We'll touch this. Really, the technology relevance continues to power the growth. So as the industry future looks bright, what is the opportunity for SanDisk? We are excited to share with you that opportunity and how we may be able to take advantage of that opportunity. It is our firm belief that we can take advantage, and we will take advantage of this opportunity based on two core strengths. First is our phenomenal and talented engineering product management and system flash expertise. And the second is our engineering platform that is set for the future growth. So I'll quickly touch on those. We have a very diverse, global, and talented engineering team with innovation at the core of what they do. This team has decades' worth of system flash expertise.
This team is purpose-built, placed strategically to optimize our product development close to the talent pools, as well as customer hubs. Now, this team has generated a large set of IPs over the last three decades, which has really served us well in client, consumer, and mobile. And now, for the last three years, this team has been focused on data center. And really, I will be talking about some of these capabilities and products a little bit later. The second strength is our intelligent, scalable, agile architecture. This architecture, this platform, has been in the making for the last few decades. And again, we have done quite well in consumer and client, as David talked about it. We've taken all those learnings.
I mean, with the client, the way I measure the success is being able to transition on the node and on the host interface transitions while maximizing the performance and power. So we've been able to do that with this platform in client and consumer. And over the last three years, we've been customizing it to work in data center. And we'll be introducing some products based on this platform today a little bit later in the presentation. But we're really excited to launch this data center platform in both AI compute and AI storage applications. Before I dive into the products, I wanted to cover how the markets are growing and how do we approach those markets and what is our strategy. First, in client, consumer and client, we have a leadership position there. And we approach it as maintaining and even growing our leadership there.
Now, the client segment really is PC, OEMs, gaming, and as well as channel. So it's a pretty large segment. And within the PC, you have performance swim lane. You have mainstream swim lane, and then the value swim lane. We have done quite well in client with the introduction of many, many successful products over the last several years and have continued to delight our customers with our brand. We can actually grow our market share if we wanted to in client. My fabulous teammate, Janet, will come up a little bit later to talk about the consumer part of the portfolio. And she's here to take our already strong brand to the next level. Mobile is a large market, but we selectively participate in mobile. It's a large bifurcated market, but we focus on a certain portion.
I'll cover that a little bit later on what that selective focus means. Lastly, data center is an exciting segment. It's really divided into two major portions: the compute part, which is compute-intensive, small cap, high performance, and then the storage part, which is the large capacity drive. These two are roughly about half-half when it comes to the market size. There we have a really aggressive approach. We are taking an aggressive approach. A lot of time that we'll be spending today talking about data center and how we are going to tackle this market. We are taking a very aggressive approach. To top all of this, we use a range of our NAND nodal mix to optimize for performance and power, performance and profitability. Let's talk about our sustained leadership in client.
We've been very strong in client over the last several years. I've traveled to a lot of customers, talked to a lot of consumers. We are really proud of what we have built here. Every one in four client devices in the world has our product from our broadest portfolio. It's not really the share and volume that is at play here, but it's really our deep engagements with our customers, our operational excellence, and our time-to-market delivery that is of value here. With these deep relationships with customers comes the trust and insights that continues to guide our roadmap and our platform delivery. We're really, really proud of what we have built in the client. Now, let's look at some of the trends which are powering the technology transitions in client. As you may know, the data demand and performance continue to drive these transitions.
First, in the QLC part, there are two major technology drivers. First is really the productivity, the Windows 11 migration, and the content generation. When you talk about AI inference at the edge, it's really shaping how we work today. 8K video enhancement, text summarization, prioritization, alert generation is really going to change the way you work. And AI at the edge will not only drive the refresh of the client devices, but it'll also drive the increase in the number of client devices. So let's look at some of the technology trends that underpin this. On the QLC, the volume or the capacity is really driving the market, specifically in the value swim lane. And QLC is a value-added drive, which really meets the requirement for BOM and margin for both us and our customers.
The performance and mainstream swim lane is really driven by performance, in this case, PCIe Gen 5. The PCIe Gen 5 is going to drive the transition into enterprise client, OEM, AI, gaming. I'm introducing two platforms today. These platforms will spun out a lot of products out of these platforms. I'm highlighting two such products here. First, our industry-leading QLC platform, which is currently shipping in BiCS6, is really going to transition to BiCS8. The BiCS8 QLC platform is really going to lead in the value segment, which is the fastest-growing swim lane in the client market. Look at the performance numbers. It's double-digit over the currently shipping BiCS6 device. We are going to ship this product in Q2 of this year. The customers are going to love this product. Second is our TLC platform.
On the TLC platform, we service both mainstream and performance. David talked about WD_BLACK. This is the platform, the TLC platform. Really, here, we are making two transitions. Not only are we making the transition from BiCS6 to BiCS8, but we are also transitioning the interface from PCIe Gen 4 to PCIe Gen 5. I would like to draw your attention to the WD product that we have here. What is amazing to see is we are saturating the PCIe Gen 5 line rate, 14.5 at 7 W. 7 W is what today's current shipping Gen 4 drives, PCIe Gen 4 drives, consume. This is absolutely going to be the best-in-class product in the industry. The gamers and the OEMs are going to love this product. This product is going to also start shipping in Q2 of this year.
So let's switch gears and talk about selective focus. As I talked about mobile, we selectively focus in mobile. It's largely bifurcated into two major form factors. First, the MCP, which is a multi-chip package. And the second is discrete. Now, we don't participate in multi-chip package for various reasons. But discrete is the one that interests us because it's a premium segment. And discrete is not only mobile, but it's IoT, automotive, embedded, and everything. And there, we have a very strong portfolio with our UFS and eMMC devices. Now, the work that we do to produce and participate in the mobile market is a function of the platform work that we do in client. We leverage that work. And then we do a light touch lift to make it like a UFS and eMMC device and ship it to the market.
So really, really, the development strategy is tied to our market participation strategy. And lastly, the automotive is one of the most exciting and, in my opinion, one of the highest potential segments from an exabyte point of view. Although today, it starts with a smaller base, but with the advent of autonomous driving, it's just a matter of time before the cars become data center on wheels. There is so much innovation happening in this segment. There is a matter of time before the exabytes will exponentially grow, as I talked about, ADAS, autonomous driving trends. Now, traditionally, we have participated in this segment with our legacy eMMC and UFS devices. With the release of our UFS 3.1 last year and our PCIe Gen 4 NVMe SSD has now given us the complete portfolio in automotive. We can catch the next growth in this sector.
Now, with our first UFS 4.1, first-to-market UFS 4.1, which we'll be releasing this April, will truly put us in the driver's seat of this segment. Pun is intended there. OK, a lot of focus in data center. I would like to take you through what is it that we're doing in data center. We're really excited to share with you some of the work that we have done. Before I do that, last summer, we came out with the AI data cycle and what role the storage plays in that AI data cycle. Here's a representation of an AI data pipeline, really drawn to show how the storage is being used and how the storage may get deployed into the modern infrastructure. Let's unpack this a little bit. First, as the raw data comes in, it gets cleaned, formatted, and labeled.
Now, then this data really gets transferred to the data lake. The clean data gets transferred to the data lake, and then from the data lake, this clean data is loaded into the cluster of GPUs for training. Now, this is done step by step until all the clean data is loaded. Now, once the data gets trained at the inference stage, what happens is that this data gets augmented with domain-specific knowledge. This is to enhance the accuracy of the model. This process is called retrieval-augmented generation or RAG. This RAG-specific data also gets vectorized and loaded into the data lake, and finally, as the inference is happening on the current model, a new model is being trained and new data is being generated. That's the cycle. Now, what role does the storage play in this?
So attached to the cluster of GPUs for checkpointing, training is really our high-speed compute eSSD, what we call compute eSSD for the training purpose. And then for the data lake, the storage device, which is a massive pool of storage, the data lake, we call it high-capacity storage eSSDs. I would say a picture is worth a thousand words. Here is a real-life example of an AI data cluster. Really, this is a 32K GPU cluster, which I would characterize as a mid-size, medium-sized cluster. And there are two storages at play. The first, like I said, the storage that has a one-to-one relationship with GPU and ensures that the GPU is utilized to its max is our compute eSSD. And I want to take a little bit of divergence and talk about this compute eSSD because we have been shipping it.
Last summer, we announced the release of this product. We've had tremendous, tremendous momentum behind this compute eSSD product. This is our leading Gen 5 compute eSSD. It's compliant with all the features. More importantly, this SSD has performed phenomenally well over the benchmark that matters for AI. What we are showing here is the MLPerf benchmark. It's showing the AU in one GPU and two GPU configuration. Really, the AU is the accelerator utilization. That's what MLPerf benchmark is using. As you may all know, the worst thing that you can do is throttle the cycles of GPU. This is the measure of how well the GPU is utilized with your storage device. You can see we did 99.3%, which, when we measured, this was absolutely the best.
To further it, we were the first one to be certified by NVIDIA on the Blackwell RAG Skill System. This is tremendous. I mean, this is a first for us. And this has really helped us build the momentum that I talked about with this product. And further, we have been in production with one major hyperscaler with this product. And we have several qualifications going on with our system partners as we speak. So we have really, really tremendous hopes on this product. And we'll follow up this product with another PCI Gen 6 refresh. So we've really established a strong pipeline in compute. So now, going back to my example, and I really now want to focus on the data lakes, the data lakes, and how the storage plays in those data lakes.
So under this GPU cluster is really a large pool of storage that is connected via a high-bandwidth network. Now, that high-bandwidth network is the data lake. And it has tremendous consumption from a storage point of view. In fact, in this example, about 100 petabytes plus. So the question is, what type of storage device is ideal for this? So let's talk about some of the constraints when it comes to the infrastructure build-out, how our customers see the storage for this large pool of storage data that is data lake. First, the space is really scarce. What we hear from our customers is that they want really, really large capacity drives to build out because the RAG density is of utmost importance for them. That plays into their TCO equation.
Secondly, as I talked about the data lake, the bandwidth on that attached network the data lake is attached to needs to be saturated, so really need large sequential throughput to transfer the data to the cluster of GPUs because you don't want to throttle that, so performance is of utmost importance and last, perhaps more importantly, is the power efficiency because that is a big, big variable in the TCO equation. Now, performance efficiency, be it the performance efficiency in the idle mode for cold data refresh, be it the performance efficiency when you're pushing large sequential throughput per watt, or be it the power efficiency to really granularly define and set the power states of storage device, and really, there, for all these constraints, QLC is a perfect fit.
It checks the boxes on high capacity, checks the boxes on high sequential throughput, and really checks the box on power efficiency. So we'll talk about that now. As we unpack the AI data pipeline and what role storage plays, I'm really excited to announce our latest technology platform, UltraQLC, which is really custom-built based on our decades of experience and our current learnings to really be deployed in the modern data infrastructure while not compromising on density, performance, and power efficiency. It's really built around those three themes. So what is UltraQLC? The full dissertation will take a long time. But in brief, in essence, UltraQLC is a technology platform built on three pillars: NAND technology, BiCS8 NAND, BiCS8 and future NAND, customized controllers, and advanced system design. Alper talked about the scaling, the two terabit NAND die. It's really foundational to our system design.
Our customers desire very large capacity QLC drives. And really, this two terabit die is foundational, which helps them to scale their services while maintaining their physical footprint. This two terabit uses custom SanDisk 3D NAND process, which really produces the c . And this smallest die, then we can package into very small form factors, really optimizing for that RAG density. So NAND is foundational. And we are very happy that BiCS8 NAND is here. And you will see the product announcement that we're going to do. We've also been very focused on data center, like I mentioned. And at the heart of this, it's really our optimized controller. This optimized controller is optimized for QLC data path, is to manage the QLC data path, as well as to run some really custom accelerated services that are required to manage QLC.
I'll talk a little bit about that later. Lastly, we pride ourselves optimizing, working with our NAND, working with our controller. Over the years, we have developed these custom enhancements, algorithms that really help us bring the best out of the QLC. I'll touch on that shortly here as well. First, the customized controller. Underneath the multicore architecture of our QLC controller is really these domain-specific hardware accelerators. They exist to manage the QLC. These accelerators perform hardware functions that otherwise would be performed in firmware. You can imagine firmware is operationally heavy, can affect your latency, performance. These accelerators provide that balance to manage some of the accelerated service that are required for QLC. Second, SSDs are also always challenged with performance because of power scaling. Our controllers have the dynamic frequency scaling feature.
We do this by management of multiple clock sources, which allows us to really dedicate the power allotment to the operations that require peak performance. Then we also have built this controller to scale the capacity. In fact, we have 64 dies per channel on this controller, programmable, which gives us a path to 512 TB SSD, 512 TB SSD with this configuration. It's a good thing, but it's also a bad thing. As you know, that when you load up a lot of things, capacitance increases. You fail to achieve the line rates. To counter that, we have an integrated bus max controller built into our controllers that helps manage that pin capacitance. We can load that many dies. We can fully maximize the channel and get the capacities that we need.
So these are some of the many, many things that we have in our optimized QLC controller. And lastly, on the system optimizations, the advanced system design, this is our secret sauce that works with the technology team, works with the hardware, and brings it all together. Some of the examples on these algorithms is how we write to QLC, how efficiently we write to QLC. QLC takes very long to program. So how do you effectively write? How do you manage the error rate? How do you manage the LDPC engine? Are some of the examples. But I want to draw your attention to one of the examples, one such example. QLC generally suffers from a lot of refreshes. This becomes pronounced when you're talking about 128 TB drives, 256 TB drives. Imagine the amount of time that you spend just refreshing the content.
So how do you tackle that? The way we have come up with the solution on this, working very close with Alper's team, with our system design experts, we have advanced read thresholds that we constantly train as the device age, and the net result of deploying advanced read threshold, maximizing where we read it, optimizing where we read it, when we read it, allows us to do 33% less refreshes on a 128 TB drive, on a 256 TB drive. That is huge, and that's where we say that we maintain the performance because of algorithms like this, so this is the summation of UltraQLC. Again, going back foundational to it is our NAND. On top of that, we have controllers and really optimized system features that give that no compromise-like experience, and we believe that's a game changer for the enterprise.
So I'm happy to share the two products. Not talking theoretically, I'm happy to announce our PCIe Gen 5 storage device. The code name of this was, by the way, Stargate. We came up with it before U.S. government, like three years ago, just so you know. But it's SN670. I'll take the official name. It's really a great product. This is our PCIe Gen 5, and compared to the current shipping Gen 5 drives, you look at the performance numbers; it really, really has significant performance enhancements on the currently shipping Gen 5 storage device in the market. Again, remember, we talked about the performance. We talked about the power efficiency. So you're getting the more performance on the same power envelope; we have better power performance efficiency. So this is going to do great in the marketplace, and we are doing this in every form factor.
You've got the longer E3.L U.2. So this will be in our customer's hand in a couple of months from now. This has been in the making for the last three years. These are being offered in two capacities, 128 with a usable 122 TB and 64 TB. We'll be shipping it in Q3. But like I said, a lot of customers are going to get their hands on this product soon. So where do we go from here? The 128 is really just the start of this journey. As I talked about, as David has talked about, how the AI is going to revolutionize the use for storage, 128 TB is just the beginning of it. We'll do a quick turn. We'll release the 256 TB early next year. We have a lot of our customers that are excited about the 256 TB capacity.
And we've talked to the industry giants if they have confirmed our timing. And then we see the path, like I mentioned, with this current existing platform that we have, a path to 512 TB in calendar year 2027. Now, I just wanted to show the potential of this platform with certainty. We see a path to 1 TB. I'm not putting a date to it. But we certainly see a path to 1 TB . And that will be something when we release 1 TB . Remember when I mentioned to you that I started working on four megabit flash to see a 1 TB being launched will be a momentous occasion for all of us. And we do see a path for that. And we'll get there. So in summary, we have a great strategy on how we tackle the growing segments of the market.
We also have sustained leadership in our client, consumer. We'll continue and maintain and even grow our leadership in those. But really, we are super focused on making data center. And this is just not me talking. You saw our momentum with the compute SSD. We're gaining momentum. We started that journey about a couple of years ago. We are gaining momentum in storage with the release of 128 TB capacity that I just showed you. And fundamentally, what gives me immense belief that we are going to succeed and lead the future is really the mingling of technology, the system know-how, but more importantly, the people who have done it over the decades. And we are going to lead the future with that. Thank you for your time.
All right, I'm back. I'm going to spend a few minutes talking to the newest member of our executive team.
Give us a few minutes here. You're going to enjoy this. And then we're going to give you a break. But I want to invite Janet Allgaier out. Janet, come on out. Welcome, Janet.
Thank you.
So how long have you been at SanDisk, Janet?
Let's see, five weeks.
OK, five weeks. Got it all figured out.
Absolutely.
And did you come here to do NAND design with Alper?
No, I don't think so.
You don't have a 2 TB die in your pocket?
No, I don't. I don't carry it around.
What about ASIC design? You're going to do ASIC design with Khurram? He's a pretty intense guy.
I think I'll wait a few weeks on that one.
You'll leave that to him?
Yeah, I think so.
OK, so what is your expertise? Why are you here?
So my expertise is really in the consumer area.
I spent 25 years at Procter & Gamble. I worked on a lot of iconic brands, from Pampers to Olay to Always, Gillette, Secret, and then my favorite, Old Spice. So I worked on that, and I now get to work on SanDisk.
So when we were talking, when you worked on Old Spice, you came up with this very funny ad about the man on the horse. That was like 10 years ago, right?
Yeah, over 10 years ago, actually, and
That was you. That was your creative, and I actually was watching the Super Bowl. And it was still there.
I know.
It's pretty amazing.
It is.
Pretty incredible.
I don't think the horse is alive, but it was a really great campaign, just one of those great moments.
OK, so why did you come to SanDisk?
Because I think you're amazing.
OK, give that a little time.
We'll see how that gets modified.
No, because I did think you were amazing and also I love the industry, learning about it, don't know a lot about it yet, but will want to learn more, intrigued by it, and then I love this brand. As I mentioned, David, I was probably one of the first consumers that moved from film to digital when I bought my camera, and I loved SanDisk was my first flash product, and it's never let me down, and I always buy SanDisk, so I love the brand, and I love the opportunity.
OK, so you've been here five weeks. You've had a chance to take a look. What's your starting point with the brand? We've got some information behind us here. Revenue share, 73% aided brand awareness. I mean, is that good? What is your starting point?
How do you think about that as an expert in this?
Right. What I've learned over the years is to always start with really the fundamental of the brand equity. And what I mean by that is, what does it stand for? Where did it come from? What's its heritage? So if you look back at all the brands, what is it rooted in? And this brand is a really powerful brand. It's a pioneer. It's been an industry pioneer, I think, of flash and synonymous with SanDisk. And it's a very powerful brand. If you look around out in the lobby, it's right there. It's really strong. Red's a fantastic color. So it's an industry pioneer. It has global leadership presence, 45% global revenue share, category leadership, great relationships with customers. And then you pointed out it has a 73% brand recognition. That's a global number.
And that is very, very strong. And I know you're going to ask me, like, well, where can you take it? And I'd love to see it somewhere in the '80s or in the high '70s. But that is a really strong starting point for the brand.
OK, I talked earlier about this brand. It helps make the company more resilient from a cycle point of view. Quite frankly, inside the company, people talk about the brand in ways that are highly technical. There are things like, for example, we're able to use more of the wafer because we can grade out the wafer. And we can ship more. Khurram has all this expertise in building controllers for the consumer business. So it allows us to do things like ship more than 99% of the wafers end up in a product somewhere.
So it gives us all these great financial and technical dynamics. But you come at it from a different perspective, right? You come at it just like Alper and Khurram have their own process as professionals about how to think about a brand. How do you do it? Where's your starting point? How do you come after this? Where do you think you can take it?
Yeah, so I don't want to be the brand with all the bits, per se. I want to be the brand that the consumer looks at and says, that's what I need. So I really start with putting the consumer at the center and understanding the needs. And I always fall in love with the brand and its consumers, no matter what the brand is. And I think really delving into understanding what do consumers need? What are their articulated needs?
What do they talk about? And then their unarticulated needs and watching them. And then really taking those insights and turning them into action. This brand is a very trusted brand. And I think that's a great starting point for me. So, like I said, I start with the brand equity, always put the consumer in the center. What is the consumer? What's the lingo they use? What are the insights that they're talking about? And what's the online chatter that they're on about? So I start with that. And then I move quickly into what sort of innovation. And you said it well in the early part of the presentation. Innovations are lifeblood. So I look at the innovation. And I was really excited to see the best-in-class innovation.
If you look up on the slides, we play, and this is distinct for SanDisk. We play across all of the forms. We've got the cards, the SSDs. We've got the USBs. So we have a very broad portfolio, and we have a very deep portfolio. We have a lot of price offerings, and we have the breadth. And that makes us an easy choice for our customers and our consumers. So that's one thing, and then I look at the delightful design cues and innovation. What do we have that's unique? In fact, I don't know if we have a lot of gamers in here, but I have three, four huge gamers, and they all have this SanDisk Carabiner PSSD, which we show up here. And it's got this design cue.
You can put this on your backpack or anywhere else you might need to hitch it and always have with you. So I like a lot of these cues, what I call delightful design cues. And then I really like our strong go-to-market partnerships. So that's the third thing I'll look at is the go-to-market. And what are we? How are we talking to consumers? So those are pretty much the three places I start. So you've got a number of levers you can start to look at of how to accelerate, take what already is a great business, see where we can take it next. Yeah, I think we have a lot of latent demand in this brand. And there are a lot of levers. And I'm going to mention a few. But there's more, I'm sure, that I'll discover after my five weeks.
But one of the levers is we have some very tried and true consumers. We've caught the waves with these consumers early. We have the camera enthusiasts that came along with the film to digital conversion. We also have, as you mentioned, David, a very large gaming presence, very big user base there, and I was astounded to find out there's about 3.4 billion gamers, and I was like, can we check that number? Because that's almost like half the globe that's gaming, so that was a big number. We have a very strong presence there, and then we have the creators, and this is the emerging wave of do-it-yourself creators. These are content creators. They're making a life, a career out of doing their own videos, their own photography, and they have increasing demand, so this is about 300 million today, but ever increasing.
So I feel like we have this great user base to start with. And then there are some additional levers on there. So if I think about two levers, the gamer, they're demanding more and more complex, bigger, longer games. And that means bigger games, more data creation, as well as the do-it-yourself content creator. They're getting very sophisticated. And making an hour of film or two weeks of filming themselves, they really need the data storage and creation or the needs that come with data creation. So that's another big lever, the do-it-yourself. And then there's an area that I think is latent. And it's really in the attachment. I remember when we first got introduced and we were talking about attachment. And you're like, Janet, there's only 4%-5% attachment in the global market. What can we do with that?
I'm like, we could do a lot with that. I think the first thing is penetrating why. Why is attachment low? I think there's an education. I think consumers would like to use more of these products. I think they need educating on how to use them. Oh, I get it. I can put it in my mobile. I can download my photos. I think there's a large education with use cases that we could really provide to the consumers and drive some further consideration and trial. I'm really excited about that lever. Then, of course, everyone's been speaking about what's the next wave? And what is that wave with AI? That would be something that I want to get in deeper and understand how are the consumers thinking about that?
What insights are there in AI that could be applicable to the consumer group? So those are a few of the levers that I've seen after five weeks.
OK, Janet. So this is kind of an iconic brand that was created on this film, The Digital Transition. I have enormous respect, having been here for five years, to the people that created this brand. They just did an incredible job. As you said, we have these different waves of people that storage is more and more important. We kind of caught the gaming wave a little bit with our Black brand. Now we're trying to catch the creator wave. And of course, AI is just going to wash over all of this. Yeah, so I think there's an incredible opportunity.
I think it's great to have someone like yourself that is very focused from the consumer angle as part of the company. So what can I get? Can I get a double in this business?
See, this is why I joined. I knew David was going to throw all these challenges at me. Lucky for me, I'm looking at the clock. And it says zero. So we have zero time left.
So stay tuned. I think we're going to spend, I think this is a great opportunity for the company. I think we've really got something here. And we're really happy that you're a part of the company. We're going to take a fresh look at this brand from the eyes of somebody that spent their whole career building consumer franchises.
And thank you. Thank you for the opportunity and for listening.
All right, we're going to take a 15-minute break. When we come back, Luis is going to be up. And he's going to wrap this all up into the model, which I know all of you will be interested in. So we'll see you in 15 minutes.
Thank you.
Please welcome incoming Executive Vice President and Chief Financial Officer at SanDisk, Luis Visoso.
Hey, guys, how are you? Welcome back. I hope that you are as excited as we are about SanDisk's future. I believe that there is a lot of value we can create together. I want to thank you for your support. Thank you for your interest. Thank you for your investment. I look forward to working with you for many years to come. We're here to write the best chapter in SanDisk's history.
What I wanted to do just to get out of the way is some of the key dates. Thursday this week, we'll start with when-issued trading. Our last day as a company, combined company with Western Digital, is February 21st. Our first day on general trading, normal trading, is going to be on the 24th. We will issue our Q2 10-Q on somewhere around mid-March when it's due. With that, I wanted to talk a little bit. I won't read you the disclaimers because we went through that. I want to talk a little bit about what we've done so far. Most of the presentations have focused around creating customer value. That is super important for us, and that's a story of innovation, leading innovation. As we said, that's our lifeblood. In addition to that, we've talked about company agility.
We've talked about an amazing brand that we have, and we've talked about scale where it matters. And I'm going to be talking about that a little bit more during my presentation. But what I want to be focusing on now is how do we create value for shareholders, that other stakeholder that we all care about? And that is a story about generating free cash flow. You're going to be hearing me talk a lot about free cash flow, probably more than we've done in the past. How do we do that? It's a story of revenue growth, margin expansion, and better use of our assets, what I'm calling asset efficiency. So, let's go a little bit through the details. On revenue growth, we play in a market that is very large and growing. The market was about $65 billion by the end of 2024.
We expected to reach $100 billion by the end of the decade, and growth is secular. We have a lot of growth because people are creating more data. We're storing more data every single day, and that is the industry in which we play. In addition to that, we have some tailwinds. AI is a huge tailwind that is impacting data centers, but it's going to flow throughout all our markets. In addition to that, we see more video generation. The amount of data that is required to build a video versus a photo is several-fold, very impressive. Additional more data that needs to be stored and then provides that solution, and obviously, you have other things like autonomous driving that move, as Khurram said or Alper said, we're going to have data centers on the cars going forward, so a lot of growth there. This is a growing industry.
In addition to that, this is a story of margin expansion. As a new smaller company, we're going to be vetting every single one of our costs through the lens of value creation, either creating value for our customers or creating value for our shareholders. We need to raise the bar on cost. This is not a story of margin expansion only, but also an enabler to accelerate innovation, and as I said, we're going to be driving asset efficiency. What I mean with that is particularly CapEx and inventories. Those are the big numbers that you will see on our pages. What we want to do different than the past is we want to be significantly more proactive in adjusting our supply to demand changes. Demand will change. We need to be more proactive.
By doing that, what we aim to do is to improve our gross margins and to narrow the other ranges around that margin that I'll talk about in the next few slides. We need to eliminate the lows that we had in the past. As a non-technical person, let me summarize the strategies that you heard from the businesses. Client is our largest business, right? We have very strong market share. What we want to do there is we want to lead with TLC in performance and mainstream. And we will innovate with QLC in value. Consumer is a market where we have our largest market share. What we want to do there is strengthen those strengths that have worked with us for many years. That is an amazing brand that consumers trust. That is performance that differentiates us in the market.
That is a very diverse go-to-market across the globe. Cloud is a big market, and it's growing. We're behind. We know that. What we will do is we will disrupt with some of the technologies that you just saw, particularly with BiCS in storage with UltraQLC. We will expand our coverage in compute with Generation 5 and Generation 6 products. That is my easy way of summarizing the strategies that you have heard over the last few hours. Let me talk a little bit about the joint venture that we have with Kioxia. I know this is very unique. We touched about it over the last few hours, but I want to just explain it a little bit more. This is the joint venture that started in May of 2000, a joint venture that was formed between SanDisk and Toshiba. A lot of things have changed since then.
The market has totally exploded. Ownership of the companies has changed. We've gone through some ups and downs, maybe more than we would like. The one thing that hasn't changed is the very strong partnership between both companies where we continue to innovate. What is the value of this joint venture? There are two elements: first, supply and innovation. And as you can see on the right side of the table, together we produce one-third of the bits market for the world. That's massive scale, close to the market leader or in line with the market leader and significantly ahead of everybody else. Let's unpack a little bit more of the supply. In supply, you have these fabs, this fab capacity. 80% of that fab capacity is allocated to the joint venture, and 20% goes directly to Kioxia.
Of the joint venture capacity, 50% is allocated to SanDisk, and 50% is allocated to Kioxia. So what does that mean? We get 40% of the output, or we have access to 40% of the output, and Kioxia has access to the other 60%. What this enables us to do, no matter where you look at, we have best-in-class cost per bit, very significant advantage. This also reduces our CapEx, as Alper talked, and it enables more efficient node transitions going forward. You just have more capacity. You have two players. You can make these transitions to new nodes a lot more efficient.
On innovation, I'm not going to talk a lot about it, but what you've seen is over the last 25 years, we created amazing IP, some of the IP that just Alper was talking about, leading innovation that allows us to have a very strong position in the market. We have access to a global talent pool. As David mentioned, we have over 1,000 people, SanDisk employees working in Japan with Kioxia every single day. What do they do? They support our supply and innovation efforts. Now, the financials for the JV are a little bit tricky. Why is that? Because we don't consolidate them. Why don't we consolidate them? Because we don't have full control. We have significant influence, but we don't have control. So we follow the equity method, and that's how we reflect the joint venture in our financials.
This is the last P&L that we published in our 10Q. That's the first quarter of 2025. We deliver $1.9 billion in revenue and a 39% gross margin. So, what is included? How does the JV flow through that? All the costs that we get from Kioxia are reflected in our COGS. That includes the depreciation for the tools that we financed. It includes the depreciation or use of buildings that Kioxia owns. It includes charges for tools that we lease. I think it includes all other labor and materials. But those three things are very important to keep in mind, and I'm going to come back to those later. Our OpEx, which is obviously included in our operating income, includes all the joint efforts that we do with Kioxia. We pool the resources, and those costs are reflected.
Our share of those costs are reflected in our P&L. You will also see our net income as a result of all of that is $172 million. For those of you that care about adjusted EBITDA, we've reported $400 million in adjusted EBITDA, and that's a partial view. Let me explain. Adjusted EBITDA obviously has the depreciation add back of those assets which are in our books. Which assets are in our books? Really, the back-end facilities, the assets that are related to the back-end facilities and anything related to our offices. But that add back does not include the depreciation of the Kioxia tools that we finance. Why? Because they are not in our books. So going forward, what we will be doing is we will be providing you two views on adjusted EBITDA.
The first one, just adding back the depreciation from the tools that are in our books, and the second one, also adding back the depreciation of the tools that we financed and which are sitting in the JV tools. If we were to do that for the first quarter of 2025, that would add another $119 million just for that quarter. So our adjusted EBITDA would have been $519 million. You can then decide which number is more relevant to you, but we'll be providing you both views so you can look at it. We believe that the $519 million is more comparable to what numbers you will be seeing from our peers. Moving on to the balance sheet, this is just a few lines or a few accounts on the balance sheet. There are two accounts that matter related to the joint venture.
First is notes and investments in Flash Ventures. What does that mean? That's all the investments we've made in the joint venture net of dividends. That's how much our investments are worth. Then you will see a line that probably doesn't get a lot of attention, which is other non-current assets. Including in those other non-current assets, there is $700 million or approximately $700 million for depreciation prepayments for the buildings that Kioxia has in the joint venture. As I mentioned, we don't own the buildings. The buildings are owned by Kioxia. What we do with those prepayments, it's a funding mechanism to build those buildings. How do we get that money back? We will get that money back through discounts in our wafer purchases throughout time over the next few years. Moving on to the free cash flow.
When you look at our free cash flow, you will see these purchases or PP&E net. What does that include? That simply includes the PP&E related to our backend and a little bit of spending in our offices. What does not include is anything related to the joint venture. That's a little bit of a complicated story, so I'll explain in the next slide. That's reflected in the $48 million in activities related to the joint venture. Before explaining that further, I want to just clarify that the negative cash flow of $150 million for that quarter includes about $100 million related to the separation, particularly payments back to our parent, Western Digital, for some notes and interest, which will not be repeated after separation. So let's talk about that CapEx. So I would start with what we call gross CapEx.
That CapEx is our share of the investments in tools in the fabs. That number has been fluctuating over the last three years, and it fluctuates for basically two reasons. One is node transitions, and second, as we try to match supply and demand. On average, over the last three years, that number has been 16%. Keep that number in mind. Now, how do we fund that amount of CapEx? We fund it through a combination of external and internal sources. The external sources, which you see there, are basically government subsidies in Japan and the JV's leasing program. If you add it all up and compare it to the gross CapEx, external sources fund 43% of our gross CapEx, at least over the last three years, 43%. And then you go to internal sources. How do we fund that from an internal basis?
First, through the depreciation charges that we're getting through our wafers, because depreciation is not a cash item, but we're paying cash. So that amount of money that we're paying is taking into account as an internal sourcing. If you add it all up, all that depreciation, and you compare it to the gross CapEx, that represents 62% of the gross CapEx. So if you add 43% and 62%, we've paid 105% or we funded 105% of the gross CapEx. So what happens is we get that money back. We get the excess of the gross CapEx minus external sources minus depreciation. We're getting that money back. And that's what you would be seeing an add back in our cash flow statement. Over the last three years, we've gotten $169 million back from the JV as we funded through depreciation and external sources. Hope that's clear.
If not, happy to answer any questions at any point in time. I'm not going to spend a lot of time on our financials over the last three years. You've seen them in our Form 10. I think we've given you lots of explanations of what has happened. Just want to make a few points. One is 2022 is kind of a good year. Keep in mind that $10 billion revenue, because we are going to come back to that. You also see how we perform over the last few years. What you saw is obviously data center was very low in 2023, which then came back up. You see consumer strong and weakened. You see kind of some inventory moves in clients and a very significant ASP volatility.
This is why it's so important for us to manage supply and demand much better so that that ASP remains at a good level. You also see that we included our first half results in line with what we reported as a Western Digital company a few, whatever, 10 days ago, two weeks ago. You see that in the first half, we're seeing an improvement relative to where we were in 2024, $3.8 billion and a gross margin of 36%. Obviously, the ASP has an impact on net income and free cash flow. As I said, just in the interest of time, I'm not going to spend a lot of time on that. We can answer questions later if you're interested. Let's start to talk about the future. Where do we start today? Today, on whatever, February 24th, we'll be well capitalized.
We're going to have cash of $1.3 billion, and we're going to have a debt of $1.9 billion, net debt of $604 million. That's a good place to be. We also have an undrawn revolver, which I hope we stay undrawn just for a rainy day, just in case. What is our capital allocation? What do we want to do? Where do we want to be over the next few quarters? First, we want and believe that it's very important to achieve a net cash position, and we have a line of sight of achieving that over the next four quarters. Now, it obviously depends on market conditions, but that's clearly a priority for us. We need to get there. That's important in a business that has significant volatility, but then what happens?
Then we're going to continue to invest in the business because this is a business that is growing. There is a lot of growth in the business. We want to capture that, as we've explained. But we also want to reduce our gross debt, and we want to return some cash to investors. So those are the priorities going forward. First, get to net cash positive, then a combination between reducing gross debt, continue to invest in the business, and getting to eliminating that debt altogether and returning cash to shareholders, sorry. So what do we expect the market to do going forward? You see, this is an interesting chart because we went all the way back to 2023. You see the black line kind of what happened with data center from down to exploding driven by AI and what happens with consumers and PCs somewhere in the middle.
The important point here is going forward, we expect all segments to be growing at about the same rate. Now, obviously, this is a year-over-year chart. So when you have this huge hump in data center, we are expecting that to grow from that higher base. But we do expect growth across the markets. So what's our model? Before going into the model, let me clarify two things. What we mean with a through cycle model is the financial performance we expect to deliver as the market goes up and down, on average, what performance you can expect from us. So it's not going to be a one-year specific, but it's probably the average of, let's say, three to four years type of performance. The last thing I'll say before going through our numbers is what are we assuming the market will do?
We're assuming the market will grow mid- to high teens, consistent with what I just showed you. It's going to be similar across segments, but we expect good growth, and we expect cost reductions as an industry to be somewhere in the low teens. As we shared before, this is a little bit less than what has been delivered earlier over the prior years, so we do expect cost savings to continue, but at a lower rate than before. Now let's look at our model, which is the column in the middle. We expect our through-cycle performance bits to grow in line with the market, so over time, what you would expect is that we will be back to our historical market share. That's where we expect to be, grow in line with the market and therefore deliver our historical market share.
On gross margin, we expect to be somewhere in the 35% range. Importantly, we want to avoid those lows of the past, as I mentioned. And the way to do it is by being significantly more proactive in managing supply and demand, adjusting supply to that demand. We aim to deliver 20% operating margin, which by definition implies we will have to reduce OpEx to 15% of revenue. We know we're not there yet, but we haven't started optimizing our spending from where based on what I said. We're using a new filter of value creation, and we're going to make sure we get there over time. It's going to take a little bit of time, but we'll get there. CapEx should be somewhere in the mid-teens, consistent with where we've been over the last few years. Alper talked about our advantage in CapEx relative to competition.
How do we deliver that? I think Alper talked a little bit about that. One is our scale, second is how we think about multidimensional scaling, and third, we reuse probably a little bit over 90% of the tools used from one node to another, and if you put that all together, that gives us a very interesting advantage on CapEx spending, but as I said, it doesn't end there. Really, our success is delivering cash, so we're focusing on delivering free cash flow in the low teens. How does this model work? How do we convert bits to revenue, revenue to net income, and net income to cash? That is the model. What does success look like? Remember that $10 billion that I talked about? We think we can get back to that $10 billion. When exactly is that going to happen?
I think it may take a little bit of time depending on market conditions, but we do think that we can get there over the next few years. Once we get there, we will be delivering 20% or we expect to be delivering 20% operating margin. And we should be able to deliver at least $1.2 billion in free cash flow. Once we get there, we should be able to be at a net cash positive position, and we should have reduced our gross debt at least by half. This is what we're working to deliver. Now, looking at this quarter, Q3 is a tough quarter. This is consistent. Some of these numbers are consistent with what we shared with you earlier. It's a tough quarter as we continue to see ASP contraction and as we continue to see inventory reductions, particularly on the client segment.
We expect revenue to be somewhere between $1.55 billion-$1.65 billion. We expect gross margins to be somewhere between 21.5%-23%. As I said, ASP is contracting quarter over quarter. Costs, consistent with what's talked in the earnings call, are expected to be up sequentially, and the increases are driven by normal fluctuations that we have, plus startup cost and our underutilization charges. We expect OpEx between $395 million-$405 million. This includes all these synergies, and it does not include all our efforts of bringing cost under control. Other income and expense, or interest and other expense, somewhere between $25 million-$30 million. Tax rates, somewhere between 21%-23%, which leads to an EPS of somewhere between -$0.30 and -$0.45 for this quarter. From this base, we want and we believe we can improve going forward.
So as you build your models, we thought it would be important to give you our perspective for the calendar 2025. So for the June quarter, what we expect is to start to see bits growth again, somewhere in the mid-single digits. So we expect bits to grow as data center continues to improve, as we start to see some of the PC refresh cycle, and the consumer business continues to improve. Remember, the January-March quarter has a low consumer business, so it should sequentially just increase. We do expect ASP to start turning around in the next quarter. Why do we do that? Why do we expect that to happen? We're starting to see all these supply interventions. Demand is starting to increase, and therefore we expect supply balance will be more balanced. Costs will be relatively flat quarter over quarter as we continue to incur underutilization charges.
Things start to turn very interesting in the back half of the year. We expect significant bits growth closer to our historical levels, closer to the model that I'm talking about. This is not something out of the ordinary. It's getting back to where we've been and to what the projections that we have going forward. We do expect ASP to be significantly positive. Why? Back to the supply-demand equation. Remember, it takes about 120 days to produce a wafer. So from the moment we decide to reduce supply until it actually impacts the market, there is 100 plus days. So it just takes time for that to have a significant impact, and as of the back half of the year, we do expect our costs to be improving in line with our models, so very excited about the year.
So to close, we're very excited about the future for SanDisk. We believe we can create significant value for our customers. We told you how we want to do that. And we believe this is a shareholder creation story here. It's about generating that free cash flow, converting revenue to net income to free cash flow. We believe we can do that, and we appreciate your support. So with that, let me turn it over to David.
All right. Thank you, Luis. We got one more session for you. And this is something we don't get to do very often, which is talk about some future innovation. I talked about earlier, this kind of presentation started with innovation. It's going to end with innovation. I feel very lucky to be here. I've lived kind of a charmed life for my career.
I started at a place called Bell Labs, which was kind of an innovation factory. Before I came to Western Digital, I led a huge team that basically invented something you may have heard of called the internet, and then I came to Western Digital, and it was like a whole another group of innovators, like material science, physics, just incredible level of depth of innovation that's going on. A lot of it is going to lead to stuff you're going to hear about tomorrow in the HDD business and kind of some of the unbelievably smart decisions that have been made in that franchise that's led it to where it is now. I have enormous confidence in that business and the leadership team, and I think you're going to have a really good day tomorrow hearing about that, and also now with SanDisk, all this semiconductor expertise.
We talked about this kind of fundamental, this team that delivered fundamental innovation in the NAND business and kind of putting those together and watching what happens over the last several years. So what we're going to talk about next, I'm going to have Alper come back, and we're going to talk about some projects, two projects that have been kind of percolating within the company for a while. They're not ready for prime time just yet. They're not in the model you just saw. That's why we put them here. But we thought it was important to talk about them because we think they're very, very important in the world we all live in today. I talked earlier about AI.
One of the things I think you can take away in the whole AI landscape, just within what's happened in the last two or three, four weeks, is the amount of innovation that's going on in this space is astronomical. The architecture is changing before our eyes all the time. It is really, really a fundamental shift in the industry. So I'm going to have Alper come back up. He's going to talk about a couple of projects, one that's been kind of in the company for quite a number of years and kind of the material science area that has now worked its way along to being a little more real, and one that's just kind of emerged, let's call it, in the last 12 to 18 months. I think you're going to be interested to hear about these. So Alper, take it away.
Thank you, David. Good morning once again. And I know it's been a long morning. Thank you so much for sticking around. This is going to be the fun part. I get the honor of talking about the amazing work that our researchers and engineers do. And as David said, I'm going to talk about two projects specifically. And they have been going on in the company for quite a while. Although we've been talking about these projects with our customers and partners and getting a lot of feedback, this is the very first time we're publicly disclosing these projects. So it's a special day for all of us technologists. Both of these projects actually tackle a similar problem statement, which is related to the end of Moore's Law for DRAM. Now, DRAM technology was invented in 1967 and first productized in 1970.
When you look at its history, the next 40 years were pretty good for DRAM. It sort of rode the wave of Moore's Law and scaled really very well. But when you look back the 10-15 years or so, scaling has become extraordinarily challenging for DRAM. And today, it's scaling at about 5% or less every year. This is happening right at the time where the need for memory bandwidth and capacity are growing exponentially, especially exacerbated by some of these large language models that are just dominating the AI world these days, which are growing possibly at 10x every year or two. And this discrepancy, which is growing exponentially between what the markets need, what the applications need, versus what the technology is delivering, we call that the memory wall. And this memory wall is driving significant cost increases for compute and AI.
We've all heard about server costs or server farms being dominated by DRAM costs. And here I found this interesting research by TechInsights that they just published showing the cost ingredients of a GPU, a modern GPU today, expected to essentially be dominated by HBM, the DRAM memory. 68% of a GPU is going to come from DRAM ingredients. This is obviously becoming a major issue. Now, how do we solve this memory wall problem? There are multiple approaches one could take. The first one is you can try the brute force way of throwing money at the problem, which is more or less what we've been doing in the industry. A second approach could be to explore a DRAM technology that actually scales, something like a 3D DRAM, the way we have done it in the NAND.
This work has been going on for quite some time, but the technological challenges have been pretty daunting, and there is no clear line of sight to getting to 3D DRAM right now in the industry, or a third approach could be exploring new scalable memory technologies. This is what we have done, so here I'm introducing the SanDisk 3D Matrix Memory. We, as the inventors of 3D cross-point memory, are very proud to show you a new innovative scalable 3D architecture targeting to deliver DRAM-like performance, 4x the capacity at half the bit cost. At the core of this technology, we have a novel memory cell that allows us to scale into dense array architectures, and at a product level, we intend to attach this memory to industry-standard attach points like CXL buses, and doing so, we will use our systems and media management expertise.
Of course, most importantly, we project this technology to be highly scalable, such that the cost advantage over DRAM will increase over time. Now, this project has been started in 2017 within our research organization using our 150 mm tool set. Over the next few years, we matured the technology using increasingly more advanced test vehicles. In around 2023, we came to a point where we believed that we were ready to push it into the productization phase. Now, one challenge we have in the memory industry developing these kinds of new technologies requiring new material sets is how expensive and how capital-intensive it is to undertake such a research program. We call this the lab-to-fab gap. To bridge that gap, starting 2024, we have engaged IMEC. Since the beginning of the year, we're working with IMEC in their 300 millimeter cutting-edge facilities in Leuven.
If you're not quite familiar with IMEC, IMEC is a premier leading industry research organization. They are based in Belgium, and they are sort of the Switzerland of semiconductor industry. They bring together all the leading semiconductor companies and conduct pre-competitive research to drive industry alignment for future roadmaps. We have been a member of some of the research programs at IMEC for over 10 years and really, really excited about this new collaboration we started with them. Max Mirgoli, the Executive Vice President of Worldwide Strategic Partnerships at IMEC, is involved in a multitude of international semiconductor projects and I believe to be one of the most informed executives in the industry today. Just before this event, I was able to sit down with Max and get his insights in the collaboration and in the technology.
In a video, I would like to share that conversation with you.
Max, it's wonderful to have you here with us today. Thank you so much for making time from your very busy schedule. You had unique insights into the development of the SanDisk 3D Matrix Memory technology. In this AI-driven era of memory and compute, I'm really excited for you to share your insights on the industry and technology development. To start off with, if you could please talk a little bit about the unique structures of our collaboration between SanDisk and IMEC and the project that we call together, Project NEAR.
First of all, thank you very much for giving me the opportunity to be with you. It is wonderful to be here. Together, we started quickly crafting the project, and we identified a partnership that was based on four pillars. It was based on access. It was based on expertise. It was based on IP. And it was based on portability. When it comes to access, IMEC has a state-of-the-art 300-millimeter tools that could be used for this. And it was identified by your engineering team. The tool set that was required is available at IMEC. When it came to expertise, the two teams very quickly identified each other's strengths and were able to create the project in a way that leveraged each group's expertise to ensure the success of the project.
On IP, we crafted an IP structure in a way that SanDisk maintained ownership of the core IP that was developed in this project and when it came to portability, the two teams ensured that any process that is developed is easily transferable seamlessly to a production facility.
Max, we talked about access as a main pillar of our project together. Can you please talk a little bit more about the infrastructure investments at IMEC's Leuven facility, as well as what stood out for you with regards to Project NEAR?
One of the unique assets of IMEC that we have been able to put together over the last 40 years has been a state-of-the-art fab with over $4.5 billion worth of investment in advanced tools. Between IMEC and the technology development team of SanDisk, we identified over $700 million worth of tools that could be used for this project for the development. Further, SanDisk realized there are certain tools that we don't have, and they can bring in, and we accommodated the space for the tools that will be brought in by SanDisk. One of the areas that was identified as a key differentiator was the access to the EUV lithography tools, whereas IMEC, over the last 40 years, we have had a very close partnership with ASML, and we have the most advanced suite of lithography tools.
Having access to the EUV lithography tool for Project NEAR was able to give SanDisk an ability to accelerate the development and prove the concept in a much faster way.
Max, if we may, let's zoom out to a macro level. From your perspective, what key factors are at play in the global memory industry today?
As we all have witnessed, there are huge investments on AI infrastructure these days. Obviously, mostly they are focused on the compute side versus the memory. As in the memory side, large investments are required for very small gains technologically. Project NEAR is a perfect solution to this challenge.
Actually, we fully agree with you, and that's why we are so excited about the SanDisk 3D Matrix Memory technology. It has DRAM-like capabilities at a much lower cost point, as well as a robust scaling path.
Absolutely. In order to develop a technology that is useful and creates good products, you need to have a good coupling to the market and to the customers. SanDisk has always had a rich history of being able to develop technologies that create good products that can address the market demands. If I see the commitment of SanDisk that you have been investing in Project NEAR since 2017, and only over the course of last year, we have accelerated this development, I am confident that SanDisk has the right technology that develops the right products that address the needs of the market. And SanDisk has the scale, the expertise, and the commitment to make NEAR a success.
I want to thank Max and the entire IMEC team one more time for their partnership and the great collaboration we have with them. If you notice that wafer that was sitting between Max and I, actually, that's the wafer we run every single day at IMEC these days with 3D Matrix memory. Obviously, we are very excited about this technology. Our ambitions actually run beyond this world. I mean that literally. Last year, we actually received a grant from the U.S. Department of Defense to build the next generation strategic radiation-hard memory for aerospace and defense applications. This technology is leveraging the 3D Matrix memory architecture and materials that we just talked about. We project that with this new device, we are going to deliver 16x advantage over existing solutions that are going to enable totally new capabilities in aerospace and defense applications.
This award was granted to us within a pool of 10 applications. And obviously, we're very honored, but also super happy to get that kind of validation from the U.S. Department of Defense on this technology. Now, this is my last one more thing moment. I talked to you about these three ways of achieving a solution for the memory wall problem, but there is actually a fourth solution I want to talk about. And I'm going to call that reimagining NAND. Now, today, we talked about scaling over and over again. And by now, I think you know NAND is the most scalable technology. And the way we use that power of scaling has been mostly about optimizing cost and density in the technology. So we build these massive memory arrays and access devices in those arrays in the NAND device.
So we challenged our engineers and said, what else could you do with this power of scaling? What else can we possibly optimize for? The answer they came up with, and I'm somewhat simplifying it here, was actually moving to an architecture where we divide up this massive array into many, many arrays and access each of these arrays in parallel. When you do that, you get massive amounts of bandwidth, a bandwidth optimization. Now, what can we build with this? So we're going to build high bandwidth flash. We're calling it the HBF technology to augment HBM memory for AI inference workloads. What does this device do? We are going to match the bandwidth of HBM memory while delivering 8-16x capacity at a similar cost point. This technology is going to be enabled by our BiCS NAND technology with the CBA.
And CBA is a key enabler here on multiple vectors. And while doing so, we also developed a proprietary stacking technology that enabled us to deliver ultra-low warpage die so that we can stack them 16 high, as in this picture that you're going to see, hopefully. Now, there we go. In this picture, 16 of them stacked on top of each other. And we developed the technology over the past year or so with input from major AI players in the industry. So next, let's talk about what a device using HBF could look like. I'm showing here a hypothetical GPU using HBM memory. And as an example, this looks fairly similar to one that is a highly anticipated leading-edge GPU that is going to come out. And it's going to have 192 GB of memory.
When we started the project, we envisioned that we were going to replace a few of these HBM die with HBF to get significantly higher capacity on the GPU. The reason we thought we still need the HBM is because, although inference workloads are heavily read-intensive, you have data structures such as KV caches that need frequent updates, and we thought for those, you might require HBM die. However, as we continued our work with our customers and did a lot of workload simulations, a different view started emerging. As you all know, AI is an incredibly rapidly moving space. A lot of innovation is constantly coming, and if you are paying attention to some of the announcements from DeepSeek a couple of weeks ago, one of the technologies that they talked about is these multi-headed latent attentions. What multi-headed latent attentions do is significantly reduce KV cache sizes.
As a matter of fact, they are reporting that they are able to reduce KV caches by 93%. So any innovation like that actually now would enable us to pack 4 TB of memory on a single GPU. So what can you do with 4 TB? Think about a frontier large language model, let's say something like GPT-4. GPT-4 has 1.8 trillion parameters with 16-bit w eights. And that model alone would take 3.6 TB or 3,600 GB of memory space. I can put that entire model on a single GPU now. I don't need to shuffle around data any longer. So we think that we're going to enable with HBF the next wave of innovation in AI, such as the next generation large language models that are going to be significantly larger. Or think about a mixture of experts models that require a significant number of experts for different skills.
Or think about the next generation of multimodal models that are bringing together the world of text, audio, and video together. All of them are going to be memory-bound applications, and we have a solution for them. Let's talk also about AI at the edge. How is your experience with your AI with your cell phones or mobile phones? Mine is pretty nonexistent because actually, it takes incredible resources to run AI or reasonable AI in a mobile phone. Now, just throwing money doesn't quite solve the problem. You just don't have the real estate. You don't have the power envelope. You don't have the compute power. You don't have the memory. None of that exists in an edge device. People have been trying to solve this problem by reducing the models significantly to a few billion parameter models.
But the performance of those models has been really frustrating for the users. They just don't deliver the experience. You really want to use a significantly larger model size. And I'm going to propose here a hypothetical model of 64 billion parameters. Although this is hypothetical, Llama 3.1 or the DeepSeek 70B are very similar models to this in size and capability. And they have been very satisfactory for people who are using them. But still, you would need 64 GB of memory just to put this in. So what can I do with HBF? On a single die, I'm not talking about a bunch of stack. A single die, I could fit an entire model.
Now, I'm going to talk, in addition to this, about another innovation that we have been quite aware of and working on, but that was also popularized recently with the DeepSeek announcement called mixture of experts. Mixture of expert models divide up a very large model into multiple smaller models. And for each token generation, you actually activate only a few of those by which you are actually reducing the need for large compute. The problem becomes more memory capacity bound, but no longer compute bound. We're changing the equation on its head and getting something very interesting here. So if you were to put this mixture of models, a mixture of experts into an HBF and ask a question that I think you're all going to ask very soon, it will very smartly give the right answer.
And we are truly excited about the potential of making the impossible possible and truly enable amazing AI experiences at the edge. David, at the very beginning, talked about how transformational running AI and inference at the edge is going to be. And this technology and innovations that we talked about are exactly what's going to enable that. Now, I'm sorry, my clicker is not quite working. There we go. This is the roadmap for HBF. I've talked about leveraging our existing NAND roadmap, the super scalable roadmap, and indeed, in capacity, bandwidth, and energy efficiency, we are projecting a path forward that's highly scalable and continues in line with the advancements in the space. And finally, about ecosystem. Now, HBF is not drop-in compatible with HBM.
We have tried to make it as close as possible mechanically and electrically to the HBM, but there are going to be minor protocol changes required that need to be enabled at the host devices. We are intending to drive an open standards ecosystem for HBF. And to guide us through that journey, we are right now forming a technical advisory board consisting of industry luminaries and partners. We are bringing back SanDisk. We are incredibly excited about that. And we can't wait to disrupt the memory industry with our spirit of innovation. Thank you all very much.
All right. Thank you, Alper. Thanks to all of you again for joining us, all of you on the webcast for joining us. I'm sure after that, nobody has any questions.
So now what we're going to do is we're going to bring the rest of the group back out, and we'll turn it over to a Q&A session. We look forward to the conversation. All right, come on up.
There you go.
Give us a second.
is everything. It's the building blocks of our world. It's our environment, our technology, how we communicate, create, and innovate. A single data point won't move the needle. But the more data points we connect, the further we can go. Because it turns out the best way to forecast the future is to build it. So we did. And this is that mark. Engineered to embody innovation and defy limitation, harnessing the unknown to reveal the future. This is the new SanDisk. SanDisk. It all starts with a pixel, a single point of data that transforms into a kinetic symbol of cutting-edge flash technology built to be as distinctive and dynamic, adaptable and versatile as we are. Our slab serif is bolder to emphasize our industry leadership and powerful foundation.
The iconic OpenD letterform unites with a fresh pixel-driven S, both symbolizing the collaboration and partnership required to actualize our purpose and tap into all new possibilities. Clean lines and minimalist design mirror the speed and efficiency of flash technology. High contrast emphasizes our precision, clear vision, and relentless spirit. From a pixel to a platform for ambition, this is a mark engineered to embody our innovation and defy limitation, a mark that is not just ready for the future, but actively shaping it. This is the new SanDisk. SanDisk.
All right. Thanks again, everyone. We'll have some mic runners in the room for people that have questions. We'll take questions from the webcast as well. We'll work them all in together. I'll do my best to try and moderate. But there you go.
Yeah, thanks, David. Congratulations on getting to this event. I know it's been a lot of work. I guess I got two quick questions. I appreciate all the detail today. I guess on the last presentation, HBF, right? It's going to take a little bit of time to get used to that. But I guess if you're successful, open ecosystem development, roadmap, et cetera, I appreciate the roadmap you gave, but there were no dates on it. So I'm curious of how do we think about commercialization of that? When is your vision of that actually showing up either in a GPU or an end device application where you start to monetize that R&D effort that you're implementing? And then my second question, I'll just throw it out there right away. Luis, you talked a little bit about the asset and running equipment across nodes.
I'm curious if we look at the amount of invested capital that SanDisk has inside of the JV that's not appreciated in your numbers, how would you characterize that? Because I think you run your equipment a lot longer past the depreciation cycle. I think in the past, you've talked about $20 billion plus. Where is that today and just how do we think about that asset value? Thank you.
Yeah, OK, thanks. So on the first question, I'll say a few words and I'll turn it over to Alper. We kind of put this after the model for that very specific reason. We knew we didn't have dates on there. We're working with this project kind of emerged over the last year. And we're very close to our customers. And we have an enormous amount of R&D capability inside the company. And our customers are looking at these very same problems as they think about their GPU roadmaps and where they're going, how they're designing their data centers, what problems they're trying to solve. And we're kind of working to intercept that kind of next generation. Also, it's very much more applicable to the inference phase, as Alper talked about. So we'll have more to say about time frame. We're not quite to that point yet.
But we wanted to kind of put the ideas out there kind of the way we think we need. Like I said, AI architecture is evolving every single day. We wake up and find new things that have been announced. And I think that's natural in a market like this. It's moving this fast. And is this dynamic? And is this important? And I think as everybody knows, I think our innovation prowess is SanDisk. And it's great to be back as an independent company and bringing projects like this to the fore. But Alper, maybe you want to say a little more.
I think there's only very little to add to what David said. The only thing I would maybe point out is that the roadmap I shared is really leveraging our existing NAND roadmap. There are no inventions needed beyond what we have delivered so far. So it's highly executable, and I think the gating item indeed is going to be enabling the ecosystem, aligning with the customers at their system level, integrate it, and then bring it to the market, so we know how to execute this.
So I'll say one more thing, and then I'll let Luis answer the second question. A lot of stuff you saw today, I think, was like a bunch of different innovation vectors coming together that arrive at a bunch of products that are kind of game changing. So if you think back what Khurram talked about on UltraQLC and these very high capacity enterprise SSDs we're now going to bring to market, it starts with BiCS8. If we didn't have wafer bonding, we wouldn't have the interface speeds to do things as fast as we're doing them. So that's like a pretty big foundation. And then on top of that, we made the important decision to build a 2 Tb die probably how many years ago, Alper?
It's been a while.
It's been a while. Many years ago, somebody had to decide, let's build a 2 Tb die. So now we've got a 2 Tb die. It's got all this great capability. And then Khurram gets a hold of it and builds a controller that takes three years to build an ASIC. And while you're building that, you're thinking about, I'm building towards this die I know I'm going to get. I know I'm going to get this BiCS8 with this very fast interface speed because I'm building the CMOS separately. And you have this compounding effect. They can take all their systems expertise. And all of a sudden, you build this game changing product. HBF is a little bit like that. You have wafer bonding that gives you a level of interface speed. You can start optimizing on. And keep me honest here, Alper, because you're the technologist.
You get the interface speeds, and now all of a sudden, you start to think about what can I do with this? You start to think about stacking. Well, we already know how to bond two wafers together. You bond them together in interesting ways, and it solves a warpage problem, so there's all these innovations that accumulate, and all of a sudden, you find out you have this kind of game-changing innovation. That's what's so exciting about where we sit today.
Yeah, I think to your other question, you're right. We've been able to use our tools for much longer than the time we depreciate them. It's difficult to assess what would be the replacement value kind of to your question. But we've done different estimates depending on, where do you do that? Do you start from what do you assume the land is worth? Do you have an infrastructure in the site? So a number of $15 billion-$20 billion is consistent with what we've estimated. But those are rough numbers on that.
Thank you, Wamsi Mohan, Bank of America. Thanks for doing this presentation. Maybe first one for you. Just as you look at the new SanDisk, you've said many times here that scale matters. How do you think about potential future consolidation in the industry? It seems fairly fragmented. And to achieve some of these goals around through cycle margins, how realistic is it in still a pretty fragmented industry? And would love to get some thoughts around China in that context. And then I will follow.
OK. So first of all, I mean, one thing you should take away from today, we are incredibly excited about SanDisk. We think we have incredible innovation. We have scale through the JV. Our path, I mean, is very, very bright. So there's just no two ways about that. I've also been very open since I came here. If consolidation presents itself, of course, we're open to that. We're a public company. We're for sale every day, as I say. So if it happens, it happens. But we are very focused on the path we have in front of us. And we're very, very excited about it. China is a long conversation. There's a lot of geopolitical things that are going on there. I think it's a market that's important to us. As we talked about, you saw the back end we have in Penang.
We have kind of a twin of that in Shanghai that we now own through a partner in JCET that we've executed in the last year, so we feel very good about where that's situated for China. We think it's an important market. You'd have to ask a more specific question for me to opine more on China, but it's something we have to pay attention to every single day, given the geopolitical issues that are going on between China and the United States. We do spend a lot of time on that, but it's a great market for us, great workforce there, important part of the footprint of the business.
OK, great. Thanks, Dave. Just to follow up maybe on these through cycle margins, we haven't really seen these kind of through cycle margins in the industry in the past. And I know you've given a lot of reasons around what is changing, both from your expectations around demand growth and cost structure and other variables. But maybe you can help us think through what you think are potential ranges or scenarios around peak and trough gross margins as you go through the course of cycles for both gross and operating margins. That'd be great.
Yeah, what I would say is if you look at the last two quarters, not to go very far, we've been at 36% gross margin. So it's not that we're trying to get to a level where we've never been. We're right there. Or we were right there just a few quarters ago. And as I said, we want to reduce the volatility around that number by managing supply and demand. It's difficult for me to give you kind of what do I think the lowest number could be. But we definitely want to avoid the lows of the past. And as I said, we're increasing the focus on operating expenses. Now, how do we justify the expenses through the lens of value creation for our customers and for our shareholders? So we're doing a zero-based budget. We're really challenging everything.
We'll make sure that we're investing the right things, which starts with innovation because that's the lifeblood of the company. It starts with building a strong brand that consumers trust. Everything else, we're going to challenge.
Thank you.
All right.
Thank you. Tom O'Malley with Barclays.
Hey, Tom.
Appreciate the event today. Thank you very much for putting it on. I wanted to ask just on some of the short-term guidance that you guys gave. So you talked about the industry kind of transitioning from Q2, where you're in a little bit of an oversupply dynamic, to getting to kind of 5% undersupply in the second half. You showed a chart with data center last year, some tough comps there. But maybe the underlying assumptions in some of the other large markets to get there is just the drop in utilization enough to get to that dynamic? Or are you expecting some things in end markets in the second half?
I'll make a few comments. And then Luis and Ivan can as well. Look, I think it's fair to say we expected kind of the PC refresh to happen sooner than it did. Like I said at the beginning, it's going to happen. I mean, whether it's last quarter or next quarter or the quarter after, I think this AI wave is going to wash over the entire industry. We're kind of doing all the things we have to do to position the business for that. I mean, the first is have the right products and be in the right position. And then be ready when it happens. But I think the thing that's changing, and it's a little bit relates to Wamsi's question as well, is we're changing the way we run the company. And I think it's not just us.
It's like we need to get supply and demand better balance on an ongoing basis. We need to get a better return for the innovation that we're putting into this business. It's incredible. I mean, you sit and you listen to Alper and you listen to Kurram and all the intellectual firepower that's going into this industry. It's got to have a better return, and I think to do that is to manage the business differently. I'll equate this back to the HED business I've been running the last five years. If I would have sat on stage in 2002 and told you that those gross margins are going to be in the high 30s%, nobody would have believed me. But what you have to believe is you can change things, and that's what we do. That's what we're in the business of doing, is changing things.
What Ivan talked about is we're changing the supply dynamic right now. We're not waiting. In many ways, NAND is a weird business, quite frankly, as somebody that's managed an enormous number of technology franchises. First of all, you tell everybody what your costs are. Who does that? What business tells everybody what their costs are? And then you go have a negotiation on what the price should be. So we're going to stop doing some of that stuff. The whole idea of the cost downs was to grow the market. Well, I mean, look at Khurram's chart. It was 2005, the industry shipped two exabytes. Is that right? You were here. I mean, now we're shipping hundreds and hundreds and hundreds and hundreds of exabytes. Like demand's not our problem. The problem is we need to get an economic return for this.
And the rule of thumb, well, you must run the fab. We must optimize on unit cost minimization at all times. Well, why? Other industries don't do that. So let's learn something from other people. And let's run this business differently. And I think we're going to do all those things. And I think some of our peers are doing those things as well. And I think this industry has got a huge, huge future. It's a big, big, big ocean. If you want to catch the big fish, you fish in the deep waters. We are in very deep waters. And we get this thing right. And it's going to be amazing.
Maybe as a segue to that, there was a gross margin question earlier. There's a lot of variables that obviously go into that. But the OpEx side, I think, is something that's much more within your control. So you talked about moving to 15% over time. Maybe the cadence of that in the near term, where are you seeing kind of low-hanging fruit in terms of operationally where you can improve things? Obviously, that goal didn't have a timeline on it. But maybe in the near term, where are you seeing areas for improvement?
Yeah, as I said, we're looking at everything. We're starting with a zero-based budget. And G&A would be an area I would first look at, obviously, because that's needed, obviously. I'm part of G&A. But it's not adding what is unique about the company. And even within innovation, we need to make sure that we're being as efficient as we can. So we're looking at everything and the obvious things, consultants and those type of things, are we really seeing the value that we need? There was a lot of expense related to the separation as well. We won't need that going forward. That work is behind us. So we're really turning, as they say, every stone. We'll turn them twice. We'll look through every rock. And we'll find efficiencies. We want to create the most value for our customers and for our shareholders.
As David said, then we need to capture more of that value. Because we're creating that value, but it's going somewhere else. We need to capture that.
Thank you.
Let's go back right next to that back there. I'm sorry. There's no Al.
Thank you, Mehdi Hosseini, Susquehanna. Two follow-up. The previous time, or the last time SanDisk was independent, there was a lot of focus on firmware and controller technology as a way to differentiate. How do you see those two elements evolving and being part of a strategy looking forward? And my second follow-up has to do with the JV. David, can you remind me who actually owns the IP for all these BiCS architectures? Is that the JV that owns the IP? Or is that Toshiba? If you could just give us an update, that would be great.
So the second one is very straightforward. All that IP is in the JV. So it's jointly owned. It's owned by us and Toshiba, our Kioxia. So we have full rights to all that intellectual property. We jointly develop it. The teams work hand in hand. In fact, although the output of the JV is 60/40, the R&D spend is 50/50, which reflects the joint ownership of the intellectual property that's created. Look, I think on your first question, I think you saw it here today. I'll let Khurram make some comments. I mean, that's a huge part of what SanDisk is. You've been doing it for 25 years. And I think what you saw today was an announcement of an architecture of a new controller that uses all that accumulated knowledge applied to the enterprise SSD market. Now, when you build a controller, it takes a while.
Khurram, you want to say a few words about that? And we're now just releasing this into the market.
Yeah, I mean, to David's point, I think today what you heard on enterprise culminates a lot of work that went into our firmware and controller strategy. Look, I mean, we're always looking for improvements. The underlying base technology that we manage is NAND. So how do you develop your IP that takes advantage of that across the segments that we participate in? That's the secret sauce. We have addressed all the verticals by making control and firmware. The next phase is how do you optimize it across segments by leveraging as much as you can? The second point on that is Alper talked about technology like HBF and stuff, where we bring our strengths to make the logic die that manages the core of flash array. So I think that's where I see ourselves heading, is to optimize our control and firmware strategy.
And then looking at new avenues where the strength that we have built is to generate more value for the company. I guess the part that I'm struggling with is, as I look at past 15 years, on the consumer side, the industry has been very susceptible, vulnerable to the ups and downs in the supply or supply-demand balances. So I don't see, or I can recollect a time where a controller on the consumer side was able to mitigate these supply-demand challenges. And on the enterprise side, I also see how SanDisk or Western Digital market share has fluctuated. So is there anything else you can offer me where these technology innovation, differentiation, and the design wins would be stickier?
So look, I think if you look at the consumer business in the last, I mean, it's a cyclical business of NAND. So you can't completely escape that. But if you look at the performance of the business in the last downturn, I mean, our gross margin, even though it was negative, was 20, 30, 40, 50 points ahead of the competition. And it was because of that consumer business. That consumer business went to 80% plus of the profit of the business. So although it's still cyclical, it's not nearly as cyclical as the rest of the market. So that's what we can offer you as far as this idea of resilience and dampening the volatility of the business. And I think we have an opportunity going forward. We talked about, I mean, Janet is here.
We're going to take a fresh look at this business through the eyes of a consumer branding expert, and not just through the eyes of a NAND team of where we can take this franchise. I'll go over here on this.
Hey, it's Tim Arcuri at UBS. There's sort of two messages here. One is that you're going to improve profitability through the cycle. But the other is that you also want to gain share, particularly in data centers. So I guess the question is, when push comes to shove, which one of those two will you prioritize? I mean, it sounds like it's going to be profitability. But that's what a lot of us struggle with when Samsung is the one who controls all the supply or a lot of it.
So we want to get to our kind of historical share. We talked about share goes up and down year over year. So we want to think when we say grow with the market, it doesn't mean off of where we are today. It's like integrated over the last three, four years is what we think our long-term share target is. And again, I talked about this with optionality. We want to have great products in every market we play in. And we've made progress in enterprise SSD.
Now we're going to make a lot more progress because we're launching a whole new set of products that have a lot, as I talked about earlier, a combination of BiCS8 and CBA technology and a 2 TB die and all the controller work that Khurram talked about and bringing all of our expertise of everything we've learned over 25 years about how to build the best controller for QLC to build the best product. Once you have that pillar built out, then it's just mix every quarter based on what the best outcome is going to be. Like during the downturn, if we would have said we wanted high share in enterprise SSD, that would have been a terrible place to be because that market was the worst.
So we want to build out that optionality and then get our fair share of the bids for what we have to supply and get the optimal return for that. That's one of the ways that we get better through cycle profitability is having enough agility and optionality in the business for the bids we do have to get the highest return for them based on where the market is at any given time. And the market changes every couple of quarters. I mean, we've seen that over just the last couple of years about how the ebbs and flows of each segment of the market have been very different from the profitability point of view. It's actually quite stunning when you look at the business from the inside. You can see like 50 points of margin difference from worst to first as far as where you're putting bids.
So you want to get the most optionality in the parts, have enough optionality to put as much supply as you can in the places that are going well.
Got it. And then depending on how the tariff situation evolves in China, do you have enough capacity in Malaysia to service all of your non-China demand?
Yeah, we have plenty of capacity for that. And we have capacity for expansion in Malaysia. That's a very, very good facility.
Great. Thanks.
Thanks. And thanks for all the information, Steve Fox with Fox Advisors. Two questions. One is just on the. You've talked a lot about bandwidth improvements that you see. You've talked about read-write capabilities across the presentation. Putting aside HBF for a second, how do we think about just what kind of roadmap you're on to improving bandwidth and read-write in order to make your products more attractive? And then just secondly, on the gross margin, can you talk about some of the swing factors? Do we think about MIX, like Dave just described, as not really changing securely? Or is there opportunity for gross margin to improve from MIX over, say, a three-year period? Thanks.
Khurram, Alper. Khurram, you want to talk about?
We can. Alper, you want to go first? And then I can put my.
Sure. I mean, at the product level, of course, you start with the technology, the underlying technology first. I'll touch on that. And then Khurram talks about the product level improvements. I think I showed you how BiCS8 significantly leads competition, not only in current generation as well as the next generation of products. And we very much intend to continue that leadership throughout. I mean, the things you talked about, read-write bandwidth, power efficiency, are primary considerations in the way we design. And the advantages we have delivering that bandwidth as well as the I/O speed are all going to continue. We believe we have sustained advantages. And just look at the numbers today. They're so much better than what's next in line. And we intend to be there throughout.
Yeah, I mean, it's really the technology, I/O capability that leads the way. And then we design our solutions to match that. And when you look at it, what Alper talked about, we have a significant runway ahead of us in terms of improving the performance. We've got great toggle mode roadmap in terms of programming speed. I mean, we've got great runway ahead. And that's what's allowing us to make these transitions to the host interfaces, like going from Gen 5 to Gen 6, is really what the technology team provides. And then we build the ASIC that can match it. So we feel pretty good about it for several years to come.
On your gross margin question, yes and no. This is a very interesting situation. You can always improve your gross margin through MIX. But you optimize for today. That may not be optimal for tomorrow because prices may fluctuate in different segments at different rates based on supply and demand. So I think, as David was saying, what we want to have is optionality so that we can place our bids in the best possible way so that we can convert those bids into revenue and into cash flow. So we want that flexibility. And we will be optimizing on an ongoing basis. But that optimization will change over time.
Thank you.
May. No, up front. You're the white.
Thank you for the opportunity. Asiya from Citi Research. To what extent do you think consolidation in the industry is needed? I mean, we talked a little bit about fragmentation. But to what extent do you think consolidation from SanDisk's perspective is needed to drive some of the target models that you've laid out today?
The target models we laid out are not. They're based on operating SanDisk as it exists today, so we feel very, very good about that model, and we feel very good about the ability to execute upon that. Obviously, in any industry, if you can drive consolidation, there's various levels of synergies you get, and so I think you can make a general statement about almost any industry that consolidation leads to some level of synergy creation, or you wouldn't do it. I just managed the HDD business for five years, and people give me this rule of thumb, like, oh, there's fewer players, so there must be more profitability, and that's not necessarily always true. I think profitability comes from the ability to innovate and capture the value of that innovation and solve customer problems, and of course, structure is part of it, but it's not all of it.
And so I think we have an awesome opportunity with the innovation roadmap we have. I mean, I think what we laid out here today is like maybe some very surprising innovations that we're talking about driving in a market where we have some work to do in enterprise SSD. We've always been strong in client. We've always been strong in consumer. But it takes a while to build. Khurram's stuff, I have a lot of grace for Khurram. I used to do that job when I was coming up through as a professional of running engineering teams. And it takes a long time to build an ASIC. It's not something you just wake up and do. It takes years to get that done and tape it out and get all the firmware done and get software done.
We're just entering that point on enterprise SSD, which hasn't been our strength. So that gives us a lot of opportunity there. And then I think we have a lot of opportunity. The world changes every single day. I mean, I think AI is just massively transformational. And as I said earlier, you wake up one weekend and you find out somebody in some far part of the world has changed the model again. And from my point of view, I've been doing this almost 40 years in different industries. It's not going to be the last time that happens. This is going to happen over and over and over again. And I think we've got an awesome opportunity to participate in that innovation cycle. And we bring that innovation to market. And you solve really hard customer problems. You get rewarded for it.
That's what we're going to focus on doing. All right, right there. Next, we'll take one from the webcast. I don't want to ignore the folks that are online if we have questions there. Jasmine has.
Yeah, hi. Thanks for taking my question. It's Krish Sankar from TD Cowen. I have two questions. One is, you spoke a little bit about the CBA technology, which kind of makes a lot of sense. But if I remember right, I think it was something that was pioneered by YMTC. And it looks like a lot of other NAND folks are also going to adopt it. So I'm curious, what is your unique advantage when it comes to CMOS bonded array? Is it less cycle time? Is it fewer wafer passes? Any color that would be helpful? And then a quick follow-up.
I can tell you a comparison point between the other example that you have given because you can't just look at a single technology ingredient and say, oh, they have it too. For example, the data we have as publicly shared on YMTC's implementation with 232 layers of memory compared to our 218. So they have more layers. So they're more expensive. So you expect them to be smaller. Their die size was about 25% larger. Our lateral scaling memory density was more than 30% denser. So that's the huge advantage. Essentially, you can think about that as a cost advantage. So they're like two generations behind. It's not that you grab an ingredient of a technology and say, oh, I got it too, versus how do I get this and turn it to a true competitive advantage in terms of cost, performance, and power.
And when you look at all those metrics against some of the examples you have given, we lead by a wide margin. It is true that the CBA is probably the path of the future. And the industry is probably going to converge over several generations. But I'm really happy about the way we are approaching in terms of continuously using this as an innovation platform and finding more ways of doing more with CBA. I didn't talk about all of it today. But some of the things we talked about in HBF, et cetera, they all do that. It's not just plain vanilla HBF. We've done a lot with CBA. We've taken it to the next level. And we will continue doing so. There is an incredible innovation roadmap ahead of us. And we're super excited about that. But it's what you do with it, not just an ingredient.
Got it.
Look, I'll say one thing on this. And this, like any technology industry, is really hard. I mean, you have a lot of respect for all of your competitors. And you have to be relentless in innovating. And that's what we're going to continue to do. And with the relationship we have with Kioxia, and the level we can invest in this, which is as much or more than anybody else in the market, and been doing it for a very, very long time, we're going to continue to relentlessly push that innovation vector forward.
Gotcha. And David, I just had a quick follow-up. I understand you're doing very well trying to innovate balance share and profitability. I think the last analyst day in 2022, you guys targeted around 20% ESSD market share. I understand we're under-indexed on that side. Are there any quantifiable share goals you could share in terms of what to expect in ESSD? Thank you.
Yeah, I don't think we said 24. I think it was 16. But I don't know. You may be right. I may be wrong. Look, we want to get, I mean, I think the first goal in enterprise SSD is to get our fair share, which is our bit share. And we're going to drive to that. We've already talked about this fiscal year. We want to drive 15% to 20% of our bit share. We're halfway through the year. We're on track for that. It's a dynamic market. So we have step goals to go into that. We're really waiting for this multi-year development effort to play out. But we're going to hold ourselves accountable to those goals. I will say I'm a little hesitant as a manager to be careful with market share goals, not because I don't want to be held accountable to them.
But sometimes when you put them in place, organizations do strange things. Like they start discounting products just to get share because they think share is the goal. The goal is profitability. The goal is free cash flow. And what we want in enterprise SSD is maximum optionality to play in as big a part of the market as we can. And we think we're on a very good path to be able to do that. All right, let's take one from the webcast. Jasmine.
Thank you for taking the question. How fast will BiCS8 supply ramp? What % of bits will account for by end of calendar year 2025 and 2026? And will CUA maturity be a factor in initially limiting cost and bit scaling?
Yeah, I'll take the first part. So I think I had it on the slide, but I didn't mention. So at the end of fiscal year 2026, we expect to have 10% of our shipments.
25.
At the end of 2025, of fiscal year 2025, we expect around 10% of our shipments to customers to be BiCS8. At the end of fiscal year 2026, we expect to be somewhere between 40% and 50%, which means that by the end of fiscal year 2026 or beginning 2027, BiCS8 will be our prominent node.
Maybe if I fill in on the second side of it, I think the question was, is CBA going to be a yield limiter?
Will CUA maturity be a factor in initially limiting cost and bit scaling?
I think they meant CBA because we don't use CUA in BiCS8. The answer is very clearly absolutely no. BiCS8 probably has been the fastest-to-yield technology that we had in the company historically. And its yields have been going steadily very fast, way ahead of our expectations. And CBA is absolutely not a ramp limiter. It's just been fantastic technology for us. So we're very proud of it.
All right, how about in the room? Any more in the room? In the front. It's like a live earnings call or something. All right, everybody in the same room? It's good.
Hi. This is Yang for Karl Ackerman from BNP Paribas. Thank you very much for the presentation, and congrats on the new phase of your business. So my first question is for Kiran, so I hope you can clarify. You mentioned that you have a new product designed into NVIDIA's Blackwell. I think it's DC SN861. And can you elaborate? Is this product in the current B200 NVL72, or it will be in the next B300 NVL72 rack?
Yeah, thank you. So the process, the way it works is obviously this certification is an enabler. And what happens depends on you have different flavors of offering from NVIDIA that ODMs build on top of. And then these ODMs get orders from their customers. So what's happening is once you're on the approved vendor list with this certification, then when the ODMs are picking the design, they pick vendors from the approved vendor list. So that's how we are getting into qualifications with the system partners that are building on top of NVIDIA, not necessarily the NVIDIA build. So that's where we are gaining the momentum in the market.
Understood. Thank you. And my second question, so I want to double-click on the competition. On a high level, I think you talked about the market is growing kind of slower than we previously, a few years ago, we were thinking about. We now talk about high teens, big growth. A few years ago, we were talking about high 20%. But the market is still somewhat kind of fragmented. And how do you find confidence that other players will execute the same way you think, with the same discipline in supply investment? And how do you think about the competition from smaller players, like the Chinese players, if they would increase investments and try to play a catch-up in technology?
Ivan, you want to?
Yeah, sure. I think that obviously, the demand growth somewhat depends on the time period. But I guess as an example, talking about some higher rate of demand growth in prior periods, well, what was the profitability? It doesn't necessarily correlate to better margins, better returns, better cash flow. So as we talked about, getting in the, whether it's mid-teens or high teens, bit growth, depending on what time period we pick going forward, there's plenty of growth there for our business to expand margins and expand returns. I think that's the key here. Now, relative to competitors, obviously, we can't speak for competitors. We can see what's been done publicly already. And I think you've seen everyone's if you look at the aggregate CapEx, it's pretty benign right now. And I think we've heard certainly announcements or discussions about reducing utilization, adjusting supply because of the price headwinds now.
But that's obviously out of our control. And we control our own business. And as we talked about, we see those price headwinds. And we've taken that action to reduce supply output as a result of that.
Thank you. So Ivan, let me ask you a question. So you were until recently outside any particular company watching the industry. What did the last downturn cost the industry?
The cumulative loss was about $40 billion of cash flow.
OK, so let's not do that again. I think anybody like, let's go do that again. If we all can't learn, I mean, I don't know what it takes. Right?
Thank you. That makes sense.
And by the way, we did the best of everybody in that. As far as financial performance, because of the consumer business, we did the best of anybody. And we don't want to do it again either. So let's all not do that again. We can't speak for. I prefer to call them our peers. We can't speak for our peers and never will. But what we're saying is how we're going to run our business is how we're going to run our business. And after you, I'll go back to the webcast. Go ahead.
If I take a dip?
Yeah, go ahead.
Thanks for hosting today.
Of course.
C.J. Muse with Cantor Fitzgerald. First question on capital intensity. A lot of today's talk has been about the advantages that you have. So I'd love to hear your thoughts on what you think industry-wide and capital intensity is. And then for you guys, what do you think is permanent versus perhaps what is temporary? And then as part of that, how much of it is just perhaps just accounting-wise with the JV? And then I guess just on the second question, spoke a great deal around the mix shift to BiCS8. Would love for you to perhaps elaborate on what other drivers will help cost down over the coming three to five years. Thanks so much.
I think on CapEx, what you saw is in some of the earlier slides from Ivan, particularly, is the industry spending somewhere around 25% of revenue on CapEx. And as I explained to you, we've been spending about 16% of revenue over the last three years. And we expect to be somewhere in that same range going forward. I also explained why we think we're lower than the industry. I talked about our scale. I think I referred to the work that we're doing in R&D on scaling and multi-layer scaling and lateral scaling, particularly as a factor of reducing our CapEx. And I also mentioned that we reuse a lot of our assets. Some of them are fully depreciated. And you keep on using them. Over 90% go back to the next nodes. So that's why we believe that some of this is sustainable.
I also had on my slide, we expect to be slightly above our ongoing model in the short term as we transition to BiCS8, but nothing significant.
So Ivan, do you want to add anything about the industry from CapEx or any point of view?
No, no. I just think that, yeah, you see, again, a structural change where you can't manage this business running 50% CapEx to sales as we saw in that sort of peak 3D era transition. And we're through that now. And I think you see, obviously, we're running more efficient. The whole industry's running just less aggregate capital and less aggregate supply. And as we talked about pretty extensively, we just have an efficient model and scale where it matters.
So, C.J., on cost downs a little bit. Look, I think the cost downs are slowing down. I'm not saying NAND is hitting a memory wall or something like that. Far, far from it. In many ways, the R&D is extraordinarily productive. We can build new nodes and, I mean, we proved that in the transition to 3D. It's very easy to build new nodes and oversupply the market almost by accident. It's just like a crazy number of bits come out the other end. But I do think cost downs are going to slow down a little bit. I think that's reality. When you look at the roadmap of the technology, everything gets harder. And I think the way we manage the business, where we talk about taking supply interventions, as Luis says, in his words, take supply interventions sooner means costs are going to be more variable.
So you're not just going to be able to model 15% down or 12% down or 11% down, whatever it is, and think that happens quarter over quarter. And I think this whole industry of talking about cost to expand the market is not a demand problem. We need to talk about cost less. And we need to talk about profitability more in capturing value. And that's what we're going to start doing as opposed to showing up and starting talking about cost all the time. Oh, everything. Again, I go back to five years ago in the HDD industry. It just fascinated me. Every quarter, we showed up at the customer and was like, OK, everything's going to be down 3%-4%, right? And you're like, what are you talking about? Who does business this way?
So I understand in storage, you want to decrease the cost to increase the TAM. Like I said, increasing the TAM is not our problem. We're on our way to $100 billion TAM. The idea is get the economics right, get the capital intensity right. I think this is going to become a more nuanced conversation between us. And we'll talk about it and how we guide. But it's not just going to be these very predictable level that we've seen in the past, kind of incrementally, every quarter we go down, we go down, we go down.
And if I just can add, if you don't mind, David, I think that for me, it comes back to we'd rather have price. I mean, you want to get price leverage and benefit as opposed to shipping one more bit. And part of this cost, the shift in the cost paradigm in the industry too, means that's even more so the case right now. And I think that's what you see in the actions that we've taken.
I can keep going on this for a long time. It all starts with innovation. You've got to show up with something that is valuable. And I don't think there is any doubt we are showing up with something that is extremely valuable to our customers. Our customers are building multi-trillion-dollar businesses on top of our innovation. Now, a lot of other innovation and a lot of their own spectacular innovation on their own, they deserve everything that they build. But we're a big part of that ecosystem. And we're going to show up with that innovation. And we're going to show up with it in a way where we think we need to capture more of that value. All right, Jasmine, on the webcast.
Of course. A part of this question asks about the supply and cost dynamics, which you just addressed. But let me rephrase the last half of the question. They're asking about how investment in technology and CapEx decisions are made in the JV.
Yeah.
Want to do that, or you want me to?
No, go ahead.
Well, it is a collective decision within our JV. We essentially align between the technologists as to what our roadmap, technology roadmap, is going to be. The development of the technology is a 50/50 ownership. So we show up with as many people as they do and collaborate on development of the core process technology as well as design of some of these dies that you have seen. We own half of that. And then we sit down and look at how we each sort of have needs for specific features. And then we go design those and get it into production. And when we get into production, then there are aligned decision points as to what capital deployment plans will be within the JV. And we go decide and execute that.
Yeah, there's different forums where we review our CapEx spending with the JV. So there is good escalation. We have a yearly plan. We then approve each time smaller and smaller increments so that we can control our CapEx together. Now, to be clear, if there is ever a disagreement, we're not forced to proceed with the investment. So we have that ability to say no to an investment if we're not convinced at the end of these discussions.
OK. Let's take one more from the webcast and one more in the room. And then we'll wrap up.
I think we're all set for.
We're all set. OK, so somebody in the room, a lot of pressure for the last question.
Just a quick question on the CapEx side. You saw your CapEx going up this year with BiCS8, I guess, to the mid to high teens. Does that come down next year? Or how do you see that CapEx going out? And same on the Ultra BiCS side. How's the CapEx there, I guess? And I'll put in a follow-up in here as well. The HBF, are you working with any of NVIDIA or any of the custom silicon ASIC guys in trying to qualify that HBF roadmap on a long-term GPU or a custom silicon? Thanks.
Yeah, I thought you were going to be super happy I gave you 25. Now you're asking for more, obviously. So I mean, at the end of the day, we expect to be somewhere in the range that I mentioned over several years.
So there will be a little bit of an increase as we transition to BiCS 8, but nothing crazy. So I wouldn't put too much on that. Do you want to take the other piece?
The answer is simple. Yes, we're working with multiple customers, and we want to, after this announcement, hopefully increase the number of engagements we have. It's really important to establish that ecosystem enablement system enablement so that people design for it. It's going to be an architectural evolution, and these things take a little bit of time, but ultimately, I think the advantages are pretty huge, and what we can deliver through this is pretty meaningful advancements in the AI space, so looking forward to that.
OK. All right, we're going to wrap up the webcast first. As you guys will stay here, we'll talk about, we have lunch, but let's wrap up the webcast in the whole session in general. We really appreciate all of you spending time with us this morning. We know it's a big investment of your time.
Hopefully, you see we're very excited about where we're going with SanDisk. We're only a couple of weeks away, less than a couple of weeks away now from this company trading as an independent company again. We're super thrilled to be here, and we're glad you're spending time with us, and look forward to a long and fruitful ongoing dialogue. So thank you very much.