Good morning, everybody, and welcome to the first corporate presentation of the 2024 Deutsche Bank Technology Conference. I'm Ross Seymore. I head up the semiconductor research efforts for DB, and I'm here with my coworker, Melissa Weathers, who handles the semi cap side of things. We're very, very honored to have Micron Technology on stage with us today, both CEO Sanjay Mehrotra, as well as Manish Bhatia, the EVP of Global Operations.
So Sanjay and Manish, thank you very much for joining us here. So we'll get started with a bunch of different questions about the core technology, the spending, AI, all those sorts of things. But I just wanted to start off by knocking out some of the near-term trends. What are you seeing in the near-term trends? I know you're just about to finish your quarter very, very soon, so you might be limited in what you can say. But how are you seeing the near-term trends, you know, bit growth, pricing, general market demand trends in your various end markets? Any sort of updates on those would be great.
First of all, let me say, Ross and Melissa, thank you for having us here today. And let me also remind you of the safe harbor statement. In the presentation today, we'll be making some forward-looking statements, and there are risks associated with those statements, and we file our risk factors in the SEC filings that we make from time to time, and we just ask you to refer to the risk factors in those filings. So as you said, we close our fiscal fourth quarter of 2024 actually tomorrow.
I will not be providing any updates on fiscal fourth quarter here, but I will tell you what we said last time, that overall, you know, data center is one with robust demand trends, and Micron has a strong portfolio of products from DRAM and NAND, from HBM products, high-density DIMMs, solid-state drives, and of course, our low-power solutions as well, so a very strong portfolio, well suited for data center applications, and data center is driving the growth enabled by AI, and of course, we see that traditional servers growing this year as well, and of course, AI servers have strong growth.
A nd then, you know, with respect to smartphone and PC markets, what we have said before is that with respect to the end market, say, a sell-through of those devices, those markets remain stable. And of course, AI, as we look ahead, drives content growth in smartphone and PCs as well, and we have shared some of those details earlier. So the smartphone and PC customers have built some inventory, and that really is primarily for three reasons. They see the pricing trend up, going up, so in preparation for that, that's the first reason. The second really one being that, they see strong data center demand.
They understand the HBM factor that is a tailwind to the industry demand-supply environment. They see supply tightness. So in preparation for supply tightness, they have taken on some inventory as well. And the third one, of course, is that AI content in smartphone and PCs continues to grow, and they look at AI smartphone and PCs penetrating further as we go ahead over next few quarters and of course, next few years, and in preparation for their own demand, increased content in AI devices. So these are really the three factors that smartphone and PC customers have built some inventory.
We have shared before, and as is well known in the broad industry reports, that consumer retail channels, industrial, automotive tend to be relatively weaker right now, as well as China has some weakness too. But overall, there are long, strong, longer-term trends in industrial and automotive markets, again, driving healthy content growth as we look ahead into the future. So when we look at all of these factors and all of these trends, when we look at our FQ1 , we think our bit shipments will be somewhat flattish to slightly up in FQ1 versus FQ 4. And as we have discussed before, we see that our gross margins will sequentially strengthen as well.
And these trends of AI and the tight supply environment, we see that continuing in 2025 as well, and that's why we say that we will have substantial revenue record in 2025 and, of course, robust profitability in 2025 as well versus 2024. And we are very well positioned with our technology and product portfolio. We believe that a best position in the history of the company with respect to the product portfolio, and that's why we are very much focused on driving the mix of our sales toward strengthened profit pools, leveraging the strong portfolio and driving the revenue mix, towards strong financial performance, in 2025 as well.
A nd one of the big factors that I would just like to highlight here, Ross, is HBM. HBM, of course, is driving strong growth, and Micron is well positioned with our HBM product. I'm sure we will talk about that as well. It's an exciting market. I mean, last year, in calendar 2023, HBM consumed about 1.5% of the industry bits, and the HBM TAM was, like, $4 billion. When we look ahead at 2025, we see HBM consuming around 6% of the industry bits and industry TAM in 2025, greater than $25 billion.
So it's exciting market. Micron is well-positioned with our HBM product, and HBM has, given the 3-to-1 trade ratio that is there with HBM, it has the impact on tightening the supply market and providing a ripple effect on pricing for non-HBM market as well. So we see a healthy demand supply environment in 2025, and we see a healthy pricing environment in 2025 as well, leading to our projection of substantial revenue record and robust profitability in 2025.
Great, thank you very much for that. I think you just provided the perfect segue to AI/HBM. So why don't I just hand it over to Melissa to dive a little deeper into that topic?
Yeah. Yeah, thank you so much for taking the time this morning. I wanted to ask a question. You guys are basically booked for calendar 2024 and calendar 2025 on that HBM supply. I think customer behavior, customer willingness to commit to that kind of capacity, is something that has been mixed in prior cycles, where long-term supply agreements can sometimes be viewed as a risk for adding too much capacity. So as we think about customer demand in calendar 2024 and 2025, like, how are you thinking about building out your HBM capacity? How are you thinking about managing that supply, specifically on the HBM side? We can get into the non-HBM demand later.
As we have shared before, our FQ3 was the first quarter of HBM revenue shipments, where we had more than $100 million of HBM revenue, the very first quarter of HBM volume shipments. We had shared that even in FQ3, the very first quarter of HBM revenue shipments, our gross margin contribution was accretive from HBM. Accretive to our corporate margins, as well as accretive to our DRAM margin. We see, of course, this trend continuing. We have said that in our fiscal year 2024, we will be shipping several hundred million dollars of HBM revenue, and again, that fiscal year 2024 ends tomorrow. For fiscal year 2025, we have said multiple billions of dollars of revenue.
So our trajectory of HBM is strong, and we have also said that we, of course, are sold out with our HBM capacity in 2024, as well as in 2025. And in calendar year 2025, sometime in calendar year 2025, we expect to be reaching our share of HBM to be in line with our DRAM share, which is in the low twenties. So we are being disciplined in managing our overall business, managing the mix of the business, and as I said, extremely focused on maintaining our bit, overall DRAM bit share, as well as NAND bit share. Maintaining that bit share and managing the mix of the business toward the greater profit pools, and HBM is certainly an important, has an important role to play in that regard.
Well, today we're into that. Can you talk about that decision to maintain share and, potentially at the risk of not gaining share, especially as we think about the competitive environment in HBM? Like, what is your willingness to potentially gain share in that market specifically versus maintaining?
So let me be very clear. In HBM, of course, we are gaining share, right? I mean, our share in HBM so far has been pretty low, and going from a very low share to our share lining up with our DRAM share sometime in calendar year 2025 is, you know, tremendous opportunity for us. We are excited with our strong product of HBM3E, which is industry's best product when it comes to power, that is 30% lower.
This strong product, along with our manufacturing capabilities, enables us to get to our share to be in line with our DRAM share in calendar, sometime in calendar year 2025. So clearly that is gaining share in HBM as we look ahead, and that's an important part of our strategy. I was talking about maintaining overall bit share across our total portfolio, and of course, managing the mix of the business across the end markets, from data center to the edge, while maintaining the bit share, managing it toward higher revenue opportunities, toward higher profit opportunities, and it's all of that combination that leads us to say that calendar year 2025 will be robust profitability.
You mentioned about the difference in shares. Of course, you're gaining share in the HBM side of things, but kind of two questions. The punchline is, why couldn't you have more share in HBM, and I think that's where Melissa was going, than you have in traditional DRAM? Let me preface that a little bit, I guess, retroactively. At dinner last night, we talked about the technology advantage you have with the HBM3E side of things in across a number of different vectors. Can you, one, talk a little bit about the technology advantage? Two, how persistent is that? And three, coming back to the punchline, if it is such an advantage, why is there a cap at that kind of low twenty share?
So I will have Manish comment on the technology advantage related to HBM. But again, I would like to emphasize that we are going from a very, very low share of the HBM market that we currently have, and we are extremely focused on getting to a HBM share in calendar year 2025 that is in line with our overall DRAM industry share. And this is a big opportunity. We are going from $100 million of revenue in FQ3 to multiple billions of dollars of revenue in calendar year 2024.
So absolutely, we are focused on capturing this opportunity, and we are pleased with our execution here. And of course, we will continue to manage our overall mix of the business, which is an important consideration for us as well. Overall mix of the business across our entire portfolio, addressing data center, PCs, smartphones, automotive, industrial, and consumer markets.
And so, Ross, as we talked about at dinner last night, you know, the foundation for HBM is the memory technology, right? And for us, in HBM3E, we're launching on our 1-beta platform, which actually is the industry's most advanced DRAM platform or technology node. And we've been mature on that node for 18 months now. So we have a very strong foundation, very high performance, low power capability inherent within that DRAM technology node, and that's kind of the, as I said, the foundation for us to build HBM on top of it.
And then, you know, there are; o ur engineering team has done a great job of designing in additional features that give us that 30% power advantage over our competitors for that product. So we really have a great product from a technology and design perspective. From a manufacturing perspective, the other two areas that are important part of our capability, both for HBM3E, 8-high, 12-high in the future, are the packaging technology that we have. We have a lot of experience with packaging, with 3D stacking, going back to Hybrid Memory Cube, which Micron pioneered many, many years ago, and we brought some of those technology elements together.
As Sanjay said, we've been able to achieve, really, you know, we've been able to execute against the plans that we laid out at the beginning of the ramp in March, relative to both fiscal Q3, and then, you know, as Sanjay said, several hundred million revenue target for the second half of the year. The other element of the manufacturing strategy that I think is also differentiated for us, you know, maybe we don't talk about as much as testing. HBM 3D stacked cubes is very, very challenging to test, given the nature of the form factor and the performance and the expectation by our customers of what the incoming quality specs are.
We've been designing our own test hardware for, you know, decades at Micron, and that really gives us another differentiation, we think, in terms of being able to deliver the best quality to our customers. So all of these elements around technology, design, packaging, and test are all foundational for us and we think will allow us to continue to have the strongest product portfolio as we go to 12-high, as we go to HBM4 and beyond.
And an important factor with HBM, as we have discussed before, is that HBM takes about three times as many wafers. HBM3E takes about three times as many wafers as a standard product to produce the same number of bits in a given technology node. And as we look ahead, when we go to HBM4 in 2026 timeframe, it will take even a bigger ratio. And therefore, when we look at managing our overall supply, of course, that is an important consideration as well.
And again, with respect to data center robustness that I discussed earlier, I mean, that's of course, you know, HBM is a big factor there, but it's really for AI in data center, it's really our complete portfolio of high density DIMMs, low power solutions, SSDs. All of this is what is driving our robust growth in data center, and that data center opportunity has really continued to strengthen as we talk to you today.
I'm gonna dig more into the three-to-one trade ratio that you guys have talked about for HBM3E, and maybe, I don't know, Manish or Sanjay, who would be better to answer this? But as we think about the evolution of your HBM product, how does, like, what are the aspects that would cause that trade ratio to grow? And then also, as we think about 8-high moving to 12-high, what are the manufacturing challenges and that we should be contemplating as we think about, like, how you're allocating your capacity?
Manish, go ahead.
Sure. So most of the biggest reason for the trade ratio is that the performance requirements of HBM memory are much higher than the performance requirements of standard LP memory and DDR memory. And the packaging design with the through-silicon via structures requires more space on the die as well. So the combination of the high performance requirements and the high performance kind of coupled with the high bandwidth nature of what through-silicon via interconnect drives makes the die size larger than a standard product, and that's the biggest driver.
Other factors are the yield challenges when we're trying to stack, you know, 8-high going to 12-high product. Those, the yield challenges in the assembly and test process are greater than they are for standard products. But the biggest factor is this, you know, performance and IO capability. And that's gonna continue to grow, a s HBM4 gets defined and finalized, you're gonna see a significantly higher number of IOs and significantly higher performance capability, and that's gonna mean that this gap between standard products and HBM products will continue to grow over time.
In terms of 8-high to 12-high, you know, the, I talked about the packaging technology and the challenge of stacking the die. There's really two factors in packaging that are more complex. One is the fact that you're having to align more die further up, and ensure that you have good connectivity to all these vias that are interconnected. The second is that the Z-height of the overall package doesn't scale quite as much as the, as the, you know, one point five nature, so of the twelve to eight. So we end up having to handle thinner die, and that requires more precision within the assembly process.
So these are. And then, you know, I mentioned the testing capability. Now, you have to make sure that 12-die stack, you don't have any defects in each of the memory. Otherwise, you have to throw away 11 good die because of one bad. So these are all the elements that make the manufacturing of 12-high more challenging, but we feel good about where we are, and we feel good about our, as I mentioned earlier, about our capabilities to be able to progress and continue to lead in the HBM space.
Maybe the last question on HBM. HBM is a lot more customizable, or I think going forward, there's been a desire for more customization, so that customers can allow for more IP integration. How does Micron enable this trend of customization? What are customers looking for on the HBM?
Absolutely. That's an exciting opportunity for HBM. When we look at HBM4, 4E, that enables us to bring potentially more some of the customer logic into the HBM product solution. Keep in mind that HBM solution consists of a stack of eight or 12, as Manish has discussed, memory die, and then there is a base die, which is the logic die. In that base die, unfortunately, with HBM4, 4E kind of solutions, where you can bring customer logic into the base die, and that brings memory and logic closer together, and that's goodness in terms of driving performance, in terms of driving power efficiency.
Working closely with the customers, and we are working on those kind of solutions right now, and you know, these are customization opportunities with HBM4E. We are working closely in terms of defining the architecture and the features related to this, and absolutely, it is an opportunity that gives us customization, opportunities, which really brings even greater value to the customers. Today, HBM really is a key enabler in AI accelerators, AI solutions, and HBM4E, where this customization opportunities will bring even greater value to the customer. So it's a very exciting opportunity, and we work closely with our customer ecosystem to address the roadmap of this for the future.
Why don't we use that as a transition point from the HBM side to the more traditional DRAM side of things? Maybe kind of a conceptual question first. Is HBM a substitution for, or an, a kind of a compounder of the traditional DRAM side, especially when you put the capacity decisions that you have to make between them and then the pricing ramifications?
I mean, HBM is a key enabler of high performance that is needed for AI solutions. But keep in mind, these AI solutions, these AI accelerators have HBM, and they have the low-power traditional DRAM as well. And so that's why, as I mentioned earlier, for our AI opportunity, we have a strong portfolio of products going from HBM to low power, to high-density DIMMs, as well as, of course, SSDs. So this is a really tremendous opportunity here.
And as we manage the mix of the business between HBM and non-HBM on the DRAM side, of course, we keep in mind the trade ratio that is there between HBM and standard products. Of course, we in a disciplined way manage our CapEx. HBM assembly operations require unique CapEx toward HBM, different from standard products. So we, of course, manage that mix, focused on CapEx, on return on investments, and driving our overall portfolio as well.
How do you think about the capacity growth or lack thereof on the traditional DRAM side? I believe you've talked in the past about kind of exit rate, fiscal 2022 to exit rate this year and how it's come down. How do you think about that going forward?
Manish?
Yeah. So, I mean, I think one of the things that's important to keep in mind about the overall growth cycle, we've said that we expect the DRAM industry to be growing in mid, in the mid-teens over time, and that includes the fact that HBM will have a, you know, have a trade ratio, right? So that, you know, implied growth could be higher, but the inclusive HBM mid-teens is the growth rate that we see for DRAM and for NAND in the high-teens.
And, you know, that I think, you know, you asked a question before about, you know, is this augmented or is it, you know, is HBM an augmented part of growth or is it a substitute? I think that the overall AI trend is going to have, you know, a positive impact on all the end markets, because there's, you're going to see generative AI implemented in all different end applications. Sanjay talked about smart PCs and smart AI PCs and AI smartphones, but even as you think about automotive and, you know, getting towards increasingly autonomous capabilities in cars, generative AI is going to be deployed there.
You're gonna need more memory, you're gonna need more, you know, the generative AI suite of capabilities there, and certainly that's gonna be favorable for memory as well. So all different real, you know, major end markets are gonna end up seeing a generative AI impact. I think that's gonna be, you know, positive, but, as I said, our plans are based around a, you know, mid-teens long-term CAGR for DRAM bit growth and high-teens for NAND.
Let's dig more into that AI at the edge topic. This is something that we talked about last night. As we think about, the compute happening at the edge versus in the data center, do you see a significant shift in spending, like, away from DRAM towards NAND, away from HBM towards low power? How would you frame the opportunity, let's say, three years from now, when we're actually starting to have phones that have significant-
Yeah. So, I mean, on the edge, particularly smartphone, PCs, automotive, we absolutely see content growth driven by AI. Again, AI is memory intensive from data center to edge. And edge, of course, low-power solutions are critically important here. Our low-power DRAM, very, very important. Our managed NAND solutions that go into mobile phone, extremely important as well as we look at this AI opportunity. We have shared before that, and you have seen actually in some of the announcements of AI-enabled phones during the course of the year, that smartphones that are AI-enabled will have about 50%-100% higher content compared to the flagship phones last year.
There have been phones already introduced that have 16 gigabytes of DRAM. You know, last year, primarily around 8 gigabytes. So strong growth in AI-enabled smartphone. Same story in AI-enabled PCs. about 40%-80% increased content compared to the average PCs today. So, AI definitely drives stronger DRAM content, and as generative AI creates more data opportunities, we see NAND storage content increasing across the edge devices as well. So this is, again, an exciting opportunity. Certainly, low-power DRAM and managed NAND solutions are clearly required, and we have a strong portfolio there.
Maybe we can switch to the profitability side of things. I guess I'll start back on HBM. What do you view as; c learly, you're adding a ton of value with your HBM offerings. That's been accretive to your corporate average gross margins or margin profile. What do you see as the through cycle profitability boost from HBM? And as we think about the value that memory adds going forward, especially in this age of AI, do you think that maybe the floor on your gross margins could be higher than what you've seen in past cycles?
So what I would say is that certainly AI and certainly HBM drives a boost to the gross margins of the overall business. No question that AI and HBM are accretive to the gross margin trends. And of course, you know, the key is managing the supply and demand discipline. And this is, you know, as we look ahead at AI-related opportunities, we see in 2025 a tight supply environment, certainly contributed by the trade ratio of HBM as well.
We see a healthy environment for pricing. As we go ahead, all of these are important factors with respect to managing the profitability of the overall portfolio. But nonetheless, we remain extremely focused on getting the value of HBM, as we talked about in the future, with the customized HBM solutions adding even greater value, and we very much remain focused on delivering the financial model that we have discussed before.
You mentioned last night, in part of even that answer, that you had about $100 million or just over $100 million of HBM sales in fiscal 3Q, and you talked about several hundred for the year and billions, multiple billions, not to try to put words in your mouth next year. Overall, if it's already accretive, I would assume early stage yields will only get better from here. So two parts to the question. One, how are you feeling on yields? And if it's already accretive on that ramp, I would assume that it will be significantly more accretive, or is that an incorrect assumption as the LTAs that you have somewhat offset those cost downs that you would build in?
Manish will comment more on the yield part. I feel very good about our team's execution on ramp-up of capacity and yields, and I also feel very good about Micron's position with HBM for not only just this calendar year, but calendar year 2025 also. Not only the volume is sold out, but our pricing is largely determined for all of 2025 as well.
And that pricing, as we said, we see it being accretive to our overall gross margins. And again, Manish can comment on the yield aspects here. And regarding the several hundred million of revenue from HBM, I just want to be clear that that is in our fiscal year 2024, which ends tomorrow, and multiple billions of revenue in our fiscal year 2025. That's what we project. Manish, go ahead, comment on the yield aspect.
No, I mean, I feel good about, as Sanjay said, and we both feel good. I feel good about where we are with yields. I feel good about where we are with the production capacity ramp and the capability to be able to enable the targets that we set, in terms of, you know, the revenue we were going to execute on in fiscal 2024. And I feel good about the outlook for the rest of calendar 2024 and the ramp from here that enables a multi-billion-dollar ramp for fiscal 2025. Yes, Ross, you're right, yields will continue to improve, right? And that will continue to bring costs down.
And, you know, I think the other thing to keep in mind is we are projecting that our core DRAM margins will improve as well, and pricing environment will continue to be tight. So that means that, you know, we should see gross margin improvement across the board across our DRAM product portfolio. But definitely feel good about where we are and being able to deliver on the objectives that Sanjay outlined for next year, and feels great to have our customers this excited about a product so far in advance. It allows us to plan with confidence, allows us to focus around, you know, certain key customer applications and really deliver value for them.
And I would like to just underscore the point again that HBM has, of course, robust margins and accretive to the business. But the trade ratio that HBM has, that three-to-one trade ratio today, really contributes to the ripple effect on pricing in the rest of the business as well, and that's what really makes the overall DRAM supply environment tight and makes the overall DRAM pricing environment healthy as we look ahead to the rest of calendar year 2024, as well as to the calendar year 2025.
Do you think from that perspective, that people focusing on the bit shipment growth a little or the people are focusing on that a little bit too much and a little bit underappreciating the kind of aggregate pricing side for that dynamic as well?
I mean, clearly, the supply-demand environment is extremely important, and the pricing dynamic that we have discussed is extremely important with respect to the health of the business. And, overall, we see, you know, healthy, demand environment as we look ahead at 2025 across our various end market segments. And again, I think keep in mind that a penetration of AI-enabled edge devices will continue to grow, and those AI-enabled edge devices require more content. And, you know, we have talked about that AI-enabled smartphones and AI-enabled PCs will grow and will continue to grow within the mix of the total smartphones and PCs, and all of that is a favorable trend for demand for memory.
I'll just wrap up in the last couple minutes that we have on CapEx. So in your most recent quarter, you gave a lot more color on your outlook for CapEx next year. Can you remind us how you're thinking about balancing adding incremental fab capacity, adding clean room space, doing construction versus WFE spend?
So we said at our earnings call that our CapEx in our fiscal year 2025 will increase meaningfully versus fiscal year 2024. And what's important to just understand there is that we did say that of the increase in fiscal year 2025 on CapEx, a vast majority of that will be related to construction CapEx, as well as HBM assembly capacity CapEx. So that's an important factor. So the increase, vast majority, is going toward portions that are preparing us for not bit growth, but with HBM, you know, increasing the mix of high-value solutions, and with construction, addressing the capacity needs of the future.
For the latter part of the decade, technology transitions will not be sufficient by themselves in driving the bit growth requirements to meet the AI demand that is projected through this decade, and new capacity will be required, and that's where some of the construction CapEx will be going into. A nd of course, as we manage our CapEx, we remain extremely focused on maintaining supply-demand discipline. We follow the mantra of maintaining our overall bit share of DRAM and NAND as stable.
Of course, within that bit share, increasing the mix of the business toward higher profit pools of the industry, and extremely focused on ROI when we look at our CapEx decisions. When we put CapEx into construction of the facilities, keep in mind that WFE in those facilities will absolutely be very much driven by our assessments of the bit growth, demand growth in the future. We have a lot of flexibility in managing, continuing to manage our CapEx in a prudent fashion.
With that, I think we are right on time. So Sanjay and Manish, thank you so much for coming to the conference and sharing all your insights with us. It's gonna be a fascinating time at Micron, so thank you.
Thank you.