Micron Technology, Inc. (MU) Earnings Call Transcript & Summary

Thank you for standing by, and welcome to Micron's post earnings analyst call. [Operator Instructions] I'd now like to introduce your host for today's program, Mr. Farhan Ahmad, Vice President of Investor Relations. Please go ahead, sir.

Thank you, and welcome to Micron Technology's Fiscal Second Quarter 2023 Sell-side Callback. On the call with me today our Chief Business Officer, Sumit Sadana; our EVP of Global Operations, Manish Bhatia; and our CFO, Mark Murphy. As a reminder, the matters we are discussing today include forward-looking statements regarding market demand and supply, our expected risks, results and other matters. These forward-looking statements are subject to risks and uncertainties that may cause actual results to differ materially from the statements made today. We refer you to the documents we filed with the SEC, including our most recent Form 10-K and 10-Q for a discussion of risks that may affect our future results. I'll now turn the call over to the operator to begin the Q&A session.

[Operator Instructions] And our first question comes from the line of Vivek Arya from Bank of America.

I had a 2-part question. One is how far is Micron in terms of your cash cost in DRAM and NAND? And what do you think is the delta for your competitors right now? And then maybe one for Mark. Mark, even when I take your inventory write-down expectations for May, it still seems like the ending inventory would be $7 billion, $7.5 billion. That would still be higher than what you had exiting the last fiscal year. Is that a comfortable range to be in? Or do you think that still leaves open the room for more inventory write-downs in the future?

Let me start maybe and then go to the cost question. So the question of we're not comfortable with where inventory levels are and separate and apart from the write-down question, Vivek. Well, we do have elevated inventory levels. We expect, pre write-down -- the DIOs have peaked in the second quarter, and we would expect, over time, for the supply-demand balance to improve on -- customer inventories improving and volumes increasing sequentially and eventually working inventory levels down. But it's going to take some time. And we would like -- certainly, targets have not changed around 100 to 110 days of inventories on hand. It's, again, going to take some time. The question of write-downs is a function of the quantity of inventories and the cost, of course, but also pricing. So that's why we made the point in the call to indicate that write-down assumptions are sensitive to those factors, including price. And if prices in our outlook are better than expected, then we'll have less write-downs of inventory potentially even less in the third quarter. If inventory or price projections worsen versus our current view, we could have larger write-downs, for example, in the third quarter, maybe the fourth quarter. It just depends on the shape of the recovery, and we have a certain view, which is incorporated in the write-offs that we took.

Okay, Vivek, and then on your first question about cash costs, I can take it. This is Manish. So we've said that we are not yet at cash costs for either DRAM and NAND. So really, the underutilization actions that we've taken have really been around managing the overall supply-demand environment, but managing our inventory and trying to manage our supply to come down in line with the demand that we see. And with regard to competitors, I can't really comment other than to say, and we've given some disclosures in the past about how our technology leadership, 1-alpha and 176 last year has definitely helped us to become very competitive in costs. So we feel good about where we are and where our cash cost position would be relative to others as well.

And if I could quickly follow up. Mark, any -- I know it's never a simple question, but modeling your gross margins in the past Q3 is quite challenging. Any rough guidelines you could give to us so that we, at least, have some kind of boundary conditions on how to model gross margins for the next several quarters?

Yes. It's a good question, Vivek. And anticipating the question, I did, in the other call, make an effort to provide a profile on gross margin. And of course, as we just discussed, it's difficult because at these levels of profitability and the gross margins will be very sensitive to pricing assumptions and cost actions such as underutilization. So with that qualification, we did say, if you look at our reported margins, we would expect the 2Q to be the trough. And then on lower inventory charges, in third quarter, we expect a sequential improvement. And then in the fourth quarter, as mentioned, based on our current pricing assumptions, we would have a small or no write-off. And then, of course, as the underutilization charges, they weigh on as of fourth quarter and first quarter and then improve, gross margins to then improve through the year. That's on reported. Now if you strip out just the write-downs in the second quarter and third quarter, the second quarter would be 7.3% gross margin; the third quarter would be negative 7.5%. So clearly, that profile is down. And then -- and that's just a function of the pricing environment and, again, the cost of underutilization, which includes period costs. Now with this view of pulling those charges out, those write-downs out, under this view, we would trough in the second quarter and then we would begin to improve off these low levels as we get volume leverage on our period costs, as the underutilization effects play through, and then most importantly, as customer inventories begin to improve and inventories begin to come down and the supply-demand picture begins to improve. So clearly, the supply-demand balance and supply coming out of the system and in concert with volumes increasing sequentially is an important part of the recovery here.

Just wanted to make a quick clarification on the pricing comment as well. As Manish said, the pricing is not close to cash cost in the last couple of quarters, Q2, Q1. But in the guidance that we have provided in Q3 with the pricing that we have assumed there and the pricing trends we are seeing in the market, particularly NAND, definitely close to cash cost for Q3 and getting closer in terms of DRAM as well. But just wanted to highlight that cash cost is obviously different from variable cost. The variable cost is much lower than cash cost because certain cash costs are fixed in nature. So wanted to just provide that clarification.

[Operator Instructions] And our next question comes from the line of Vijay Rakesh from Mizuho.

Just a quick question. Just going back on the inventory side. As you look out exiting May quarter, given the incremental $500 million of write-downs, and I know you mentioned 25% wafer start cuts now and the CapEx tweaks, I guess. Would you expect inventory exiting the quarter to be closer to that $7 billion -- low $7 billion range? And I have a follow-up.

Yes. On a dollar basis, we would expect inventory levels to remain elevated for some time but begin to come down through '24. So -- but again, it depends on the nature of the recovery, and we do expect bit shipment volumes to increase through the year and inventory days certainly to -- on a pre write-off basis to come down, and then on a dollar basis, eventually begin to come down.

Got it. And then on the write-downs between the February and the May quarter, just wondering what would be the split between DRAM, NAND -- DRAM and NAND. And in DRAM, is the write-down that you've taken more on the DDR4 or DDR5 side since you'd expect some DDR5 demand to probably pick up into the back half? So might be some benefit from that, but any color there?

Yes. We just report these write-downs at the corporate level and given how we run the business on precedents and other factors. So we don't break it down by technology or by BU.

[Operator Instructions] Our next question comes from the line of Sidney HO from Deutsche Bank.

So I'm still trying to figure out the underutilization charges you are embedding in the fiscal third quarter guidance, trying to bridge the gross margin from Q2 to Q3, excluding that inventory write-down you talked about maybe down 15 percentage points. Is there any color you can provide just for fiscal Q3? I know you talked about $900 million for the fiscal -- for the full year.

Yes. We're just providing the full year. If we look at -- I just provided the full year. If you give me a few minutes, maybe go to the next question, I can maybe give you a little bit more detail, but let me work that out.

Okay. The reason I asked that, just to give you a background, is because if I look at that down 15 points, let's assume half of that $900 million is included in fiscal Q3, that's already 12 points of margin difference. So that implies the ASP has not really declined that much, but I'm not sure if I'm aware -- I'm missing. But maybe a second question, while you look at that, are you still expecting to have production cuts now 25% itself, 20% through the end of this fiscal year? Are there any plans to go beyond the current fiscal year?

Yes. Let me maybe answer your first question. I've already got the answer on the first one. So of the $900 million, not quite half of it is in the third quarter. And then we've got amounts split between some second and some fourth. And as it relates to -- we've not commented on utilization beyond '23. Sanjay mentioned on the call that we may. We'll have to evaluate it based on market conditions and so forth.

Yes. That's right.

Okay. So the plan is still through this year.

And Sidney, one thing I do want to clarify is that the underutilization charge that Mark mentioned, a portion of it is going to be in the write-down amount, including as well. So just -- like just -- that is a piece that you have to take into account in...

Yes. Just to be clear, and I think I mentioned this on the main call, that the reason the proportion is higher to the total underutilization charges because of that write-down accounting and its effect.

So part of that $500 million of write-downs is this -- is the underutilization charges?

Part of the $500 million would include utilization also, those higher costs. Because utilization -- underutilization creates higher cost inventory. So by definition, as you're writing that inventory down, part of that charge would be that underutilization-related cost.

Got it. Got it. Maybe a quick follow-up, if I may. If you exclude the inventory write-downs, can you say what a DRAM gross margin is negative in fiscal Q3? And at what point does it make sense to cut further production or completely stop production? Probably not realistic since you want to maintain your share.

Yes, I mean, we are not giving out specific gross margin breakout -- breakdown between DRAM and NAND, particularly with all of these charges that we are talking about related to underutilization, inventory write-down, et cetera. So definitely, we'll provide more color when we report results. But having said that, broadly speaking, our goal is to certainly keep flat bit share at customers and to not drive the pricing down to gain share just we had a [ follow-on ] on price, and we try to just keep our bit share flat and manage our business prudently.

Simply one more thing that Mark mentioned on the main call was that the gross margins, excluding the impact of the write-downs in Q3, was 7.5% negative. And obviously, the DRAM margins are a lot better than NAND. That is something that we can say, while we are not providing you the details and split between the DRAM and NAND gross margins.

And Sidney, just one other thing on the underutilization charge, which I mentioned. That's inclusive of costs that are absorbed in inventories and also period costs. And just keep in mind that if you're then translating what portion of that goes into the write-down, it's clearly only the portion that's related to the inventory costs. So just wanted to make sure that distinction was noted.

Our next question comes from the line of Harsh Kumar from Piper Sandler.

I was just curious, very roughly, what is the price difference between DDR4 and DDR5 right now that you guys are seeing? And then also maybe you could help us think about how much of your DRAM revenues is DDR5 right now? And how much do you think or when do you think the cross-over will occur for you guys?

Yes. So in terms of DDR4 versus DDR5, we still have a healthy premium between DDR4 and DDR5. DDR5 does sell at a premium. The premium is bigger in the data center than it is in the client space. And both data center and client DDR5 is going to continue ramping through this current and next calendar year, and we expect crossover to happen in both segments around mid-calendar 2024 from a bid perspective. The crossover will happen earlier on a revenue basis, but will happen around mid-2024 on a bit basis between the two.

[Operator Instructions] And our next question comes from the line of Ambrish Srivastava from BMO.

Mark, I know you've been very helpful in answering all these questions on the inventory. I had a couple of clarifications. At the beginning of your prepared remarks, you said this is not due to obsolescence. And then did I get it right that in the fiscal third quarter, there will be $300 million included in that from the $1.4 billion write-down?

Yes. The reference to the $300 million, Ambrish, was the $1.4 billion creates lower-cost inventories. And so the $300 million was in reference to the benefit associated with that lower-cost inventory sell-through. In other words, had we not written it down, we would have $300 million less income in 3Q versus since we pulled those inventory costs forward into 2Q, we now have $300 million higher income in 3Q.

Got it. So it's not necessarily it didn't turn...

That's separate, of course, from -- yes, the inventory charge in 3Q is the $500 million. That's an incremental inventory charge.

Got it. So the $1.4 billion and $500 million charges that you've taken now, we should continue to see benefit of that, right, if this inventory is not being -- is not going obsolete. You said there's no risk of obsolescence then.

Correct. It's permanently lower-cost inventory.

Yes. Got it.

Over the period that we sell.

Yes. And it will move through and pass on it as per the turns in the business.

Got it. And you will call it out explicitly, right, every quarter, then you get the benefit, correct?

I mean, we haven't decided, like we'll see on the proceeds.

Okay. And then on -- sorry, just on the minus 7.5 number that you gave for fiscal third quarter, is that inclusive of the benefit from the $300 million? Because you said strip out everything and fiscal 2Q quarter will be...

No. I just said that's stripping out the write-down.

Okay. Got it. Got it. And then I had a quick one for Sumit on the CapEx side. Sumit, this is a question I'm sure you get, we all get, what happens to your competitiveness as you push out your nodes? What's the right way to think about it in terms of specifically on the DRAM side?

I'll have Manish answer that question .

Sure. So Ambrish, I think we gave some good color on the call about the progress that we are making on our -- both of our new technology nodes that we announced last year, 1-beta for DRAM and 232-layer for NAND, and feel very good about what those nodes are going to be able to deliver in terms of bit per wafer gain over the prior nodes, both of which were also good nodes. So we feel really good about the intrinsic technology capability. And then now the yield ramps that we have been able to demonstrate are reaching targeted yields faster than prior nodes, really faster than any nodes in our history. So we feel very good about our ability to demonstrate technology capability. And now what we're preparing to do is to enable those -- both of those technologies across a broad portfolio of products. For example, in 1-beta, we talked about in our prepared remarks the LP5 part that's going to be generating revenue for us later this year, but we're also going to be utilizing that technology across DDR5 and compute, across high-bandwidth memory, graphics, et cetera. So we're kind of getting ready to use that really strong node across the portfolio. And similarly, in NAND, we've started in client SSD, and we'll move the 232-layer into mobile, into data center as we go. So we feel very good about that. And then as we look forward to the next-generation nodes, for example, 1-gamma, making very good progress on 1-gamma. And in particular, that's the node we're going to introduce, the EUV. And we've actually already taken delivery of, we've mentioned before, of our first-production EUV tools in Taiwan. We've actually already started using those tools, to some degree, on limited production in our 1-alpha. And it's already, within just a couple of quarters of having landed the tools, demonstrated the ability to match yields with our multi-patterning immersion technology on 1-alpha that's been in production now for a couple of years. So really, really good progress on EUV, and that gives us confidence that when we introduced 1-gamma, it will also be a strong introduction with good inherent technology capability for bit per wafer gain and cost reduction as well as it's going to ramp predictably and provide a strong cost reduction for -- across the portfolio.

[Operator Instructions] And our next question comes from the line of Karl Ackerman from BNP.

Two quick ones, if I may. Just -- but I was hoping you could discuss your China exposure by product category. My understanding is that you're primarily tied to mobile. But if you perhaps rank order the remaining segments, that would be very helpful as we think about the pending recovery in China.

Yes. I mean, we definitely have good dealer business in China. As you said, mobile customers, very strong center in China for mobile OEMs. We also have a very strong automotive business there. As you know, the EUV vehicle -- sorry, EV vehicle growth, not EUV. EV vehicle growth is really high in China. The penetration is very high. The uptake is very high in these cars, especially at the medium-to-higher end, have a lot of electronics, semiconductor components in them. So we have a good position. As you know, we are worldwide market share leaders in automotive. So we continue to have a good position there as well. And then, of course, we have customers on the client side as well as on the data center side, a broad range of customers. So we are pretty diversified across many different segments. And we expect that as the China recovery gains momentum after the reopening, we do expect that there will be some demand uplift coming from China later in calendar '23.

That's helpful. If I may, for my second question. You -- Mark, you indicated 2024 CapEx would decline from 2023. Perhaps maybe there's a question for Manish, but is it fair to suggest your implicit outlook is that CapEx will be all brownfield or upgrades? And then you would achieve projected bit demand, which should recover in fiscal '24 through the inventory on your balance sheet, unless maybe demand were to recover faster than you currently expect? Maybe just discuss that a bit more color, that would be very helpful.

Sure. So as we -- we have taken multiple levers to be able to reduce the supply. We've reduced CapEx and WFE in particular, as we talked about, more than 50% for FY '23. And we've said that for '24, it will be even lower. And we've also reduced utilization within our facilities. And then the last one is we've held inventory -- we're holding inventory on our balance sheet. So as we look forward into '24, the first thing we'll do is bring our inventories down, and that will help us to be able to meet demand and keep up as the market increases. The bit growth increases that we see -- that we talked about occur, our inventories will come down. And then as we start to feel comfortable that we're -- we have a trajectory to reach our inventory targets, of course, we can increase our utilization rates to be able to increase supply as well. And then the -- then CapEx for supply growth would be the kind of the next lever and the last lever. So really, right now, even in '23, the CapEx that we have, the WFE that we've put to work has been primarily -- actually, more than primarily, I'd say, exclusively for technology transition and technology learning on these 2 new nodes that I mentioned. And that will be what we'll be focused on, and even as we go through '24, is bringing the benefits of the new technology to the product portfolio, whether that's 1-beta or whether that's 232-layer.

[Operator Instructions] Our next question comes from the line of Mehdi Hosseini from SIG.

A couple of follow-ups. I'm just looking at quarterly CapEx trends throughout several cycles. And I come up with an estimate of $4 billion annualized CapEx just from maintenance, just what it would take to kind of run utilization rate at 70% to 80%. And actually, this is obviously making a broad guesstimate. But would -- is the $4 billion annualized maintenance CapEx in the ballpark?

I don't know that we've really kind of given a definition. I think different folks have different definitions about what maintenance CapEx involved. So I don't think we're really prepared. I think there's sort of...

Maintain a bit, but back to zero.

Yes. So if you think about the CapEx and the way we're -- the elements that we have, the big buckets are, of course, WFE, which primarily is for technology transitions. And when implemented the way we typically have, that does provide bit growth. So the technology transition, WFE does provide bit growth. That's the primary source of the bit growth that we would target within our supply. There's the facilities requirements that we have for implementing those technologies and adding the new equipment. And those tend to be a little bit lumpier, and you do those sort of well in advance of when you need them. And in fact, I think we mentioned that the mix of our CapEx in the second half of this year is moving more towards -- construction is becoming an increasing part of [ that in ] the second half of the year. In fact, more than double -- more than twice what it was in the first half of the year. So sort of starting to prepare a clean room space for the future. And then we have the assembly test and other R&D and other areas that we have that are -- to be able to either provide us with good cost reductions on -- for managing our assembly test costs or to provide R&D capabilities for the future. So these are kind of the big buckets that we typically define our CapEx side.

Yes. Mehdi, the CapEx is not really like a consumable. So it's more -- it is for technology transition, product enablement and advancing to more advanced nodes, which is giving you better costs and better production.

And more bits.

And more bits. So you shouldn't -- we shouldn't think that just for maintaining the bit production, we need to invest in CapEx.

Got you. Okay. I was just trying to get a sense of where net cash would go to as I try to forecast free cash flow. Okay. And then just looking beyond the current environment, obviously, there's a lot of moving parts with the write-off and underutilization charges. Let's say, February of calendar year '24 Q2 FY '24, and you have your, let's say, some of the DDR4 that are [ cost downs ]. And then DDR5, that is at a premium of the DDR4. If all equal, even if prices were to go sideways, given this inventory adjustment, if prices were like stable, you should be able to have a better margin profile, excluding all the onetime charges.

Yes. Mehdi, we're not providing specific guidance that far out. And we will, as we get closer to the date and the market situation becomes clearer. We did give you a profile on how to look at the back half of this year and expected improvement on gross margin through '24, assuming that all these factors we laid out move in the direction that we expect. And as you point out, particularly if you're talking about second quarter '24, some of the write-downs that would occur in third quarter, that lower cost inventory would be passing through in the second quarter of '24.

Can I ask a clarification question? Do you mind if I just have a quick follow-up?

Go ahead.

In your previous comment -- [ this is regarding the ] mix here. As you think about the DDR5 premium for both PC and server applications, would that be -- that premium, would that be off of a market pricing for DDR4? Or would that be off of a cost-adjusted DDR4 that you have?

Yes. I mean, I think a lot depends on the market environment and competitive behavior. But generally, obviously, DDR4, as you know, versus DDR5, DDR5 tends to be a bigger die, more expensive, at a module level as well because it has further integration at the module level compared to DDR4. So It's higher ASP, but also higher costs. So those are some of the factors that come into play. And like you said, I mean, the DDR4, DDR5 transition is going to be taking place, and there is more inventory in DDR4 than there is in DDR5. So a lot more DDR5 deployment in the future comes in as new purchases from our customers. That's another thing to look at. But I think, broadly speaking, I think you have a number of data points that we have provided to you, which include our bits bottoming in fiscal Q1 of this year, our before write-down days of inventory peaking in fiscal Q2, our data center revenue bottoming in fiscal Q2, customer inventory is improving. So when you zoom out of all of this, we've also told you that looking at 2025 calendar year, we expect record TAM in that year. So we do expect that looking out multiple quarters and into '25, there is going to be a very substantial amount of growth that we are expecting in that time frame in the -- particularly in the late '24, '25 time frame, and obviously, continuing to focus on significant improvements in profitability and trajectory of free cash flow in the interim from here to '24.

Got it.

One other thing, our 1-beta -- you asked about DDR5. Our 1-beta technology is optimized for DDR5 in terms of the architecture and the technology. So we feel very good about what that will eventually position us for as the market transition.

[Operator Instructions] And our next question comes from the line of Brian Chin from Stifel.

Maybe a 2-parter on sort of demand. I don't think too many people have really asked about that. But was the lion's share of the reduction in the calendar '23 memory bit demand forecast, is that related mainly to data center? What is your also underlying assumption for data center bit demand growth in calendar '23? And then sort of the second part of that is, does your expectation for a mid-single-digit decline in the calendar '23 PC TAM, is that a sellout or a sell-in figure? I ask because sell-in sounds like it could be weaker than down mid-single digits. And I think sell-in ties more into kind of activity levels at OEMs.

Yes. I mean this is definitely a good question, and we've focused on lot of these dynamics with our customers on an ongoing basis. So in the last 3 months since our last earnings call, we definitely have been continuously assessing, as we always do, the level of inventories our customers have and its impact on demand for the near-term planning over the next several quarters. And so the degradation in our calendar 2023 outlook for DRAM and NAND came as a result of the assessment of the inventories, particularly in the data center, and also the pace of progress of inventories. So the inventories have been improving. So it's a matter of the rate and pace of that improvement that we had to adjust for in our TAM, but also the extent of the pullback in growth estimates for both PC units as well as smartphones. So smartphones going from a low single-digit increase to low single-digit decrease in units in calendar '23 over the last 3 months. And on the PC side, yes, mid-single-digit decrease, that puts the PC units -- and this is more of a sell-through comment, puts the PC units at a level that is very consistent with the level they were at pre-COVID in 2019 from a global PC unit sales perspective. So that's how we think about the PC business. And certainly, different customers have different amounts of channel inventory. So their own sell-in versus sellout rates will be different. And our demand expectations are based on a bottoms-up view of what we are feeling our customers' sellout is going to be because that's going to -- sorry, sell-in is going to be into the channel because that's going to determine the consumption of DRAM and NAND.

Okay. Yes. Okay. Okay. Great. That makes sense. Maybe just a quick thing tying in technology kind of [ separate ]. But in terms of mix, I know you won't want to be too specific about percentages and crossover, et cetera, but when next year, what do you kind of plan for wafer inputs to really begin to shift toward 1-beta-and 232-layer technologies? I know it may become more observant once you kind of add up some of the utilization charges and et cetera, right, FIFO inventory, et cetera. But I'm wondering if there's sort of a -- kind of a particular point in next year that you kind of anticipate not to start to pick up some inertia.

It's too far out, Brian, right? Like at this point, I don't think it's worthwhile. But like you said, as demand picture improves, we will have more clarity on it. And we just have to see time play out and as we turn back under utilization, and that is currently being implemented. We will also ramp on 1-beta at that time when the demand improves. .

[Operator Instructions] And our next question is from the line of Srini Pajjuri from Raymond James.

Mark, I'm just trying to figure out the cash flow for next quarter. Obviously, you got some help from working capital in the February quarter. So if you could maybe help us if you're expecting somewhat similar benefit from working capital in the next quarter or 2, I think that will be really helpful.

Yes. I think, listen, on free cash, I think it's helpful to maybe step back and -- this has been such a sharp and sudden downturn. And we know all the factors, and we've been aggressively, and with even greater intensity through the fall, taking actions, CapEx reductions, cutting operating spend, lowering utilization, and you even heard some additional actions announced today. So we've also, over the past several months, used our balance sheet to bolster liquidity and make sure that we're making the best long-term decisions for the business. And so that's -- that sort of gives us some backdrop as to where we are. I mean, our actions are helping improve the financial picture from what it would be otherwise. But it's clear that we need the broader market to recover and, more specifically, more supply to come out of the system. But on free cash flow, we do expect free cash flow to improve from here, slightly at first going second to third quarter as reductions in CapEx are partially offset by weak operating cash flows. We have low volumes and challenging pricing, so that's weighing on operating cash flow as the benefits of the receivables have waned and the inventories remain elevated. Over time, though, we see free cash flow improving sequentially by sustaining our capital discipline and on, importantly, improving operating fundamentals. Specifically, we do see shipments continuing to increase from here, as we've talked about, customers replenish inventories, inventories begin to decline across the industry and channel, supply-demand balance to improve and pricing to revert to more sustainable levels. It's not sustainable where it is. So between our actions and a healthier industry, we see fundamentals improving. And we're certainly focused on returning to positive free cash flow within quarterly positive free cash flow within fiscal '24.

Got it. But it doesn't look like you're assuming your inventory levels to decline on an absolute basis in the May quarter.

Right. I mean the sort of elevated inventory levels are, in part, contributing to this free cash flow use that we have, and we're certainly reducing CapEx and spend and other things to improve the situation, and we do expect slight improvement in free cash flow in the third quarter relative to the second, but it is still negative significantly.

Got it. Got it. And 1 question on the mechanics of the underutilization, maybe for Manish. As we go through the next few quarters and as demand starts to recover, just curious if you expect pretty much all of the 100% of the underutilized capacity to come back. Or at what point some of the capacity becomes permanently impaired or move to the next node, I guess? And then as you bring that back, is there any incremental CapEx that we should be aware of?

Srini, I'm glad you asked that. It's actually definitely one of the considerations that we have when we think about underutilization is, as we implement these new technology nodes, we can utilize some of this idled equipment to help us be capital efficient as we implement the next generation nodes, right? So that's always a balance that we're thinking about. I wouldn't think of it in terms of any sort of impairment. I don't think that we're in that window. Our reuse nodes, node-to-node is very good. So I wouldn't -- I don't think that idled equipment ends up getting impaired. That's kind of a first-order consideration. But the ability to utilize some of the equipment to be a little bit more efficient as we transition to new nodes because, as we mentioned, new nodes could provide new product capabilities, whether that's higher performance products or, as we mentioned, DDR5 capability or others, we're trying to make sure we balance the whole picture on utilization, CapEx and demand for new technologies across our portfolio.

I guess as you bring that capacity back online, is there any incremental CapEx, Manish, that we should be aware of in the short term?

Not -- I mean, not specifically for bringing online, but it depends on how we -- I mean, we have multiple options of how we bring that capacity online in order to implement new technology nodes or to just replace the capacity that we had taken down before.

This does conclude the question-and-answer session as well as today's program. Thank you, ladies and gentlemen. You may now disconnect. Good day.