Good morning, everybody. My name is John Vinh.
Semis here at KeyBanc Capital Markets. We're pleased to have Micron with us this morning. We've got Mark Murphy, CFO, and Samir Patodia from Investor Relations. Welcome, guys.
Thanks, John.
Thanks, John.
Great. Hey, Mark, maybe you can just kick us off and maybe just give us an update on what are some of the key demand trends you're seeing across your end markets right now.
Sure. Happy to do that, John. So first, let me start with safe harbor. I'll 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. The NAND and DRAM markets overall are on an upward trajectory towards what we believe will be a substantial revenue record for Micron in our fiscal 2025, and with significantly improved profitability. Micron is in the strongest position it's ever been in with our technology leadership and product portfolio. As far as our markets in data center first, data center demand remains very robust, driven by AI server units and also a recovery in traditional server units. Micron has a tremendous portfolio strength in AI server products. So we're especially well positioned in this market.
If we go to PCs and smartphones, as we talked about on our earnings call, PC and smartphone customers built inventories for three reasons. One, they were getting out ahead of price increases in the industry for memory and storage. And they also are positioning themselves for anticipated growth in AI PCs and AI smartphones. And then lastly, there is an expectation that the memory complex remains very tight, especially on the leading edge, and any incremental supply is going to be directed towards data center. So they are buying out ahead of that. As noted in broader industry reports, automotive, industrial, retail, consumer markets have weakened. In China, economic activity and consumer buying patterns are weak or uneven at best. It depends on the market. We forecast bit shipments to be flat sequentially in the November quarter for both DRAM and NAND.
On gross margin, sequentially, August to the November quarter, we expect gross margins to be up around 200 or up around a couple of hundred basis points, consistent with what we've said before. Maybe lastly, give an update on our HBM ramp. Our HBM is progressing very well on the ramp. Our yields are trending well. Our capacity and output is on track to deliver what we expect to be several hundred million of HBM revenue in this fiscal 2024, to be multi-billion HBM revenue in fiscal 2025, and then achieve our share target, which is to have HBM consistent with our DRAM share at some point in calendar 2025.
Great. Thanks for that update, Mark. Maybe we could just touch base on kind of November bit shipments being flat. That seems maybe slightly below kind of normal seasonal. What are the assumptions there? Is that a function of some of your customers buying ahead from an inventory perspective? And I know you kind of walked us through kind of the key end markets. Can you also touch base on where you think your customers' inventories are across servers, PCs, and smartphones?
Sure. Sure. Maybe start with what we are doing is price is going up. We have price going up in DRAM through 2025. And so where we're seeing that price is not as robust as we would like, we're walking away from deals. So we're seeing that, and we're just taking a very disciplined approach. And we know that we can sell that product for more later in the coming months. And we also need that inventory to supply our expected demand in 2025. I'll talk about that in a moment. As far as inventories at our customers, we had talked some time about data center inventories, particularly traditional data center servers. So that inventory improving in the first half of calendar 2024. And that's happened. So data center inventories are healthy. We're seeing the resumption of traditional data center growth in the back half here of the calendar year.
In PCs and smartphones, as I mentioned at your first question, we do see those inventories were built up in the past quarter. Again, for what we think are healthy reasons. One, they see prices increasing in the memory space, wanted to get out ahead of that. They do anticipate growth in AI PCs and smartphones, which are, as you know, more memory-intensive, memory-rich. And then they see the tight market, and they want to make sure they have supply. On our own inventories, I think what's important is that we are ramping HBM. And we've talked about before that there's a significant amount of silicon required to ramp that product. We see a 3:1 trade ratio. So that is going to limit our bit supply growth in 2025. So we believe that our bit supply growth will be less than demand in 2025.
We will need the inventories that we exit this year and enter next year. We'll need those inventories to serve the market in 2025. We've sort of been in this 150 days on DIO, which we believe will be around there for the next quarter or so. Then we will begin to decline through second, third, fourth quarter of fiscal 2025, and then eventually reach what we believe will be our target inventory levels, which we target about 120 days.
Great. You mentioned your AI PC customers building ahead of a potential AI cycle there. How are you thinking about just the AI PC cycle and AI smartphone cycle this year? And maybe can you just talk about what you guys expect in terms of the increased content loadings there versus non-AI PC and smartphone products?
Yeah, certainly data center has gotten a lot of the attention. The demand has been very strong. Product portfolio is very rich. But the edge is an exciting place for memory and storage on AI at the edge. And there's a lot of very compelling consumer features that we think will be introduced. And we're seeing that now as we work with customers. And I think it could be very helpful to a refresh cycle, which is sort of long in the tooth at this point anyway. It may help strengthen that cycle. We'll see. But as far as AI-enabled devices, there's a significant uplift in memory content. We've said that in a smartphone, we see 50%-100% more DRAM than a non-AI device. On a PC, we see 40%-80% more. I think the minimum to run a Windows Copilot+ , I think, is 16 GB.
Significant uplift from standard Windows.
Great. Are there any questions? Great. So Mark, it sounds like you guys are making some good progress in terms of improving your yields on HBM, and I appreciate the update there. Can you just clarify kind of the gross margin expectations for HBM through this cycle? I know in the just reported quarter, you had said, "Hey, HBM gross margins are already above DRAM gross margins, above corporate." Should we expect that to be the case throughout this entire cycle?
Yeah. So you're right, John. We already indicated our first $100+ million of revenue that we had in the third quarter of HBM. That was accretive out of the gate to the corporate, of course, and also DRAM margins. So it's a great product and yielding well enough already to be accretive to the business. As I said in the first question, our yields continue to trend well, too. So as we're ramping, we're seeing, obviously, we believe that an accretive gross margin will continue. Now, we see as this sharp ramp happens, you're actually going to see our cost downs get impacted. So I just want to point that out. And actually, you see some in the first quarter, our cost downs will actually be flat to up. But our previous guidance of muted cost downs for 2025 still applies. And we will ramp this product.
We're investing to get our share. Then we expect it to be accretive. It's a higher-priced product, of course, because the costs are higher. But it's also a very differentiated product and impacts the performance of our customer system. So we would expect that on average or most of the time to command a premium to other products. And the cadence on these HBM products appears to be following the customer's annual cadence at this point. So we'll have HBM 3E right now, 8-high stack. We expect next year for there to be 12-high stack, and we're sampling on that. And then the year after, we'll have HBM 4. So extremely difficult product, getting more difficult each generation, meaningful performance differences and impact on the customer systems. It helps that Micron is a technology leader in this case.
Great. I think there were some headlines kind of overnight that potentially some of your customers' products could be delayed. I'm just wondering from your perspective, are you seeing any sort of changes in timing and scheduling of the ramp of your HBM3E products?
We work closely with the customers, and obviously, we can't comment on something that they've not stated. Our product is sold out, so there's no change in our plans. We're ramping. Our ramp is on track. Our yields are trending well. We're sold out. Pricing is largely set through 2025. So we just continue to ramp and work to our share target in 2025.
Maybe I want to talk about kind of your target for next year on HBM3E, where you talked about getting to market share equivalent to your traditional DRAM share. Maybe can you just talk about what your assumptions are for getting there? More specifically, I think we're hearing about concerns about Samsung struggling to get qualified on HBM3E. If they were to continue to struggle there, would there be a potential opportunity for you to upside on some of those targets for next year?
First of all, our investment assumptions and all the statements we've made, I can't comment on Samsung. We assume that Samsung will be in the market and successfully deploy products. So that's our assumption. And our assumption, of course, as we've stated, is consistent with the rest of our business. We work to maintain stable bit share. And in this case, our share aspiration or our share comments on HBM are that we expect to be consistent with the rest of our DRAM business and be in the low 20%. So that's the comment there. The HBM market, it's about 4% of industry bits this year. It's likely to be about 5% of industry bits next year. We've said that the TAM in 2025, we expect to be over $20 billion.
And so that lines up with the comments we've made about several hundred million this fiscal year, multibillions next year, next fiscal year, and then our share. And so we'll see that ramp. But it's an exciting market. It's growing quickly. We are in a great position. And of course, there's all the supply-related dynamics that are positives as well.
I know, Mark, you're probably obviously working very closely with your customers in terms of kind of securing agreements for next year. I think you said they're largely secured. If your customer is keen to you and asks for more bit capacity next year on HBM3E, can you just talk about what your ability to respond and to react to that is?
We're focused on getting in the equipment required for our HBM capacity. So the assembly and test equipment for HBM. And we're in a steep ramp, and we're working through that ramp, and it's going well. Beyond what we stated as our plan and share target, there's limited ability to upside on that. We're doing well while we're doing. And then also on the front end, keep in mind that there's this 3:1 trade ratio. So if we do more HBM, then that's going to take bit capacity from other markets. And so we're tight on the front end, and we're putting in place this HBM-specific capacity. And then you heard on the call the greenfield capacity we need to do. So Micron, like the others in the space, the capacity is very tight.
The last thing I would say on your question of upside is, again, I want to reiterate that we have our share target for HBM is this. We're investing and planning to have that HBM share be consistent with our DRAM share.
Great. Maybe you can touch on HBM4. I think you've indicated that you think the trade ratio increases. We've heard that potentially could go as high as 5- 6 to 1 versus HBM3E. And I'm just wondering, what do you think are kind of the longer-term impacts to the memory industry as this trade ratio continues to increase?
Yeah. We've not quantified HBM4 trade ratio. We've just said that it will be meaningfully higher. So it's getting more challenging. In HBM3E, the die are bigger, so you have some yield fallout there. The stacking of the die creates additional yield fallout. So you have this significant amount of silicon capacity required to produce bits. And if you look at, you're going from 8-high to 12-high, and then there's complexities in HBM4 that would drive that trade ratio even higher. So I think it's obviously a big deal in the market. It's a contributing factor, obviously, to the tightness in the market. And that's going to be a persistent issue with the space for the next several years because AI-driven growth will continue for many years.
I think what's noteworthy, just maybe a little bit of a sidebar on this, listening to the earnings calls this cycle on the big four spenders on data center, we all know they've increased their spend from $150 billion last year to over $200 billion this year. And they'll be probably $250 billion or more next year. What's interesting is, I think it was Microsoft mentioned, half of that spend is only half of that spend is for actual servers, that half of it's for greenfield and space. So that bodes well for all the server-related space available and continued spend. So I think this is a multi-year build-out that will consist of server refreshes and new servers and the space for those new servers. And I think that bodes well for HBM and the broader memory industry.
Great.
You talked about walking away from deals based on price. Is there any deal that you actually cannot walk away from that you have to do versus ones maybe you developed for types that you would walk away from?
I mean, every situation is different, and there's commercial decisions to make. We have customers or partners, and we want to make sure we're meeting their needs. But on the other hand, if we don't think the price is reflective of the value we're bringing, if we don't think the price is robust enough given the market conditions, we're going to walk from that business because we believe that in the coming months, we'll have the ability to sell that product. And we also know that given our outlook on the business, we need inventories to supply us through 2025 until we have sufficient production capacity. So we feel we can both meet the customer needs and also maintain our price discipline, and that's what we're striving to do.
We'll have time for one more question.
Okay.
Mark, maybe just last question from me is just, can you give us an update on how things are progressing with EUV and incorporating that into One Gamma?
It's very much on track. I mean, as I think we've said, and you just mentioned, we'll insert it into 1-gamma. We already have tools in place, a few of them. And they're already actually incorporated into our flows and older technologies. And so we're already yielding on the equipment. We know how to run it. I've personally seen it. It's pretty impressive. So we'll insert it at the right time, which for us is 1-gamma. And we'll work to optimize that in that deployment.
Great. Looks like we're out of time. Thank you, Mark.
Yes, sir.