Hey, well, I guess we can get started. Good morning, everyone, and welcome to day one of the Deutsche Bank Tech Conference. I am Sidney Ho. I cover semiconductor, semi-cap equipment, and IT hardware at Deutsche Bank. Our first company today is Micron Technology. Micron is, as you all know, one of the largest memory suppliers in the world. Today, we are very excited to have Micron CFO, Mark Murphy, as well as EVP Global Operations, Manish Bhatia, joining us today. So, Mark, before we get into Q&A, why don't you give us a brief update as to what you are seeing in the marketplace, maybe highlighting things that may have changed since you provided guidance on, I think, June twenty-eighth?
So great to be here, Sidney, and thank you, everybody, for joining us today. I'll start with the safe harbor. I'll be making, or will be making forward-looking statements. Those statements have risks and uncertainties associated with them. I refer you to the risk factors disclosed in our public filings. So the business is trending in line with the expectations that we shared at our earnings call in June and a recent conference. Industry bit shipments continue to increase, and the supply response has been broad-based. On industry bit shipments and our bit shipments, yeah, we're past the bits bottom. And in fact, we expect to have record bit shipments in this quarter. On pricing, we're at an inflection point. We see pricing bottoming about now.
As we've said, you know, before, we expect in calendar 2025 for the industry to be record TAM, and that is with more normalized profitability as well. You know, on other items, on July 26th, we did announce that we began sampling our HBM Gen 2 product. Can also refer to that product as HBM3E. You know, subsequent competitor announcements show that, you know, Micron will have a product that leads on specs when it's ramping. On August 7th, also supporting data center, we announced a memory expansion module product both uses and supports the CXL standard. And then finally, most recently, we did complete CHIPS grant filings for a fab in Idaho and a mega fab in New York.
So with that, we'll take the rest of the Q&A.
Okay, great. Well, yeah, thanks for that update. Maybe just a few follow-up questions on what you just mentioned. You talked about price inflection, expecting price decline, inflection in the second half, and in fact, today you updated to be just about now. Can you add some color as to the scope and magnitude of the price increases that you're expecting? Are you seeing that both in DRAM and-
We are. You know, the market is playing out as we had expected. You know, severe downturn beginning last summer, broad-based supply response as a result of that downturn that has helped begin to stabilize inventory levels and thus pricing. And so now we are with inventories in a better position in many markets and volumes continuing to increase. We are seeing now, you know, first it started in pockets, but now we say generally, we're seeing pricing bottom.
Okay.
We expect that to, you know, pricing to strengthen through the second half of the calendar year, you know, assuming these supply discipline holds and also that the volume continues to expand.
Okay, great. Maybe on data center recovery, you talk about data center inventory will likely normalize around the end of the calendar year or somewhat thereafter. What gives you confidence? Is that still the right timing? What gives you the confidence that this time is actually happening?
That inventories will normalize?
Exactly.
Well, I mean, so far, again, I would just point to, you know, our view on how this has played out. You know, we have seen broad-based supply response. We are seeing inventories at lower levels in many markets. PCs and smartphones, for example, have been in pretty good shape for some time. So, as demand picks up in those markets, customers will replenish their inventories. Data center, which had more elevated levels of inventories, we do expect those inventory levels to be in a good spot in the end of this calendar year, maybe beginning of calendar 2024. So, you know, and so then the drawdown of inventory should happen at the producer level, and that is happening.
We are seeing, you know, our DIO continue to drop, and see that trajectory of improving inventories, you know, through the rest of this year into next year.
Got it. Okay, go ahead.
Just a little bit on that. I think Mark's point about playing out the way we saw it. I think we've been pretty consistent about talking about this. We were probably the first in the industry to see this coming, to take action in terms of supply. We saw this starting with mobile phones, you know, during the lockdowns in China in spring of 2022. So inventory started to build in mobile, consumer space, then eventually moved over to data center later. And the recovery in inventories has played out as we expected, starting with the consumer space. Mark mentioned mobile and PC segments now, relatively normal, heading into the seasonally strong part of the year. So I think, you know, we sort of have seen this movie before.
We understand how the cycles play out, and I think we have a pretty good read on the fact that data center will get better next. And the other part to keep in mind about data centers, it's not just one, you know, amalgam of. It's an amalgamation of a bunch of different technologies. There's D4, there's DDR5, there's. We're going to talk about high-bandwidth memory in a few minutes. There's even some LP that's going into data centers. So there's different markets, and the inventory positions are different there. I think as the market moves more and more towards some of these newer, more advanced solutions, like DDR5, that will definitely, you know, also help with the recovery and demand from the data center.
Great. That's great. That's good. You guys talk about DDR5, and looks like things are picking up. Big growth is growing well. What is the situation with DDR4 in terms of inventory in general?
Well, the DDR4, there's more supply in DDR4 than DDR5, and there is a transition underway. So, you know, DDR4 will continue to be used and will continue to be used for, you know, a long time. But we are seeing this transition to DDR5, and we're in a great position there. We've got DDR5 right now in our 1- alpha. We're transitioning, our next DDR5 products will be on 1- beta, which with Manish here, we can go into detail on that process technology is outstanding. But this mix shift into DDR5 is occurring, and data center, we saw DDR5 as a percent of DRAM shipments more than double Q2 to Q3.
And then we expect DDR5 volume to cross over DDR4 at the end of the first calendar year, 2024, or first quarter, calendar 2024, versus mid-2024 for the rest of the industry. So we feel great about our position, and the migration of data center to that latest.
Okay.
Yeah, I'll just add that Mark mentioned it, 1 -beta is really going well for us. You know, clear leadership in the industry, you know, mature yields already across the board and you know, great power, great performance, specs across the portfolio that we're leveraging it in. It was really uniquely defined architecture and a lot of the specs around D5. So, as the industry moves towards D5, we think it's going to really be a differentiator for us in terms of our ability to address the D- you know, DDR5 opportunities, as well as LP5 opportunities, HBM opportunities, and GDDR7 eventually as well.
Got it. I do want to talk about HBM in a second, but before we get into that, for NAND market, it seems like it's consensus that things are a little slower compared to DRAM. Can you give us an update on that market? And it sounds like pricing is also starting to improve, but are there pockets of strength and weaknesses that we, you guys are looking?
Yeah, the NAND market, no, no real update on the NAND market. You know, the profitability and pricing is, is worse than DRAM. There's higher inventory levels than NAND. DIO is higher. Yeah, we are seeing volumes increase there, and there is supply response as well. So, so we do, we do believe that, the same sort of recovery is underway, but it will, will take longer given worse situation.
Okay. And last questions before I open up to the audience. I want to ask about the China, the CAC restrictions. Has there been any more clarity or visibility in terms of what's being restricted? Could there be more negative surprises there? And talk about your progress in terms of also flexing up share at some of the key customers that you guys talk about in the past.
Yeah, no update on the CAC as well. I would refer folks to our latest 8-K, which made no additional update beyond that in earnings. It's a sensitive situation with a lot of uncertainty, but you know, we have sized what the potential risk is, and we're obviously working to mitigate that risk. And we're having good engagements with customers and others to try and mitigate impact. You know, we've been in China a long time. And have served customers very well in China, provided a lot of value to the Chinese semiconductor ecosystem and Chinese consumers. So, you know, we look forward to continuing to supply the market and are working, of course, under the current strain.
Okay, great. Maybe a pause here to see if there's any questions on the floor. If we do, we can run the mic over to you.
Yeah. Ian Murray, thank you. Mark, listening to you for the last year, this is a nice change in tone at the beginning of the conference. But, I have two questions. One is, you know, we all hear about HBM and, and the, AI servers, and I'm wondering how much DDR5 is required for the AI servers. You know, what, to what extent is that a, a bump up? And secondly, could you talk about your, HBM solution? You know, the general consensus now is that Hynix has the best solution, and they're taking most of the market. You're just at 5% or something, and, and they have this stacked with the liquid in between. What, what is your answer to that? When does it come? You know, where do you expect the market share to end up, please? Thanks.
Maybe I'll start.
I'll take the first one.
Yeah, I think start and then she build on it. So, it certainly is a better tone than this time last year. I mean, we were headed, you know, in the midst of, you know, the steepest part of the downturn. We knew it was happening, but many others didn't. So we were, you know, lonely acting early at times. But we acted early and so are in a good position today, as a result. And so today, the recovery has started. The supply-demand balance is better. Pricing, as we've talked about, we believe, has bottomed. So, you know, we look forward to the future. An exciting part of that future is data center, of course, and the products serving that market.
Yeah, your question was good because, you know, serving that market takes a suite of products that we think we're very well-positioned in. So, you've got high-capacity DIMMs, you've got the standard DDR products. We've talked a lot about DDR5. LPDDR, so Low Power DDR is being introduced in the data center or in servers. So, you're seeing various combinations of these products being used to best or to optimize the performance of the servers. And then finally, you've got, of course, HBM, which has a lot of attention, and we're really excited about that product. It's built on we believe the best technology in the business and our 1 -beta process technology. So, we've announced that we've begun sampling.
We're very pleased with what we see. The customer engagement's very deep, and we will be ramping that product in calendar 2024, and we think we'll have meaningful revenue on that product by the second half of the calendar year. And, you know, we expect to have on the HBM3E, we will have, you know, better than our share of the overall DRAM market in that, in that particular product. So, we believe.
Yeah. So, Ian, you're right that today, Hynix and, you know, does have a strong position relative to the existing solutions for both HBM2E and the first generation of HBM3. Our product is really a leapfrog with HBM3E. The performance is 50% higher than today's HBM3 solution, and even relative to recent product announcements by competitors, our performance is higher than that. Our power, we believe, will end up being lower than competitive offerings in HBM3E as well, 2.5 times lower than previous generation on a power-per-bit basis. So, we really think our product will be differentiated and will be an industry leadership platform. As Mark just mentioned, we expect to be able to have very strong share performance within HBM3E going forward. Now, how did we do that?
The foundation is the technology and the fact that we have leadership with 1 -beta well ahead of competitors. Typically, we scale to new technology nodes because we're trying to drive cost. In this case, it does give us a cost advantage, but beyond that, it's able to... Because of the performance of the device and the size of the memory array, we're able to actually have higher capacity die. Our HBM3E is higher capacity than existing HBM3 solutions, so we can get more memory capacity closer to the processor, because of the ability to shrink using the 1 -beta technology into the standard package. We're also able to provide, you know, better performance, and power trade-off because we have a lower power inherent in the technology.
Then the other, I think, big differentiator is we're so far ahead on 1 -beta. We are mature, shipping in high volume across, you know, other products, have all that learning that will go into our HBM process flow, as we ramp and take that and grow that revenue, as Mark mentioned, in calendar year 2024. So I think that 1 -beta really gives us a lot of the fact that we're so far ahead, and it's such a great technology foundation, gives us the confidence to be able to build a strong product offering on top of that. Now, on top of that, we do have a lot of great design features and packaging features that we're pretty confident in, are going to be able to provide, you know, enable that leadership, and enable us to ramp successfully, next year.
And we're investing with confidence. We mentioned on our last call that we're based on the confidence and the strong response we've seen from customers to our early samples and our specifications, we're increasing investment for next year, our plans for investment next year, for advanced packaging capacity, to be able to meet this demand.
Yeah, maybe I'll just a follow-up question. You talk about HBM ramping, sampling right now, ramping in calendar 2024, and meaningful revenue in the second half of 2024. HBM is a little different than the normal DIMM in terms of design cycles, right? So, you probably have to work with your, partners a little earlier. Can you talk about that process? How long does it take versus a regular DIMM? And once you've designed into an HBM, is it, is that something that your customer can switch in and out just based on prices, or pretty much locked in for that?
I'll talk about it. I think that you're right, Denis. It's a longer process, not just because the product is such a high-performance product, that it requires a lot of tuning by the customer on their side to be able to optimize the solution, but also because it's a multi-party process, as you're aware of. Because there's the foundry that makes the silicon, as well as does the integration for our customers combined with our process. So it's a you know more complex overall integration.
Yes, it takes longer time, but it does create a, I think, a more of a customized solution in terms of the tuning that we work together with to be able to optimize the solution for the, you know, for the end application. And that tuning happens on both sides, and tuning happens in terms of the way the system interoperates, as well as even just in some of the manufacturing areas that each side needs to address to be able to allow this very, very high-performance solution to be optimized.
Okay. Any other questions on the floor?
Hi, Amy Sutter from Deutsche Bank. I just wanted to get some color about, to the extent you have it, on what you mean about calendar 2025. I mean, I think this was a bigger downturn than anyone expected, obviously, and coming out of COVID, there's a lot of not great visibility into, the supply chain in general. But look - when I look at what's happening with HBM and the market opportunity there and all the investment in data center, is there a chance that we can get back to previous margins? Like, how, to the extent you are willing to talk about the margin framework as we come out of this cycle, would be really helpful.
Yeah, we're not going to comment in any sort of detail on margins out that far. What we can say is that, you know, on the demand side, as you point out, there are very strong growth drivers in the space. And, you know, I think the best example is that, you know, we're at record bit shipments today. You know, emerging from a downturn, we're at record shipments. So that just points to the strength overall of memory and the need for these products. And those growth drivers are durable, so automotive will be continuing to grow and a lot of content there.
Smartphone and PC, you know, they've been in the doldrums for some time, but eventually there'll be a cycle, replacement cycle on those, and that will recover. And then, of course, you've got data center, which is currently, you know, demands exceeding expectations on AI-related server purchases, training and inferencing. But eventually, you know, the broader server, you know, compute infrastructure will need to be, you know, built out, refreshed. So again, very durable growth. So on the demand side, volumes, you know, we feel very good about, you know, the multiyear trend in the business. Now, on the supply side, you know, the response has been very strong. All the competitors have announced reductions. They're all minding inventories.
They've all cut CapEx. We've all reduced utilization, sort of actions that are unprecedented and appropriate given the downturn that we experienced. But it's a positive sign on the industry to deal with this because, you know, the pricing is not sustainable at these levels. We cannot invest in the needed capacity and technology developments required to drive the rest of the technology space with pricing where it is. So pricing, as it's beginning to now, should recover, and that will, of course, help margins. So we're, you know, we're investing appropriately for that good future in the business, where volumes are higher, pricing is better, cash generation is stronger.
So we'll provide more on near-term margins and then, you know, more on the longer term in the business, the earnings.
I'll maybe just add a little bit on the supply side there. I think, yes, the industry did have, you know, you know, added too much production capacity relative to demand signals we saw through the pandemic-induced, you know, increase or inflation in demand signal that we were getting. But that doesn't change the fundamental driver that technology is getting harder and harder, both in DRAM and NAND side. Technology is getting harder; cadences are more challenging to be maintained in terms of new technology introduction. CapEx still continues to grow per node, you know, node on node, to be able to transition capacity. So that's, you know, that's a stabilizing factor, I think, for long-term bit growth in the supply side. And then high-value solutions. You know, we talked about DDR5, we talked about HBM.
You know, those are both solutions where the bits per wafer, bits per die, are more, are lower, and in HBM's case, significantly lower than on a standard D4 or LP4 technology. So as the industry transitions to these higher value solutions, that I think will also be a stabilizing factor as we look forward in terms of the pure bit supply growth capability for the industry.
There's one more question.
Yeah, just looking forward to 1-g amma, are you going to continue to use EUV sparingly, as in as few layers as possible? Or do you think there is a need to kind of build hard and build out a very large fleet of EUV tools? And to what extent is your footprint ready, to sort of receive those tools, because you have a relatively older footprint than some of your competitors. Thanks.
Sure. First of all, I think we really hand it to our technology development team, you know, for making the decision to be able to push EUV out to our 1-g amma node. I think, you know, that's one of the reasons that our, I mean, our immersion-based multi-patterning solutions for both 1- alpha and 1-b eta have performed, you know, superbly. I think, you know, we will. I think our lithography expertise is excellent. Our judgment in terms of which layers need EUV, which ones we can, you know, optimize better with multi-patterning is very good, and I think we'll implement it well within 1-g amma there as well.
In terms of the readiness, what we've talked about already, well, we've mentioned this publicly before, that we've already taken receipt of production tools in Taiwan, where EUV will be first introduced on 1-g amma in 2025. We've already started exercising those tools on our 1 -alpha just to demonstrate that we understand how to and to come up the learning curve on the new materials and you know processes that are needed to be able to be successful with EUV. We're already matching yields on 1 -alpha EUV layer against multi-patterning layers. So, we feel very good about our learning rate, and when we introduce it, we'll be you know successful.
In terms of our ability to receive it, you know, we have new fab space that we've built in Taiwan. We have new fab space that we've built in Japan, both of those within the last five years. And when we built that new space, we enabled EUV with that. That space was built with EUV in mind. So we were very, we're ready for, you know, however many EUV tools we need to receive to be able to grow in 1-g amma and into the future.
Okay, well, maybe I have a few more questions on the product side. Clearly, a lot of interest in the HBM side of things, but you've also talked about high-capacity DIMM, the 128 GB, I think that's what you're referring to. Maybe you also talk about LP being deployed in data center. How is Micron positioned, and how do you differentiate that with competitors that have similar solutions?
Sure. So, you know, I think high-capacity DIMMs are, you know, certainly becoming a larger part of the market and, you know, exciting part of the growth in AI servers. So, we actually already have an offering. Mark mentioned our DDR5 1-alpha solution today is on 1-a lpha. We have a 96 GB offering, that's already, you know, shipping in volume and production to customers.
As we transition to 1 -beta, you know, we expect to have a 128 GB offering that will be competitive, and well, in terms of capacity, it'll be matching what others have in the industry, but again, benefiting from the 1 -beta—all the strong capabilities we have on that 1 -beta technology to be able to provide our customers benefits on power and performance. I think the difference in our strategies between the strategy between us and our competitors is that we've targeted to have monolithic die, high-capacity solutions. We think that those optimize costs as well as power and performance better than the multi-die stacking that competitors are currently using to get to the 128 GB capacity point.
So, our 1 -beta offering, which will be next spring, which will be coming out next spring, will be a monolithic solution, again, based on 1 -beta, and we're able to do that because of our leadership and time-to-market with 1-b eta and our ability to get that larger monolithic die onto, you know, into the package for the DIMM solution. So, we feel very good about ours. And that was sort of some of the thinking, you know, about what solutions we introduce when, because we saw that our 1- beta was going to be such a great node, well ahead of competition. We, you know, yes, there are some high-capacity DIMM spots that we're not in today. We think our 96-gigabyte solution is helping us in a lot of them.
When we get to our 1- beta solution next year, we think we'll be very well positioned for the 128-gigabyte technology or 128 gigabyte module capacity.
Great, thanks. Maybe I just wrap it up with a couple of financial questions for Mark. Does the improvement in DDR5 and AI demand change the way you think about CapEx for fiscal 2024? What is the... Maybe just to give us a sense, what's the range of supply growth that you can do and without increasing that you can support in fiscal 2024, calendar 2024, whatever time?
So, there's no update to what we said on fiscal 2024 CapEx this time. You know, it's a continual process, assessing how to most optimally deploy capital, how to spend our R&D, how to spend CapEx at the right time, at the right place, and the right amount. And that is ongoing. So, for example, on the last call, we mentioned that, you know, we're increasing our HBM-related CapEx spend because we see the market need is immediate. We have a great product, so we need to be able to supply that to market. We've not changed our CapEx. We're looking for, you know, other ways to get that CapEx out, push things, whatever we can do to be mindful of that.
I mean, the fundamentals of the business is still weak, and so we're being very thoughtful about what we spend and getting returning to free cash flow growth. You know, we did say that next year we do expect WFE to be down year-over-year. And so, you know, we'll give. We're about a month out from earnings, so we'll give a more complete update at that time.
Okay, great. Maybe just one more last short-term question. On gross margin, do you still expect gross margin on the reported basis not to turn positive until second half, or is it things are improving a little bit, that outlook is a little better?
Yeah, again, we're about a month out from earnings, and we'll give a more complete update at that time, including profile of gross margin spend, but or gross margin. But we do, you know, as we said on the last call, we do see gross margin on a reported basis, continuing to, you know, become less negative and then eventually turn positive in 2024. As I mentioned at the opening, you know, you know, right now, the business is trending in line with what we shared at earnings and in the last call. Yeah, so the volume and pricing trends are consistent with that, and so therefore, we would expect to do as good or better than what we had did before.
Okay, great. Just to close, what are some of the key messages that you may want investors to take away from today? What are some of the areas that you think investors may have underappreciated with regards to the Micron story?
Just quickly, then Manish can add. I think investors should take away that, that we responded very well during the downturn, and, you know, in a disciplined manner and in many ways led the industry, and, and we, you know, minimized the financial impact of what is, you know, maybe the worst downturn in the industry's history. And we did that in a way that we continue to invest for the future, and we're very excited about the future. We talked a lot about it today, and we couldn't be more pleased with the position we've got, and, of course, are working to capitalize on, on what we view as a very, very positive trends in the space.
Just to add, I mean, I think the technology and product portfolio of the company has never been stronger. I think we talked a lot about 1- beta today, but 232-layer NAND also, both 1 -beta and 232-layer NANDs, you know, fastest programs in our history and positioning us very well for many different solutions, high-value solutions. We talked about DDR5, we talked about HBM, we didn't talk about data center SSD, but we're already on 232-layer shipping data center SSDs. So the technology product and our manufacturing ability to deliver those leading-edge technologies at scale, I think has never been stronger. And you know, I think the underappreciated part, I think, is still the multi-cycle growth story, right? Yes, we have cycles. This one was deeper than most.
But when you look over the long term, out towards the end of the decade, the fundamental demand drivers of AI, 5G, and autonomous are still there, maybe even building more steam than before. And when you look across cycles, I think, you know, the memory space is going to continue to outgrow the rest of the semiconductor.
Great. Well, we are out of time, so thank you, Mark and Manish, and enjoy the rest of the conference.
Thanks. Thanks, everyone.