Q4 2025 Micron Technology Inc Post Earnings Call

After the speaker's presentation, there will be a question and answer session to ask a question. During the session you will need to press star one on your telephone to remove yourself from the queue. You May Press Star one again I would now like to hand, the call over to Satya Kumar Investor Relations. Please go ahead.

Thank you and welcome to Micron technologies fiscal fourth quarter of 2025 post earnings analyst call on.

Satya Kumar: We are under no duty to update any of the forward-looking statements to confirm these statements to actual results. With that, we can now open up for Q&A.

On the call with me today are <unk>, Chief business Officer, <unk> EVP of global operations and Mark Murphy our CFO.

Operator: Thank you. As a reminder, to ask a question, you will need to press *11 on your telephone. Please limit yourself to one question and one follow-up to allow everyone the opportunity to participate. Please stand by while we compile the Q&A roster. Our first question comes from the line of Tom O'Malley of Barclays. Please go ahead, Tom.

As a reminder, the matters. We're discussing today include forward looking statements regarding market demand and supply market trends and drivers and unexpected results and guidance and other matters. These forward looking statements are subject to risks and uncertainties that may cause actual results to differ materially from statements made today.

I refer you to documents that were filed with the SEC, including our most recent Form 10-Q and upcoming 10-K for a discussion of risks that may affect our results. Although we believe that the expectations reflected in the forward looking statements are reasonable we cannot guarantee future results levels of activity performance and achievements, we're under no duty to update any of the forward looking statements to conform these statements Dr.

[Analyst]: Hey, guys. Thanks for taking my question. My first one is just on the state of industry of NAND. Over the last couple of weeks, you've heard a lot of reports about hyperscalers stepping in. You've heard about tightness, pricing increases. If you look at your current quarter that you just reported, you had bits down in the quarter. I've tried a couple of different ways here. Tell me if I'm crazy, but it seems like you need to have bits down again in your guide to kind of get there. Largely driven by ASPs. Is that the right way to think about the world right now? Some of your peers are talking about some big growth going into the back half. Is that something that you're going to look to ease up with potentially some capacity additions?

Results with that we can now open up for Q&A.

Thank you as a reminder to ask a question you will need to press star one one on your telephone please limit yourself to one question and one follow up to allow everyone. The opportunity to participate please standby, while we compile the Q&A roster.

[Analyst]: Anything that you can comment on to the state of the world right now. Thanks.

Our first question comes from the line of.

Sumit Sadana: Yeah, hi. This is Sumit. I'll just take the question. In terms of the NAND industry, I wouldn't read too much into the bits for FQ4. It's really just noise based on a few things, different segments moving around in our mix. It's really just some mix-driven noise for the various segments. In terms of the demand profile and what is shaping up looking ahead, certainly the large hyperscalers are looking like they will be needing significantly more storage capability for their AI server deployment because they have shortages on that side from the HDD segment of the market. The deployment of NAND SSDs in servers and data centers more broadly is going to increase in calendar 2026. We expect, driven by that tailwind for the NAND industry conditions, to start improving.

Tom O'malley of Barclays. Please go ahead Tom.

Hey, guys. Thanks for taking my question. My first one is just on the state of industry of NAND. So over the last couple of weeks you've heard a lot of reports about hyperscale are stepping in and you've heard about tightness pricing increase and then if you look at your current quarter that you just reported you had been down in the quarter and then I'll try it a couple of different ways here tell me, if I'm crazy, but it seems like you need to have.

It's down again, and your guide to kind of get there.

So largely driven by the USPS is that right way to think about the world right now and like some of your peers are talking about some big growth going into the back half. So is that something that youre going to look to ease up with potentially some capacity additions just anything that you can comment on the state of Vermont.

Yes, hi.

This is Samantha I'll just take the question.

So in terms of the land industry.

Great too much into the.

Bits for FQ4 it's really just noise.

Based on just a few things on different segments.

Moving around and our mix. So it's really just some <unk>.

Speaker #1: Thank you for standing by and welcome to Micron Technology's post-earnings analyst conference call. At this time, all participants are in a listen-only mode. After the speaker presentation, there will be a question-and-answer session. To ask a question during the session, please press *1 on your telephone.

Driven noise for the various segments.

In terms of the.

Demand profile and what is shaping up looking ahead.

Sumit Sadana: Certainly our position in the data center SSD market, where we have been hitting record share after record share for several years now and have a really strong position in the market for us to benefit from that data center SSD position, competitive position that we have. We have announced a lot of new products. We have a very exciting portfolio. We expect to be able to leverage that going forward. I think the NAND industry will improve. The improvement in the DRAM business is definitely ahead from a time perspective in terms of being tight. NAND is improving and getting tighter, but DRAM is tight already today and getting even tighter going forward.

Certainly.

Speaker #1: You will need to press *11 on your telephone to remove yourself from the queue. You may press *11 again. I would now like to hand the call over to Satya Kumar, Investor Relations.

The large hyperscale dealers.

Our.

Looking like they will be needing a significantly more.

Our storage capability for their AI server deployment.

Speaker #1: Please go ahead.

Speaker #2: Thank you, and welcome to Micron Technology's fiscal fourth quarter 2025 post-earnings analyst call. On the call with me today are Sumit Sadana, Micron's Chief Business Officer; Manish Bhatia, EVP of Global Operations; and Mark Murphy, our CFO.

Because they have shortages on that side from the HDD.

<unk> segment of the market.

So the deployment of NAND, Ssds and servers and data centers more broadly is going to increase in calendar 'twenty six.

Speaker #2: As a reminder, the matters we're discussing today include forward-looking statements regarding market demand and supply, market trends and drivers, and our expected results and guidance in other matters.

We expect driven by that tailwind for demand industry conditions to start improving.

Speaker #2: These forward-looking statements are subject to risks and uncertainties that may cause actual results to differ materially from statements made today. We refer you to documents that we have filed with the SEC, including our most recent Form 10-Q and the upcoming Form 10-K for a discussion of risks that may affect our results.

And certainly our position in the.

The data center SSD market, where we have been hitting record share after record share in.

For several years now and have a really strong position in the market for us to benefit from that data center SSD position.

Speaker #2: Although we believe that the expectations reflected in the forward-looking statements are reasonable, we cannot guarantee future results, levels of activity, performance, and achievements. We are under no duty to update any of the forward-looking statements to confirm these statements to actual results.

Mark Murphy: Yeah, Tom, it's Mark. I would just add on the supply side to continue. You know, we structurally brought down wafer outs, and NAND would continue to slow node transitions there and pace new node ramps. Still have low levels of CAPEX and still are working down inventories there.

Competitive position that we have and we have announced a lot of new products, we have a very exciting portfolio.

Speaker #2: With that, we can now open up for Q&A.

So we expect to be able to leverage that.

Speaker #1: Thank you. As a reminder, to ask a question, you will need to press *11 on your telephone. Please limit yourself to one question and one follow-up to allow everyone the opportunity to participate.

Manish Bhatia: Yeah, I was going to add, we're working down inventories. Now with our decision to exit managed NAND, there'll be more supply availability for us to focus on the data center market, as Sumit was mentioning.

Going forward, so I think the NAND industry will improve with the improvement in the DRAM business is definitely.

Speaker #1: Please stand by while we compile the Q&A roster. Our first question comes from the line of Tom O'Malley of Barclays. Please go ahead, Tom.

[Analyst]: Helpful. Just as a quick follow-up, if I look at the CAPEX guidance, you're saying kind of annualized Q1, you get to $18 billion. I just want to make sure you're saying that's a net number. Obviously, something higher in aggregate. In your commentary in the slides, it seems like very little additional NAND spend and more on the DRAM side. At a very high level, I know you're not guiding specifically. Should we be taking away from this that you're moving from $13-something to close to $20-something and that incremental is largely a function of additional DRAM spend? Thank you.

Our head from our time time perspective.

In terms of <unk>.

<unk> tight.

<unk> is improving and getting tighter, but DRAM is tight already today and getting even tighter.

Speaker #3: Hey guys, thanks for taking my question. My first one is just on the state of the industry of NAND. So, over the last couple of weeks, you've heard a lot of reports about hyperscalers stepping in, you've heard about tightness, and pricing increases. If you look at your current quarter that you just reported, you had bids down in the quarter. I've tried a couple of different ways here to tell me if I'm crazy, but it seems like you need to have bids down again in your guide to kind of get there.

Going forward, yes, Tom it's Mark I would just add on the supply side continue.

Structurally brought down wafer outs in NAND.

Continue to slow node transitions there than in pace, a new node new node ramps.

Mark Murphy: We're just going to give in that number, Tom. We're saying that we're going to go from $13.8 billion net in 2025 to, you know, a net approximately $18 billion in 2026. The vast majority of that is for DRAM construction and equipment.

Still have low levels of Capex, and so were working down inventories there so were.

Speaker #3: So, largely driven by ASCs, is that the right way to think about the world right now? And, like, some of your peers are talking about some big growth going into the back half, so is that something that you're going to look to ease up with potentially some capacity additions?

Were working down inventories and now with our decision to exit managed NAND there'll be more supply availability for us too.

Focus on the data center market estimates with much.

Speaker #3: Just anything that you can comment on to the state of the world right now. Thanks.

That's helpful. And then just as a quick follow up if I look at the Capex guidance, Youre, saying kind of annualized Q1.

Speaker #4: Yeah, hi. This is Sumit. I just take the question. So, in terms of the NAND industry, I won't read too much into the bids for FQ4; it's really just noise, based on just a few things, different segments.

[Analyst]: Thank you, Mark. Appreciate it, guys.

The $18 billion.

Operator: Thank you. Our next question comes from the line of Joseph Moore of Morgan Stanley. Please go ahead, Joseph.

Want to make sure Youre, saying Thats, a net number so obviously something higher in aggregate and in your in your commentary in the slides it seems like very little additional land spend and more on the DRAM side. So just on a on a very high level I know youre not guiding specifically should we be taking away from this that you are moving from 13, something close to 20 something in that incremental is largely a function.

[Analyst]: Great, thank you. Sanjay, early on in the call, talked about taking market share in HBM in calendar 2026, if I heard him right. I know that would sort of happen mechanically because you got to your target share towards the end of this year. You would gain share. Can you just talk about what are your share gain aspirations in HBM, if any? Are you content with holding your DRAM share? Does that need to go higher?

Speaker #4: Moving around in our mix, so it's really just some mix-driven noise for the various segments. In terms of the demand profile and what is shaping up looking ahead, certainly the large hyperscalers are looking like they will be needing significantly more storage capability for their AI server deployment.

Additional DRAM spend thank you.

Yes, we're just kind of given that number Tom So we're saying that we're going to go from 13, eight net 25 too.

Yes.

Net approximately $18 billion in 'twenty six and the vast majority of that is for DRAM construction and equipment.

Sumit Sadana: Yeah, in terms of our HBM share, you're right. We have ramped our share through calendar 2026, and we expect to be in the vicinity of our DRAM supply share in calendar Q3 in 2025. In 2026, for the full calendar year, we will expect to have higher share in HBM compared to calendar 2025. HBM has now become a very robust part of our portfolio, and we treat it like any other part of our portfolio where we focus on ROI of the portfolio, we focused on discipline investments, we focused on getting to be the best from a product perspective and creating that value with our customers and then ensuring that we are taking full advantage of the competitive landscape as it evolves over time. Those are some of the things we do.

Thank you Mark I appreciate it guys.

Thank you.

Our next question comes from the line of Joe Moore of Morgan Stanley. Please go ahead Joe.

Speaker #4: Because they have shortages on that side from the HDD segment of the market, the deployment of NAND SSDs in servers and data centers more broadly is going to increase in calendar 2026.

Great. Thank you.

Sanjay early on in the call talked about taking market share in HBM in calendar 'twenty six if I heard him right.

But I know that would sort of happen mechanically because you got to your target share towards the end of this year. So you would gain share can you just talk about what are your share gain aspirations in HBM. If any are you content with holding or do you Im sure does that need to go higher.

Speaker #4: So we expect driven by that tailwind for the NAND industry conditions to start improving. And, certainly our position in the data center SSD market where we have been hitting record share after record share in, for several years now, and have a really strong, position in the market, for us to benefit from that data center SSD position, competitive position that we have.

Yes in terms of our <unk> sure.

You are right I mean, we have ramped our share through calendar 'twenty six and we expect to be.

In the vicinity of our.

DRAM supply share in calendar Q.

Speaker #4: And we have announced a lot of new products. We have a very exciting portfolio, so we expect to be able to leverage that going forward.

Q3.

In 2025.

Sumit Sadana: We're not going to provide any more specifics beyond that, but that's how we are thinking about it.

In 2026 for the full calendar year, we will expect to have higher share in HBM compared to calendar 'twenty five.

Speaker #4: So, I think the NAND industry will improve with the improvement in the DRAM business. It's definitely ahead from our time perspective. In terms of being tight, NAND is improving and getting tighter, but DRAM is tight already today and getting even tighter.

[Analyst]: Okay, that's helpful. Thank you. Will the gross margins in HBM continue to be higher than, you know, sort of non-HBM DRAM? If given that you're locked in through most of 2025 and DDR5 is moving up, is it possible that that changes? Does that lock-in actually hurt you if DDR5 prices rise faster?

HBM has now become very robust part of our portfolio.

And we treat it like any other part of our portfolio, where we focus on ROI of the portfolio will be focused on disciplined investments. We focused on we are focused on getting to be the best from a product perspective, and creating that value with our customers and then ensuring that we have to.

Sumit Sadana: Yeah, I mean, I don't think about the lock-in as "hurting us" because the two businesses, the HBM portion of it and the non-HBM DRAM, are just different types of businesses. The HBM business, we are working with customers with long lead times, long order visibility, lock-in of volumes, lock-in of pricing well ahead of time, a lot more stable ROI over the years. That kind of a profile is helpful because no matter what the rest of the business does, our expectation is that the HBM business will have higher through-cycle ROI. We manage that business in that different way. It creates a lot of value for our customers, and we have that expectation of a higher ROI for us in our portfolio.

Speaker #4: going forward.

Speaker #5: Yeah, Tom, it's Mark. I would just add on the supply side, continue to, you know, we structurally brought down wafer outs and NAND, continued to slow no transitions there and, and, and pace, new node, new node ramps.

<unk>.

Full advantage of the competitive landscape as it evolves over time.

So those are some of the things we do so we're not going to provide any more specifics beyond that but that's how we're thinking about it.

Speaker #5: We still have low levels of CapEx, and we're working down inventories there.

Speaker #1: Yeah, I was going to add, we’re working down inventories, and now with our decision to exit managed NAND, there will be more supply availability for us to focus on the data center market, as Sumit was mentioning.

Okay. That's helpful. Thank you and then the gross margins in HBM.

Will that continue to be higher than <unk>.

Non HBM DRAM and if given that you are locked in through most of 2025 and DDR five is moving up is.

Speaker #3: Helpful. And then just as a quick follow-up, if I look at the CapEx guidance, you're saying kind of annualized Q1, you get to $18 billion. I just want to make sure you're saying that's a net number, so obviously something higher in aggregate.

Is it possible that that changes is that could that lock in actually hurt you if DDR five prices rise faster.

Speaker #3: And in your commentary in the slides, it seems like very little additional NAND spend and more on the DRAM side. So just on a high level, I know you're not guiding specifically. Should we be taking away from this that you're moving from the 13-something to close to the 20-something, and that incremental is largely a function of additional DRAM spend?

Yes, I don't think about the lock in as.

Quote unquote hurting us because the two businesses the HBM portion of it and the non HBM DRAM.

Sumit Sadana: Now, will there be times when the rest of the DRAM portfolio will become very profitable and can challenge or even exceed at times the profitability of the HBM business? Of course, that is possible. We don't think of that as a bad thing. We think of that as a positive thing that the whole business has become dramatically better. That's how we think of it as, you know, those outcomes are a good place to be because that means the whole industry is extraordinarily healthy from a supply-demand balance perspective, and that helps us drive really robust financial outcomes.

Just different types of businesses the HBM business.

Speaker #3: Thank you.

Speaker #4: Yeah, we're just going to

Speaker #5: give a net number, Tom, so we're, we're saying that we're going to go from 13 to 8 net in 25 to, you know, a net approximately 18 billion in 26 and the vast majority of that is for DRAM construction and equipment.

We are working with customers with long lead times long order visibility locking the volumes locking of pricing well ahead of time.

A lot more stable.

Over the years, so that kind of a profile is helpful. Because no matter what the rest of the business does our expectation is that the HBM business will have higher through cycle ROI. So we manage that business and that different way it creates a lot of VAT.

Speaker #3: Thank you, Mark. I appreciate it, guys.

Speaker #1: Thank you. Our next question comes from the line of Joe Moore of Morgan Stanley. Please go ahead, Joe.

Speaker #6: Great. Thank you. Sanjay Mehrotra, early on in the call, talked about taking market share in HBM in calendar 2026, if I heard him right. But I know that would sort of happen mechanically because you got to your target share towards the end of this year.

For our customers and we have that expectation of a higher ROI for us in our portfolio now will there be times when the rest of the DRAM portfolio will become.

[Analyst]: Very helpful. Thank you.

Operator: Thank you. Our next question comes from the line of Jim Schneider of Goldman Sachs. Please go ahead, Jim.

Speaker #6: So you would gain share. Can you just talk about what your share gain aspirations are in HBM, if any? Are you content with holding your DRAM share?

Very profitable and can challenge or even exceed that times the profitability of the HBM business of course that is possible and we don't think of that as a bad thing. If we think of that as a positive thing that the whole business has become.

[Analyst]: Good evening. Thanks for taking my question. Two quick ones. One follow-up on Joe's question, which is, you know, just in general, given all the performance benefits you talked about relative to your position on HBM4, would you expect your HBM4 share to be higher than it was for HBM3 in a sustainable way going forward?

Speaker #6: Does that need to go higher?

Speaker #4: Yeah, in terms of our HBM share, you're right. I mean, we have ramped our share through calendar 2026, and we expect to be in the vicinity of our DRAM supply share in calendar Q3 2025.

Dramatically better and so.

That's how we think of it as those outcomes are.

<unk> a good place to be because that means the whole industry is extraordinarily healthy from a supply demand balanced perspective, and that helps us drive really robust financial outcomes.

Sumit Sadana: I mean, we do feel we have very strong capabilities entering the HBM4 business. We certainly believe that we have the best HBM3E product in the world with the highest performance and 30% lower power consumption than any other competitor. What we did not have at the start of the HBM3E business ramp, though, is that we didn't have that installed capacity and that base capability. We were ramping from very low levels. Now, with the very strong performance from our operations team, we have been able to get our capacity, our yields, our performance, our quality to extremely good levels. We have met our market share goal that we had articulated many quarters ago to you. We are starting off our HBM4 in a very different position in terms of being able to meet the expectations and requirements of our customers from a volume support perspective.

Speaker #4: So in 2026, for the full calendar year, we expect to have a higher share in HBM compared to calendar 2025. HBM has now become a very robust part of our portfolio.

Very helpful. Thank you.

Yeah.

Thank you.

Our next question.

Comes from the line of Jim Schneider.

Goldman Sachs. Please go ahead Jim.

Good evening. Thanks for taking my question two quick ones one follow up on Joe's question, which is.

Speaker #4: And we treat it like any other part of our portfolio, where we focus on ROI of the portfolio. We focus on disciplined investments; we are focused on getting to be the best from a product perspective.

Just in general given all of the performance benefits you talked about relative to your position on the HBM. Four would you expect your HBM for sure to be higher than than it was for HBM three and in a sustainable way.

Speaker #4: And creating that value with our customers and then ensuring that we are taking full advantage of the competitive landscape as it evolves over time.

Going forward.

I mean, we do feel we have very strong.

Speaker #4: So those are some of the things we do. We're not going to provide any more specifics beyond that, but that's how we are thinking about it.

Capabilities entering BHP and for business.

We certainly believe that we have the best <unk> product in the world with the highest performance and 30% lower power consumption than any other competitor what we did not have at the start of the <unk> III E business ramp, though is that we didn't have that installed capacity.

Speaker #6: Okay, that's helpful. Thank you. And then the gross margins in HBM—will that continue to be higher than, you know, sort of non-HBM DRAM?

Sumit Sadana: We feel that from a competitive positioning, that aspect of the business is in much better shape. Now, with HBM4, there has been a lot of chatter in the market and the media about what is our HBM4 performance compared to those who are doing the base die, the foundry, etc. We just wanted to mention and lay some of those doubts to rest that we today believe that our HBM4 has the highest performance amongst any competing product out there. We don't believe that others can match this performance with this level of capability, quality, power consumption, and so on. We believe that there'll be tremendous customer preference for our products. Of course, we have increases in our HBM supply that we have created through our investments. We have confidence that we will sell out that supply for calendar 2026.

Speaker #6: And, you know, if given that you're locked in through most of 2025 and DDR5 is moving up, you know, is it possible that that changes?

And that base capability.

Speaker #6: Is that because that lock-in actually hurt you if DDR5 prices rise faster?

We're ramping from very low levels and now with the very strong performance from our operations team, we have been able to get.

Speaker #4: Yeah, I mean, I don't think about the lock-in as 'hurting us' because the two businesses, the HBM portion of it and the non-HBM DRAM, are just different types of businesses.

Capacity, our yields our performance or quality to extremely good levels. We have met our market share goal that we had articulated many quarters ago to you and.

Speaker #4: The HBM business, we are working with customers with long lead times, long order visibility, lock-in of volumes, and lock-in of pricing well ahead of time.

And so we are starting off our H being forward in a very different position from in terms of being able to meet.

<unk>.

Expectations and requirements of our customers from a volume support perspective.

Speaker #4: A lot more stable ROI over the years. So that kind of profile is helpful because no matter what the rest of the business does, our expectation is that the HBM business will have higher through-cycle ROI.

So we feel that somewhat competitive positioning.

That aspect of the business is in much better shape now with HBM for.

It is.

It has been a lot of chatter in the market in the media about what is our HBM for performance compared to those who are doing the based on the foundry et cetera.

Speaker #4: So we manage that business in a different way. It creates a lot of value for our customers, and we have that expectation of a higher ROI for us in our portfolio.

And we just wanted to mention and lay some of those doubts to rest that we today believe that <unk> four has the highest performance amongst any competing product out there and we don't believe that others can match this performed.

Speaker #4: Now, will there be times when the rest of the DRAM portfolio will become very profitable and can challenge or even exceed at times the profitability of the HBM business?

Sumit Sadana: There may be some HBM share shifts happening on the competitive landscape, but it is our belief that most of that is most likely going to be between our two competitors rather than impacting our share in any material way.

Speaker #4: Of course, that is possible. We don't think of that as a bad thing; we think of that as a positive thing that the whole business has become dramatically better.

With this level of.

Capability quality.

[Analyst]: Thank you. As a follow-up on the capital expenditure (CAPEX) side, could you maybe give us some color on the split of WFE versus facility CAPEX you saw in fiscal 2025, and whether we should expect any significant change in that mix heading into next year? Thank you.

<unk>.

Power consumption.

Speaker #4: And so, you know, that's how we think of it as, you know, those outcomes are a good place to be because that means the whole industry is extraordinarily healthy from a supply-demand balance perspective.

And so on and so we believe that there'll be tremendous.

Customer preference for our products of course, we have.

Increases in our HBM supply that we have.

Speaker #4: And that helps us drive really robust financial outcomes.

Created through our investments and we have confidence that we.

Mark Murphy: Jim, it's Mark. We're not going to provide that breakout, just that the vast majority of the spend is for DRAM construction and equipment.

Speaker #6: Very helpful. Thank you.

We will.

Sell out that supply for calendar 'twenty six.

Speaker #1: Thank you. Our next question comes from the line of Jim Schneider of Goldman Sachs. Please go ahead, Jim.

There may be some HBM share shifts happening on the competitive landscape, but.

[Analyst]: there any significant change in the facility piece of that for next year, though?

Speaker #7: Good evening. Thanks for taking my question. I have two quick ones. One follow-up on Joe's question, which is, you know, just in general, given all the performance benefits you talked about relative to your position on the HBM4, would you expect your HBM4 share to be higher than it was for HBM3 in a sustainable way?

It is our belief that most of that is most likely going to be between our two competitors rather than impacting our share in any material way.

Mark Murphy: We're not going to comment on that. As you know, we're expanding in a number of sites, including, we talked about our first fab in Idaho, and we're also equipping those fabs as well.

Thank you and then as a follow up on the on the Capex side could you maybe give us some color on the split of WSB versus facility Capex you saw in fiscal 'twenty five and then whether we should expect any significant change in that mix heading into next year. Thank.

[Analyst]: Okay, thank you very much.

Speaker #7: Going forward?

Operator: Thank you.

Mark Murphy: I should say equipping the fab and then also doing tech transitions.

Speaker #4: I mean, we do feel we have very strong capabilities entering the HBM4 business. I mean, we certainly believe that we have the best HBM3e product in the world, with the highest performance and 30% lower power consumption than any other competitor.

Thank you.

Manish Bhatia: You mean, sorry, just to be clear, you're saying in Idaho, we're in construction. We expect production there.

Jim It's mark.

We're not going to provide that breakout just debt.

Mark Murphy: Second half of 2027.

Manish Bhatia: Second half of 2027. We are investing in technology transitions and optimizing, you know, production across our footprint in Japan and Taiwan, where we have available clean-up space.

Yes, the vast majority of this spend is for DRAM construction equipment.

And any any significant change in the in the facility piece of that for next year.

Speaker #4: What we did not have at the start of the HBM3e business ramp, though, is that we didn't have that installed capacity and that base capability.

We're not going to comment on that we're as you know we are expanding in a number of sites, including we talked about our first stab in Idaho and we're also.

Operator: Thank you. Our next question comes from the line of Chris Danely of Citi. Please go ahead, Chris.

Speaker #4: We were ramping from very low levels. And now, with the very strong performance from our operations team, we have been able to get our capacity, our yields, our performance, and our quality to extremely good levels.

[Analyst]: Hey, thanks, guys. Can you hear me okay?

Flipping those.

Mark Murphy: Yeah.

Sumit Sadana: So far.

Apps as well.

[Analyst]: Thank God. I've been having phone problems all day. Quick clarification. As far as the DRAM revs go, you guys gave us the HBM breakout. Can we assume, you know, roughly the same ratio for DDR4, roughly high single digits, and then the rest DDR5? Any commentary on the relative contributions to incremental gross margin between, say, you know, DDR5 and HBM?

Okay. Thank you very much.

Thank you or I should say I should say equipping the fab and then also doing tech transitions, you mean, sorry, just to be clear youre, saying in Idaho, or Oregon construction to production. There in that can have second half of 'twenty seven, but we are investing in technology transitions and optimizing.

Speaker #4: We have met our market share goal that we articulated many quarters ago. As a result, we are starting off our HBM4 in a very different position in terms of being able to meet the expectations and requirements of our customers from a volume support perspective.

Production across the.

Our footprint in Japan, and Taiwan, where we have available space.

Speaker #4: So we feel that, from a competitive positioning, that aspect of the business is in much better shape. Now with HBM4, you know, there has been a lot of chatter in the market and the media about, you know, what our HBM4 performance is compared to those who are doing the base tie and the foundry, et cetera.

Sumit Sadana: Yeah. We had said earlier to you that the DDR4 plus LP4, we had provided some guidance earlier. We are pretty close to those levels. I'm not going to say that it's going to sustain throughout the fiscal year because the mix changes over time. DDR4, just by itself, not counting LP4, has actually been a low single-digit % of the business. It's been relatively small. We have UL'd this. We have extended some of our UL timeline based on extreme shortages at customers. Of course, the margins have become dramatically better as this shortage has continued. I think that's that. In terms of the relative margin, if I understood your question correctly, relative margin between DDR and HBM, we are not going to comment on that because while the HBM, as I mentioned earlier, they run like two different business models.

Thank you. Our next question comes from the line of.

Chris <unk> of Citi. Please go ahead Chris.

Hey, Thanks, guys can you hear me okay.

Yes, so far.

Thank God I been having phone problems all day, so quick clarification as.

Speaker #4: And we just wanted to mention and lay some of those doubts to rest that, you know, we today believe that our HBM4 has the highest performance amongst any competing product out there.

As far as the DRAM Roscoe you guys gave us the HBM breakout can we assume.

Roughly the same ratio for DDR four roughly high single digits and then the rest DDR five and any any commentary on the.

Speaker #4: And we don't believe that others can match this performance with this level of capability, quality, power consumption, and so on. We believe that there will be tremendous customer preference for our products.

The relative contributions to incremental gross margin between say DDR five and HBM.

Yes. So the we have said earlier to you that the DD.

<unk> four plus LP for.

Speaker #4: Of course, we have increases in our HBM supply that we have created through our investments. We have confidence that we will sell out that supply for calendar 2026, and there may be some HBM share shifts happening.

We have provided some guidance earlier, we are pretty close to those levels I'm not going to say that it's going to sustain throughout the.

The fiscal year, because the mix changes over time.

But DDR for just by itself not counting LP forward has actually been a low single digit percentage.

Sumit Sadana: While the HBM pricing is set ahead of time for the upcoming calendar year, the pricing on the rest of the DRAM portfolio and, of course, NAND portfolio also is moving around based on quarterly trends. We have mentioned to you that the DRAM industry is tight. As we get into 2026, we expect it to further tighten due to all of the robust demand that we are seeing and the limits on the supply growth caused by a number of factors that were discussed. We believe that tightening is going to enable improving pricing and margins on the non-HBM portion of the portfolio. That will move around. It's not really practical or possible to give you sort of a relationship between the two.

Speaker #4: On the competitive landscape, it is our belief that most of that is most likely going to be between our two competitors rather than impacting our share in any material way.

Off the business. So it's been relatively small, but we have.

You all this we have extended some of our UL timeline based on extreme shortages at customers.

And of course, the margins have become dramatically better as this shortage has continued.

Speaker #7: Thank you. And then, as a follow-up on the CapEx side, could you maybe give us some color on the split of WFE versus facility CapEx you saw in fiscal 2025?

But yes, I think I think that's that's that I think in terms of the relative margin. If I understood. You correct question correctly relative margin between DDR and HBM.

Speaker #7: And then, whether we should expect any significant change in that mix heading into next year. Thank you.

Speaker #5: Jim, it's Mark. We're not going to provide that breakout, just that the vast majority of this spend is for DRAM construction equipment.

We are not going to comment on that because while the HBM.

As I mentioned earlier, they run like two different business models, so wildly HBM pricing is set.

Speaker #7: And any significant change in the facility piece of that for next year, though?

Head of time for the upcoming calendar year.

The.

Pricing on the rest of the DRAM portfolio and of course NAND portfolio also.

Speaker #5: We're not going to comment on that. We're, you know, we're, as you know, we're expanding in a number of sites including, we talked about, you know, our first Sab in Idaho and we're also equipping those tabs as well.

Is moving around based on quarterly trends.

[Analyst]: That's great. That's still super helpful. For my follow-up question for Mark on gross margins, Mark, remember we had lunch earlier this year and I asked you about, you know, could you guys get back to 50% gross margin? You said it's possible. Who knows? Depends on a lot of things. It's going to be tough. Here we are, and now we're going past that. Can you just maybe talk about what changed or what was the key to that? What are the limiting factors on gross margin going forward? Is it just essentially a function of, you know, how many more quarters pricing keeps going up or is there something else going on as far as mix goes? Thanks.

And we have mentioned to you.

At the DRAM industry is tight and as we get into 2026, and we expect it to further tighten due to all of the robust demand that we're seeing and the limits on the supply growth caused by a number of factors that were discussed.

Speaker #7: Okay. Thank you very much.

Speaker #1: Thank you.

Speaker #5: Or I should say, I should say equipping the fab and then also doing tech transitions.

Speaker #1: You mean, sorry, just to be clear, you're saying in Idaho we're in construction, and you expect production there in the second half of

Speaker #5: Second half of.

Speaker #1: 27. But we are investing in technology transitions and optimizing production across our footprint in Japan and Taiwan, where we have available cleaner space.

So we believe that that.

Tightening is going to enable improving pricing and margins on the non HBM portion of the portfolio. So that will move around so it's not really.

Article or possible to give you sort of a relationship between the two.

Speaker #1: Thank you. Our next question comes from the line of Chris Dainley of City. Please go ahead, Chris.

Mark Murphy: Chris, since through the fourth quarter, the market conditions continue to improve. As you know, we updated in mid-August the 11th, and we reported a margin 120 basis points higher than that update. Prices continue to improve. The market conditions are very good, very tight on DRAM, and they've improved in NAND and continue to improve. The margin we reported is now above where it was in mid-fiscal 2022, and DRAM margins are higher than that mid-2022 period. The operating margin's the highest that it's been since November 2018. From here, the NAND business can continue to improve. There's been good supply response by us, and we've focused on higher-value SSD products in that business. While that business is below 2022 levels, it can continue to improve, and that will help overall margins. Furthermore, as we talked about, DRAM is very tight now, and incremental supply is meaningful.

That's great. That's that's still Super helpful. And then for my follow up.

Speaker #8: Hey, thanks guys. Can you hear me okay? So

Question for Mark on gross margin so.

Mark remember, we had lunch earlier this year and I asked you about could you guys get back to 50% gross margin you said, it's possible who knows depends on a lot of things, it's going to be tough and here we are in.

Speaker #5: Yes.

Speaker #8: So far, thank you. Thank God, I've been having phone problems all day. So, quick clarification. As far as the DRAM reps go, you guys gave us the HBM breakout.

Speaker #8: Can we assume roughly the same ratio for DDR4, roughly high single digits, and then the rest DDR5? And any commentary on the relative contributions to incremental gross margin between, say, DDR5 and HBM?

And now we're going past that.

So can you just maybe talk about what what changed or what was what was the key to that and then what are the limiting factors on gross margin going forward or is it just essentially a function of how many more quarters pricing keeps going up or is there something else going on as far as Mexico's thanks.

Speaker #4: Yeah, so we had said earlier to you that the DDR4 plus LP4 we had provided some guidance earlier. We are pretty close to those levels.

So Chris since through through the fourth quarter.

Okay.

Market conditions continue to improve.

Speaker #4: I'm not going to say that it's going to sustain throughout the fiscal year because, you know, the mix changes over time. But DDR4, just by itself—not counting LP4—has actually been a low single-digit percentage of the business.

And as you know we updated.

In mid August 11th and <unk>.

They reported out.

Margin 120 basis points higher than that.

Date, so prices continue to improve the mutations R. R.

Speaker #4: So it's been relatively small, but we have UL this; we have extended some of our UL timeline based on extreme shortages at customers. And of course, the margins have become dramatically better as this shortage has continued.

Very good.

It's a very tight on DRAM.

They have improved in NAND and continue to improve.

Yes, the <unk>.

Margin we reported.

Is now above.

Speaker #4: But yeah, I think that's that. I think in terms of the relative margin, if I understood you correct, question correctly, relative margin between DDR and HBM, we are not going to comment on that because while the HBM, as I mentioned earlier, you know, they run like two different business models.

Where it was in mid fiscal 'twenty, two and DRAM is DRAM margins are higher than that.

At mid 'twenty two period.

The operating margin is the highest it's been since November 18.

Mark Murphy: Incremental supply is difficult to bring on. You have a number of factors that are structural here. Our inventory levels are low, below targets. We've got this extended life on DDR4, LP4, which is constraining the ability to do tech node transitions. You have obviously the silicon intensity of HBM as that continues to grow quickly. Incremental capacity, as you know, to bring on a fab takes a long time. For example, our new fab, Idaho, will come on in wafer production, starts in meaningful wafer production starts in second half 2027. That constrained supply focus on diverting bits to the best products and also our continued good cost improvements, those all offer the ability to expand margins. We've indicated that we expect second quarter gross margin to be up relative to first.

From here.

Uh huh.

The NAND business can continue to improve.

Speaker #4: So, while the HBM pricing is set ahead of time for the upcoming calendar year, the pricing on the rest of the DRAM portfolio, and of course the NAND portfolio, is also moving around based on quarterly trends.

There has been good supply response.

By Us and we've.

<unk> focused on higher value.

SSD products in that business so.

While that business is below 22 levels.

Speaker #4: And we have mentioned to you that the DRAM industry is tight, and as we get into 2026, we expect it to further tighten. Due to all of the robust demand that we are seeing and the limits on the supply growth caused by a number of factors that were discussed, we believe that this tightening is going to enable improving pricing and margins on the non-HBM portion of the portfolio.

It can continue to improve and that will help overall margins.

And then Furthermore, as we talked about DRAM is very tight now.

And.

Incremental supply is.

A meaningful incremental supply is difficult to bring on do you have a number of a number of factors.

Yes that are structural here, our inventory levels are low below targets.

We've got this extended life on.

DDR for LP for which is constraining the ability to do tech node transitions.

Speaker #4: So, that will move around. So, it's not really practical or possible to give you sort of a relationship between the two.

You have obviously the silicon intensity of HBM as that continues to grow quickly.

Speaker #1: That's great. That's still super helpful. And then for my follow-up, a question for Mark on gross margin. So, Mark, remember we had lunch earlier this year and I asked you about, you know, could you guys get back to 50% gross margin?

And then incremental <unk>.

[Analyst]: Great. Thanks for the explanation, Mark.

Capacity as you know to bring out a fab it takes a long time.

And.

Speaker #1: You said it's possible. Who knows? It depends on a lot of things. It's going to be tough. And here we are. And now we're going past that.

<unk>.

For example, our new Fab, Idaho will come come on.

Operator: Thank you.

[Analyst]: I'm sorry.

Operator: Thank you. Our next question comes from the line of Vijay Rakesh of Mizuho. Please go ahead, Vijay.

Wafer production starts in <unk>.

Speaker #1: So can you just maybe talk about what changed or what was the key to that? And then, you know, what are the limiting factors on gross margin going forward?

Meaningful wafer production starts in second half 'twenty seven.

Satya Kumar: Yeah, hi, Mark, Sumit, and Manish. Just a quick question on the DRAM side. I know you mentioned 1 Gamma is starting now. If you look at the year 1 Beta, which looks like most of the HBM4 will be on 1 Beta as well, any thoughts on what the mix will be for you, 1 Gamma versus 1 Beta now versus exiting the year, fiscal 2026, let's say?

So.

So that constrained supply.

Speaker #1: Is it just essentially a function of how many more quarters pricing keeps going up, or is there something else going on as far as mix goes?

Our focus on diverting pets to the best products.

And also our continued good cost improvements those all offer the ability to expand margins and we've indicated that we expect second quarter gross margin to be up relative to first.

Speaker #1: Thanks.

Speaker #5: So So Chris, since you know through the fourth quarter, the market conditions continue to improve. And you know, as you know, we updated in mid-August the 11th and you know we reported a margin 120 basis points higher than that update.

Great. Thanks for your explanation Mark.

Manish Bhatia: Yeah, sure. Thanks, Vijay. Manish, we are very pleased with 1 Gamma ramp. We were able to achieve both mature yield this quarter as well as first revenue shipments into hyperscaler as well as other applications. We're really, really pleased with how that ramp is going. We expect to be able to have 1 Gamma be the primary area where we're going to be, that's going to be providing us big growth for fiscal 2026 in terms of supply. We're going to be qualifying across multiple different product areas as we go through the year. We're already at a point where the 1 Gamma production plus the 1 Beta production are the significant majority of our bit output. That mix will continue to grow towards 1 Gamma as we go through calendar year 2026, excuse me, fiscal year 2026.

Okay.

Alright, thank you.

I'm sorry, Thank you our next question.

Comes from the line of Vijay Rakesh of Mizuho. Please go ahead Vijay.

Speaker #5: So you know prices continue to improve; the market conditions are very good. It's very tight on DRAM, and they've improved in NAND and continue to improve.

Yes, hi.

Yes.

Just a quick question on.

On the DRAM side I know you mentioned one gamma is starting now.

If you look at your run beta, which is you expect more of it.

Speaker #5: Yeah, the margin we reported is now above where it was in mid-fiscal '22. And DRAM margins are higher than that mid-'22 period.

We'll be on one small any.

Thoughts on what the mix will be for you.

Gallons versus one beta.

Now versus exiting the year fiscal 'twenty six.

Yeah.

Speaker #5: The operating margin is the highest that it's been since November 2018. From here, you know the NAND business can continue to improve. There's been a good supply response by us, and we've focused on higher value SSD products in that business.

Yeah sure. Thanks, We just finished we are very pleased with one gamma.

Ramp it's that we were able to achieve.

Both.

Mature yield this quarter.

As well as first revenue shipments.

Hyperscale.

Hyper scaler as well as other applications. So we're really really pleased with how that ramp is going.

[Analyst]: Go ahead.

Mark Murphy: I would just maybe add one thing. We talked quite a bit about supply constraints on the last question. A very important part of our supply solution in 2026 is this ramping of 1 Gamma, which, as Manish said, is going very well. Chris, just one additional note on your margin question is, is this just improved mix to, in our business, the data center and the higher performance requirements of that market and the favorable effect of that on the business?

<unk>.

Expect to be able to.

Speaker #5: So, while that business is below $22 levels, it can continue to improve, and that will help overall margins. Furthermore, as we talked about, DRAM is very tight now.

Have one gamma b the primary.

The area, where we're going to be.

That's going to be providing us bit growth for fiscal 'twenty six.

In terms of supply and we're gonna be qualifying across.

Multiple different.

Product areas as we go through the year.

Speaker #5: And incremental supply is meaningful; incremental supply is difficult to bring on. You have a number of factors that are structural here. Our inventory levels are low, below targets.

We are.

Already at a point, where the one gamma production plus the one data production are the significant majority.

40 of our bit output and that that mix will continue to grow towards one gamma as we go through.

Speaker #5: We've got this extended life on DDR4, LP4, which is constraining the ability to do techno transitions. You have, obviously, the silicon intensity of HBM that continues to grow quickly.

Calendar year 2000, excuse me fiscal year 'twenty.

Satya Kumar: Got it. On the HBM4, given that your higher speeds and lower power, I guess the die's bigger too, and your own logic die. Would you expect the margins on HBM4 to be better versus HBM3E, I guess, equal high?

I would just I would just I would just maybe maybe add one thing we've talked quite a bit about supply constraints on the last on the last question.

A very important part of our supply solution and 26 is ramping of lung camo, which is the east side is going very well.

Speaker #5: And then incremental capacity as you know, to bring on a fab takes a long time. And you know our for example, our new fab, Idaho, will come on in wafer production, starts in meaningful wafer production starts in second half '27.

And then Chris just one one additional note on your margin question is.

Sumit Sadana: Yeah, I mean, we are very confident in the capabilities of our product. Certainly, the HBM4 cost is higher than the HBM3 cost, and the HBM4 price will be meaningfully higher than the HBM3E price. We don't really talk about margins by product line or, you know, within a product line, you know, which product has what kind of relative margin. Overall, for HBM, we certainly expect that we'll have really good ROI capability for many years for the company, not just driven by the huge complexity of HBM, but also driven by the dramatic value it creates for our customers and our intent to, you know, ensure that we are benefiting from providing that value.

Is this just improved mix to.

And our business the data center and higher performance.

Requirements of that market.

Speaker #5: So that constrained supply focuses on diverting bits to the best products, and also our continued good cost improvements. Those all offer the ability to expand margins, and we've indicated that we expect second quarter gross margin to be up relative to first.

The favorable effect of that on the business.

Okay.

Got it and then on the on <unk> four.

Given your higher speeds.

Our.

I guess, Dave with regard to in your own logic, there would you expect the margins.

<unk> to be.

Better versus <unk> I guess.

Speaker #1: Great. Thanks for the explanation, Mark. Thank you. I'm sorry. Thank you. Our next question comes from the line of Vijay Rakesh of Mizuho.

Yes.

We are very confident in the capabilities of our product and certainly been HBM forecast is higher than MBA screamed fee cost and BHP and full price will be.

Speaker #1: Please go ahead, Vijay.

Meaningfully higher than the <unk> price.

Speaker #8: Yeah, hi, Mark. Sumit and Manish. Just a quick question on the DRAM side. I know you mentioned one gamma is starting now. But if you look at your one beta, which looks like most of the HBM4 will be on one beta as well.

Satya Kumar: Got it. Thanks.

Don't really talk about margins by product line.

Operator: Thank you. Our next question comes from the line of Aaron Rakers of Wells Fargo. Your line is open, Aaron.

Our within our product line.

Which product has what kind of.

Relative margin, but overall for <unk>, we certainly expect that.

[Analyst]: Yeah, thanks for taking the question. I'll ask too as well. I guess kind of building on some of the prior questions, I'm curious, I know, Mark, you had talked about your ability to execute, I think it was high single-digit cost down with HBM in DRAM in fiscal 2025, and then I think it was teens or something, low teens on NAND. I'm curious if, you know, as we make these process node transitions as well as G9 and NAND, how do you think about the cost down curve as these process nodes materialize through this next fiscal year?

Speaker #8: Any thoughts on what the mix will be for you, one gamma versus one beta, now versus exiting the year, fiscal '26, let's say?

Bill have.

Really good.

Roy capability.

For many years for the company not just driven by the huge complexity of HBM, but also driven by the dramatic value it creates for our customers and.

Speaker #1: Yeah, sure. Thanks, Vijay. Manish, we are very pleased with one gamma ramp. We were able to achieve both mature yield this quarter, as well as our first revenue shipments into hyperscalers and other applications.

An intent to.

Ensure that.

We are benefiting from providing that guidance.

Speaker #1: So, we're really pleased with how that ramp is going. We expect to be able to have one gamma be the primary area where we're going to be that's going to be providing us pit growth for fiscal '26.

Alright. Thanks.

Mark Murphy: Just to be clear, I think on cost downs, we gave for fiscal 2025 we're slightly better than high single digits for DRAM front-end ex-HBM and mid-teens for NAND. I think as, and then for DRAM with HBM, down low single-digit % and NAND cost reductions, low teens. These are consistent with commentary we've given in the past that are better.

Thank you.

Our next question.

Comes from the line of Aaron Rakers of Wells Fargo. Your line is open to Erin.

Yes, thanks for taking the question and I'll ask two as well.

Speaker #1: In terms of supply, and we're going to be qualifying across multiple different product areas as we go through the year. We're already at a point where the one gamma production plus the one beta production are the significant majority of our bid output.

I guess kind of building on some of the prior questions I'm curious I know Mark you had talked about your ability to execute I think it was high single digit cost down with HBM in DRAM in fiscal 'twenty five and then I think it was teams or something.

Low teens on me and I'm curious if you know as we make these process node transitions as well as Jeannine <unk> NAND, how do you think about the cost down curve as these process nodes materialized through this next fiscal year.

Speaker #1: And that mix will continue to grow towards one gamma as we go through calendar year fiscal year '26.

[Analyst]: Okay. I'm curious, just structurally, the NAND market, you know, there's a lot of discussion around these high-cap enterprise SSDs pushing 250 terabytes and plus. As we look at that market, I'm curious how you guys see the average capacity trending on these AI servers for SSDs and whether we should really be expecting that to inflect materially higher as we move through fiscal 2026 and some of these high-cap drives really, you know, hit true volume.

Speaker #8: Got it.

Speaker #5: Yeah, I would just maybe add one thing. We talked quite a bit about supply constraints on the last question. And, you know, a very important part of our supply solution in 2026 is this ramping of One Gamma, which, as Manisha is mentioning, is going very well.

Okay.

To be clear I think on cost downs.

We gave for fiscal 'twenty, five or slightly better than high single digits for DRAM front in X H.

And mid teens for NAND.

Yeah.

Speaker #5: And then Chris, just one additional note on your margin question is you know is this just improved mix to in our business the data center and the higher performance you know requirements of that market and the favorable effect of that on the business?

S.

And then for DRAM with HBM down low single digit percentage and.

Sumit Sadana: Yes, I think the average capacities are going to continue to escalate rapidly. We had 60-terabyte drives and then, you know, 102-terabyte drives. We have ourselves announced at our last SMS event 122-terabyte drives, 245-terabyte drives. We are very excited with the demand that we see for these drives. We do think there will be meaningful uptake of these high-capacity drives, and they will drive the average capacities higher. I also mentioned that, you know, there is expected to be a shortage of hard drive storage for these hyperscalers. More meaningful usage of NAND even beyond previous plans should drive a positive outcome here as well as an added tailwind.

<unk> cost reductions low teens sorry.

These are consistent with.

Commentary, we've given in the past it at that or better.

Yes.

And then I'm curious just like structurally the NAND market Theres a lot of discussion around the high cap enterprise SSD is pushing 250 terabytes and plus.

Speaker #1: Got it. And then on HBM4,

Speaker #8: Given your higher speeds and lower power, I guess it dives bigger into your own logic dive. Would you expect the margins on HBM4 to be better versus HBM3e, or equal high?

As we look at that market I'm curious, how you guys see the average capacity trending on these AI servers for ssds, and whether we should really be expecting that to inflect materially higher as.

Speaker #4: Yeah, I mean, we are very confident in the capabilities of our product, and certainly the HBM4 cost is higher than the HBM3 cost. The HBM4 price will be meaningfully higher than the HBM3e price.

As we move through fiscal 2006, and some of these these high cap drives really hit.

Through volume.

Yes, I think the <unk>.

Average capacities.

Is going to continue to escalate rapidly.

Had 60 terabyte drives and then.

Speaker #4: We don't really talk about margins by product line. Or, you know, within a product line, you know which product has what kind of relative margin.

102, terabyte drives and we have ourselves announced at our last SMS event.

[Analyst]: Where do you think the average capacity is today for servers? Just curious.

Sumit Sadana: It's moving around quite a bit because in any given quarter, the mix of general purpose servers and AI servers keeps changing. AI servers generally tend to use the highest capacity that is available in very meaningful quantities. AI servers have been, now there are, you know, AI servers that are focused on just ensuring a huge amount of capacity. The different workloads of inference that are also increasing, more fragmentation is happening in the configurations that are getting shipped. Generally speaking, the highest capacity drives that the industry has been able to produce do get used by AI servers. They are going from 100, they're going from 60 terabytes to 120 and actively using a lot of 245 going forward.

100002, terabyte drives 245 terabyte drives.

And so we are very excited with that.

Speaker #4: But overall, for HBM, we certainly expect that we'll have really good ROI capability for many years for the company. This is not just driven by the huge complexity of HBM but also by the dramatic value it creates for our customers and our intent to ensure that we are benefiting from providing that value.

The demand that we see for these drives and we do think there will be meaningful uptake of these.

High capacity drives and they will drive the average capacity is higher.

And I also mentioned that there is expected to be a shortage of hard drive storage for these hyperscale orders and so.

More meaningful usage of NAND even beyond.

Previous plans.

Should drive.

Positive outcome here as well as an added tailwind.

Speaker #1: Got it. Thanks. Thank you. Our next question comes from the line of Aaron Rakers of Wells Fargo. Your line is open, Aaron.

Where do you think the average capacity is today for server just curious.

It's moving around quite a bit because in any given quarter. The mix of general purpose servers, and AI servers keeps changing and AI servers generally tend to use the highest capacity that is available and very meaningful.

Speaker #7: Yeah, thanks for taking my question. I'll ask two as well. I guess, kind of building on some of the prior questions, I'm curious.

[Analyst]: Thank you.

Speaker #7: I know, Mark, you had talked about your ability to execute. I think it was a high single-digit cost down with HBM in DRAM in fiscal '25, and then I think it was teens or something.

Quantity. So AI servers have been there are AI servers that are focused on just ensuring huge amount of capacity, but then the different workloads of influence that are also increasing.

Sumit Sadana: Welcome.

Operator: Thank you. Our next question comes from the line of Brian Chen of Stifel. Please go ahead, Brian.

Speaker #7: Low teens on NAND. I'm curious, as we make these process node transitions, as well as G9 and NAND, how do you think about the cost-down curve as these process nodes materialize through this next fiscal year?

Manish Bhatia: Hi there. Thanks for letting us ask a few questions. Maybe I know there's a couple of questions on cost down or cost reduction, but I was just curious. You've obviously talked about in the past HBM3E, some efficiencies you've gained as you've kind of moved into relative maturity from where you started. I know you don't want to provide too much detail around pricing in 2026 for 3E while your tone does suggest you should have sufficient leverage, I guess, in terms of those negotiations. I'm wondering, is there more efficiency in terms of improvement in assembly and packaging yields for 3E that maybe could also help you out a little bit in terms of margin profile next year? Thanks, Brian. We are very happy with how the HBM3E 12-high ramp has gone.

More fragmentation is happening in the configurations that are getting shipped but generally speaking.

The highest capacity drives that the industry has been able to produce.

Speaker #5: Just to be clear, I think on cost downs, we gave for fiscal '25 or slightly better than high single digits for DRAM front end x HBM.

Do get.

Used by AI servers so.

They are going from 100, they're going from 60 terabytes of 120 and actively.

Speaker #5: And mid-teens for NAND. You know, I think as for DRAM, with HBM down low single-digit percentage and NAND cost reductions in the low teens.

Using a lot of $2 45 going forward.

Thank you.

Okay.

Thank you.

Our next question.

Comes from the line, Brian Chin of Stifel. Please go ahead Brian.

Speaker #5: So, I mean, these are consistent with the commentary we've provided in the past that, you know, at that or better.

Hi, there thanks.

Thanks for letting us ask a few questions.

Manish Bhatia: We talked about being able to get our yields to maturity on 12-high significantly faster than we did on 8-high, which was really our first major HBM volume node. We've now gotten production capacity up to meaningful volumes as we kind of gave you. There will be continued opportunity, but I would say that really the biggest part of that benefit in ramp has happened as we've gone through the 12-high over the last couple of quarters. Okay.

Maybe I know, there's a couple of questions on costs down or cost reduction, but I was just curious you've obviously talked about in the past Hbm's <unk>. Some efficiencies you gain as you've kind of moved into relative maturity from where you started.

Speaker #1: Yep. Okay. And then I'm curious, just like...

Speaker #7: structurally the NAND market, you know there's a lot of discussion around these high cap enterprise SSDs pushing 250 terabytes and plus. You know as we look at that market, I'm curious how you guys see the average capacity trending on these AI servers for SSDs and whether we should really be expecting that to inflect materially higher as we move through fiscal '26 and some of these high cap drives really hit true volume.

And I know you are not you don't want to provide too much detail around pricing in 2026 for <unk>. While your tone does suggest you have you should have sufficient leverage I guess in terms of those negotiations, but I'm wondering is there more efficiency.

The improvement in the assembly and packaging yields for three that maybe you could also help you out a little bit in terms of margins.

Mark Murphy: Got it. I appreciate that. Maybe a quick follow-up. I know there's been some discussion here about your guys reiterating this 50%+ blended gross margin, first time since the 2017, 2018 timeframe. I know you wouldn't recast or you can't recast cloud and core enterprise server revenue back that far. You did give us the sequential and the year-over-year compare today. Those segments are roughly mid-50% of sales now. What roughly did they represent back then? Clearly, it must have been a much smaller proportion. That's obviously a structural shift in the business period to period.

Speaker #4: Yes, I think the average capacities are going to continue to escalate rapidly. We had 60 terabyte drives, then 102 terabyte drives, and at our last SMS event, we announced 122 terabyte drives and 245 terabyte drives.

Margin profile next year.

Thanks, Brian.

We are very happy with how the <unk> III <unk> 12 high ramp.

Is gone, we talked about being able to get our yields to maturity on 12 high significantly faster than we did on eight high which was really our first major HBM volume node.

Now gotten production capacity up to a meaningful volumes as we kind of gave you. So.

Speaker #4: And so we are very excited about the demand that we see for these drives, and we do think there will be meaningful uptake of these high-capacity drives, which will drive the average capacities higher.

There will be continued opportunity, but I would say that.

Really the.

The biggest part of that benefit and ramp as has happened as we've gone through the 12.

Speaker #4: And I also mentioned that there is expected to be a shortage of hard drive storage for these hyperscalers, and so more meaningful usage of NAND even beyond previous plans.

Sumit Sadana: Yeah, I mean, without giving you specifics for any one particular timeline, I'll just mention to you that most segments that are part of the TAM over the different time horizons in the past prior to generative AI coming on the scene had peaked around a third of the overall market in terms of segment size, right? If you think about the data center segment or you think about the mobile segment, when they would go into this big spurt, it would peak around, roughly speaking, give or take, a third of the overall market. This time around, of course, that pseudo ceiling of a third is no longer in place because the data center has become more than half and continues to outgrow the rest of the market at very robust profitability levels. We do think that AI has radically changed the landscape.

Over the last couple of quarters.

Okay got it and Oh.

Appreciate that maybe a quick follow up.

I know there's been some discussion here about sort of your guys' retaining this 50% plus blended gross margin you know first time since 2017 2018 timeframe.

Speaker #4: Should drive a positive outcome here, as well as an added tailwind.

I know you wouldn't recast or you cant recast cloud core enterprise server revenue back that far you did give us sort of the sequential and the year over year compare.

Speaker #7: Where do you think the average capacity is today for servers? Just curious.

Speaker #4: It's moving around quite a bit because, in any given quarter, the mix of general-purpose servers and AI servers keeps changing. AI servers generally tend to use the highest capacity that is available in very meaningful quantities.

But those segments are roughly mid 50% of sales now.

Roughly did they represent back then.

It must have been a much smaller proportion.

And that's obviously a structural shift in the business period to period.

Speaker #4: So AI servers have been there are AI servers that are focused on just ensuring huge amount of capacity, but then with different workloads of inference that are also increasing more fragmentation is happening in the configurations that are getting shipped.

Yes, I mean without giving you.

Best effects for any one.

Particular timeline.

And to do that.

Most segments.

That are part of the Tam or the different time horizons in the past prior to generating AI coming on the scene.

Speaker #4: But generally speaking, the highest-capacity drives that the industry has been able to produce do get used by AI servers. So they are going from 60 terabytes to 120 terabytes and actively using a lot of 245 terabytes going forward.

Had peaked around a third of the overall market in terms of segment size right. So if you think about <unk>.

Sumit Sadana: Not only is the mix due to AI growth in the data center caused the data center to become a much larger part of the TAM, but it is also a higher value as well as a higher margin opportunity. We did mention that HBM and high-capacity DIMMs and the LP that goes into the data center, these three categories alone were $10 billion in fiscal 2025, so a very large part of our fiscal 2025 revenue. These are all high-value-add portions of the portfolio. This doesn't even account for regular DDR5 that is in 64 gigabytes and below type of DIMMs that are going into the data center. It doesn't account for data center SSDs that are part of the data center portfolio. There is a lot of that shift that is going on, but it's also a shift that has a lot of legs to continue.

Data Center segment.

Think about the mobile segment when they would go into this.

Big Sport it would peak around roughly speaking give or take a third of the overall market and this time around of course, you know that.

Speaker #7: Yep. Thank you.

Speaker #4: You're welcome.

Speaker #1: Thank you. Our next question comes from the line of Brian Chen of Stifel. Please go ahead, Brian.

Pseudo sealing off of code is no longer in place because data center has become.

Speaker #5: Hi, hi there. Thanks for letting us ask a few questions. Maybe I know there's a couple of questions on cost down or cost reduction, but I'm just curious.

More than half and continues to outgrow the rest of the market.

And very robust profitability levels. So we do think that AI has radically changed the landscape not only is the mix.

Speaker #5: You've obviously talked about in the past HBM3e, some efficiencies you've gained as you've kind of moved into relative maturity from where you started. And I know you're not wanting to provide too much detail around pricing in 2026 for 3E.

Due to a high growth in the data center cost the data center to become a much larger part of the Tam, but it is also a higher value.

Speaker #5: While your tone does suggest you should have sufficient leverage, I guess, in terms of those negotiations. But I'm wondering, is there a more efficiency in terms of improvement in assembly and packaging yields for 3E that maybe could also help you out a little bit in terms of margins profile next year?

As well as a higher margin opportunity.

You did mention that you know HBM.

And high capacity dams and.

Sumit Sadana: That's also the demand coming from the data center that's creating a lot of this tightness in the industry and causing pricing to and enabling us to be able to increase pricing across all of the segments of the market, not just the data center. It's helping the profitability of the entire portfolio of the company, even though it's driven largely from the data center.

The LP that goes into the data center. These three categories alone were $10 billion.

Speaker #1: Thanks, Brian. So you know we are very happy with how the HBM3e 12 high ramp has gone. You know we talked about being able to get our yields to maturity on 12 high significantly faster than we did on 8 high, which was really our first major HBM volume node.

In fiscal 'twenty five so a very large part of our fiscal 'twenty five revenue and these are all high value add portions of the portfolio. This doesn't even account for.

Speaker #1: And you know we've now gotten production capacity up to meaningful volumes as we kind of gave you. So, you know, there will be continued opportunity, but I would say that, you know, really the biggest part of that benefit in ramp has happened as we've gone through the 12 high.

Regular DDR five.

That is in 64 gigabytes and below type of games that are going into the data center doesn't account for.

Manish Bhatia: Great. Appreciate that.

Data Center Ssds that are part of the data center portfolio. So.

Operator: Thank you. Our next question comes from the line of Quinn Bosen of Barclays. Please go ahead, Quinn.

There is a lot of that shift that is going on but it's also a shift that has a lot of legs.

Speaker #1: You know, over the last couple of quarters.

Speaker #5: Got it. And I appreciate that. Maybe a quick follow-up. I know there's been some discussion here about sort of you know your guys you know retaining this 50% plus blended gross margin you know first time since 2017.

[Analyst]: Hey, Mark. Maybe I just missed your answer. I just want to clarify your answer to Aaron's question on the cost downs. You did low single digits in DRAM inclusive of HBM and low teens in NAND. Did you say that those are good ranges to think about in 2026, or did you not give a cost down target for fiscal 2026?

To continue and that's also the demand coming from the data center, that's creating a lot of this tightness in the industry and causing pricing to enabling us to be able to increase and increase pricing across all of the segments of the market.

Speaker #5: 2018 timeframe. I know you wouldn't recast or you can't recast Cloud and Core Enterprise Server revenue back that far. You did give us sort of the sequential and the year-over-year comparison.

Not just the data center, so it's helping the profitability.

Mark Murphy: We just gave 25.

Of the entire portfolio of the company.

Speaker #5: Today, you know those segments represent roughly 50% of sales now. What did they represent back then? It must have been a much smaller proportion.

[Analyst]: Thank you for that clarification. I guess maybe just kind of a bigger picture question. As you look at the inferencing market, NVIDIA recently introduced their CPX GPU that uses GDDR7 instead of HBM memory. To the extent that that architecture takes off, do you think that lower HBM content on inferencing platforms would have any impact on your broader outlook for the HBM market over the next few years, or do you think that the CPX architecture represents sort of only a small portion of the inference market?

Even though it's driven largely from the data center.

Okay, Great I appreciate that.

Yeah.

Speaker #5: And you know that's obviously a structural shift in the business period to period.

Okay.

Thank you.

Our next question comes from the line of Quinn Bolton Needham <unk> Company. Please go ahead cream.

Speaker #4: Yeah, I mean, without giving you specifics for any one particular timeline, I'll just mention to you that most segments that are part of the TAM over the different time horizons in the past, prior to generative AI coming on the scene, had peaked around a third of the overall market in terms of segment size, right?

Hey, Mark maybe I just missed your answer I just want to clarify your answer to Aaron's question on the cost Downs you did low single digits in DRAM inclusive of HBM and low teens and Dan did you say that those are good ranges to think about in 2006 or did you not give a cost down target for fiscal 'twenty.

Sumit Sadana: Yeah, as the AI market continues to grow and evolve, it is for sure going to have a lot of different use cases come up, a lot of different types of workloads for which optimized architectures would have to be evolved. This is just another step in that direction. The AI market is going to become very, very big over time, and it is going to have needs that are not going to be one size fits all. That optimization will be necessary to scale out the capabilities in a cost-effective way. This is just one example. There will be more in the future. One important thing to keep in mind as all of this happens is that a lot of the inferencing workloads tend to be memory-bound workloads.

Six.

Speaker #4: So if you think about the data center segment, or you think about the mobile segment, when they would go into this big spurt, it would peak around roughly speaking, give or take, you know, a third of the overall market.

We didn't give a we just gave 25.

Perfect. Okay. Thank you for that.

Certification and then I guess, maybe just kind of bigger picture question as you look at the inferencing market Nvidia recently introduced their CTX GPU that uses GDR seven instead of HBM memory and to the extent that that architecture takes off do you think that that lower.

Speaker #4: And this time around, of course, you know that the pseudo ceiling of a third is no longer in place because the data center has become more than half and continues to outgrow the rest of the market.

<unk> content on infringing platforms would that have any impact on your broader outlook for the HVA market over.

Speaker #4: And very robust profitability levels. So we do think that, you know, AI has radically changed the landscape. Not only is the mix due to AI growth in the data center costs, the data center has become a much larger part of the TAM, but it is also a higher value as well as a higher margin opportunity.

The next few years or do you think that the.

CTX architecture represents sort of only a small portion of the inference market.

Yes.

AI market continues to grow and evolve.

It is.

For sure going to.

Have a lot different.

Speaker #4: We did mention that you know HBM and high-capacity DIMMs and the LP that goes into the data center. These three categories alone were $10 billion in fiscal '25.

Use cases come up a lot of different types of workloads for which optimized architectures would have to be involved.

Sumit Sadana: Consequently, as inference becomes a larger and larger part of the AI market, and over time, will likely become, you know, 80% of the AI market over the years, a lot of that being memory-bound means customers will be looking for different architectures that increase memory capacity and memory bandwidth. Depending on the latency requirements and time-to-first token and those types of different KPIs that these architectures have to hit, different types of solutions will be appropriate for various workloads. We are having a strategy to ensure that we are very strong in HBM, as we have discussed at length. We have an industry-leading product with our GDDR7.

And this is just another step in that direction.

AI market is going to become very very big over time and it is going to have.

Speaker #4: So, a very large part of our fiscal '25 revenue. And these are all high-value-add portions of the portfolio. This doesn't even account for regular DDR5 that is in 64 gigabytes and below type of DIMMs that are going into the data center.

Needs that are not going to be one size fits all and that optimization will be necessary to scale out the capabilities in a cost effective way.

So certainly.

Speaker #4: It doesn't account for data center SSDs that are part of the data center portfolio. So, there is a lot of that shift that is going on, but it's also a shift that, you know, has a lot of legs to continue. And that's also the demand coming from the data center that's creating a lot of this tightness in the industry and causing pricing to increase, enabling us to be able to raise pricing across all of the segments of the market, not just the data center.

This is just one example, there will be more in the future one important thing to keep in mind is all of this happens is that a lot of the inferencing workloads tend to be memory bound workloads.

And consequently, as entrance becomes larger and larger part of the air market and over time will likely become.

Sumit Sadana: We are the pioneers of LPDRAM in the data center, which is going to be used more and more over time, not just because of its energy efficiency, but because, you know, more industry players are going to, from an ecosystem perspective, from a CPU enablement perspective, going to start supporting LP over time. All of these capabilities that we have developed that are industry-leading will serve us really well. Our intent is to create those high-value solutions across the rich frontier of evolving architectures that our customers will be driving.

80% of the AI market.

Over the years.

A lot of that being memory bound means.

<unk> will be looking for different architectures that increase memory capacity in memory bandwidth.

Speaker #4: So, it's helping the profitability of the entire portfolio of the company, even though it's driven largely from the data center.

And depending on the latency requirements and time to first spoken and those types of.

Speaker #5: Great. Appreciate that.

Different kpis that these architectures have to hit.

Speaker #1: Thank you. Our next question comes from the line of Quinn Bolton of Needham & Company. Please go ahead, Quinn.

Different types of solutions wouldn't be appropriate for various workloads.

And.

We are having a strategy to ensure that we are very strong in <unk> as we have discussed at length, we have an industry leading product with <unk> seven.

Speaker #9: Hey Mark, maybe I just missed your answer. I just want to clarify your answer to Aaron's question on the cost downs. You did low single digits in DRAM, inclusive of HBM, and low teens in NAND.

[Analyst]: Excellent. Thank you.

Operator: Thank you. Our next question comes from the line of John Vinh of KeyBank Capital Markets. Please go ahead, John.

We are the pioneers of MPD Ram and the data center, which is going to be used more and more over time.

Speaker #9: Did you say that those are good ranges to think about in '26, or did you not give a cost-down target for fiscal '26?

Not just because of it.

Energy efficiency, but because.

[Analyst]: Great. Thanks for taking my question. I just had a couple of follow-ups. First on HBM, Sanjay on the call really went out of his way to kind of say, "Hey, we can support 2.8 terabytes per second and 11 gigabits per second." It seems like these performance specs are greater than the original JEDEC standards that were set. It looks like customers are asking for more. I'm just curious, why are customers increasing kind of the original specifications for HBM4 and then asking for higher performance? Just a quick housekeeping question from Mark. Looking at your guidance, given the midpoint of all the ranges that you provided, it looks like to get to $3.75 in EPS, it implies that interest income is positive in the quarter. Is that the case of what's implied there?

Speaker #5: We didn't give a number; we just gave $25.

More industry players are going to from an ecosystem perspective from a CPU enablement perspective going to stop supporting LTE over time. So all of these capabilities that we have developed that are industry, leading will serve us really well and our intent is to create those high value solutions across the rich.

Speaker #9: Just gave 25. Okay. Thank you for that. Clarification. And then I guess maybe just kind of bigger picture question. As you look at the inferencing market NVIDIA recently introduced their CPX GPU that uses GDDR7 instead of HBM memory and to the extent that that architecture takes off, do you think that that you know lower HBM content on inferencing platforms, would that have any impact on your broader outlook for the HBM market over the next few years or do you think that the you know that that CPX architecture you know represents sort of only a small portion of the inference market?

Frontier of <unk>.

Evolving architectures that our customers will be driving.

Okay. Thank you.

Thank you.

Our next question.

Come from the line of John <unk> of Keybanc capital markets. Please go ahead John.

Speaker #4: Yeah, as the AI market continues to grow and evolve, it is for sure going to have a lot of different use cases come up, a lot of different types of workloads for which optimized architectures would have to be developed.

Great. Thanks for taking my question I just had a couple of follow ups. So first on HBO Sanjay on the call really went out of his way to kind of say hey, we can support 2.8 terabytes per second in loving Gigabits per second.

Sumit Sadana: Okay. I'll just answer the first part of your question and then hand it over to Mark. In terms of the HBM4 specs and its evolution, no doubt that the JEDEC specs were definitely exceeded by a significant margin by some of the numbers that we have quoted to you today, with greater than 11 gigabits per second pin speed and 2.8 terabytes per second of bandwidth. We have sampled these products. Customers obviously are looking for as much improvement as they can get in the ROI that they can offer to their end customers. That typically means that if there is a way to really drive higher bandwidth, you can increase the tokens, you can reduce the time-to-first token, you can get to a lower cost per token kind of a metric, and that becomes a more attractive solution for their end customers. That is really what is happening.

It seems like these performance specs.

Our greater than the original Gerrick standards that were set and it looks like customers are asking for more I'm. Just curious why are customers increasing kind of the original specifications for each beam for it and asking for higher performance and then just a quick housekeeping question for.

Speaker #4: And this is just another step in that direction. The AI market is going to become very, very big over time, and it is going to have needs that are not going to be one size fits all.

Speaker #4: And that optimization will be necessary to scale out the capabilities in a cost-effective way. So certainly, this is just one example; there will be more in the future.

Mark.

Just looking at your guidance.

Given the mid points of volume.

Ranges that you provided it looks like to get to $3 75 in EPS. It implies that interest income is positive in the quarter is that is that the case of whats implied there.

Speaker #4: One important thing to keep in mind as all of this happens is that a lot of the inferencing workloads tend to be memory-bound workloads.

Hey, Isaac.

To answer the first part of your question and then hand it over to Mark.

Speaker #4: And consequently, as inference becomes a larger and larger part of the AI market, and over time will likely become, you know, 80% of the AI market over the years, a lot of that being memory-bound means customers will be looking for different architectures that increase memory capacity and memory bandwidth.

In terms of.

The HBM four specs and its evolution no doubt that the jet expects for.

Definitely exceeded by a significant margin by some of the numbers that.

We have quoted to you today.

With greater than 11 gigabit per second speed and to eight terabytes per second of bandwidth.

Speaker #4: And depending on the latency requirements and time to first token, and those types of different KPIs that these architectures have to hit, different types of solutions will be appropriate for various workloads.

Sumit Sadana: Certainly, the HBM4 specs that we have been able to produce will create a new bar for performance that would be needed to really enable these types of high-performing systems. That is really what has happened on that front. I'll pass it on to Mark to talk about the second part of your question.

We have samples of these products.

Customers, obviously are looking for as.

As much improvement as they can get and the rois that they can offer to their end customers.

Speaker #4: And we are having a strategy to ensure that we are very strong in HBM, as we have discussed at length. We have an industry-leading product with our GDDR7.

That typically means that if there is a way to really drive higher bandwidth you can increase the tokens.

Mark Murphy: Yeah, John, we do expect that interest to flip to income. We have greater capitalized interest, lower debt, and higher cash balances. All those will flip that line.

Speaker #4: We are the pioneers of LP DRAM in the data center, which is going to be used more and more over time. Not just because of its energy efficiency, but because more industry players are going to, from an ecosystem perspective, and from a CPU enablement perspective, start supporting LP over time.

Can reduce the time to first token you can get to a lower cost per token kind of metric.

And that becomes a more attractive solution for their end customers.

[Analyst]: Great. Thank you.

No.

That's.

Really what is happening, but certainly the HBM for.

Operator: Thank you. Our next question comes from the line of Kevin Cassidy of Rosenblatt Securities. Please go ahead, Kevin.

Specs that we have been able to produce will create a new bar for performance that would be.

Speaker #4: So all of these capabilities that we have developed, which are industry-leading, will serve us really well. Our intent is to create those high-value solutions across the rich frontier of evolving architectures that our customers will be driving.

That would be needed to.

[Analyst]: Yes. Thanks for taking my question. With the supply-demand going favoring the demand side, is there any discussions yet for long-term agreements with all of your clients or your customers?

Really.

Enable these type of high performing systems.

So that's that's really what has happened on that front and I'll pass it onto Mark to talk about the second part of your question yes.

Speaker #7: Excellent. Thank you.

Speaker #1: Thank you. Our next question comes from the line of John Vinn of KeyBank Capital Markets. Please go ahead, John.

Yes, John we do expect.

Net interest a flipped income we have greater cap capitalize interest we have lower debt, we have higher cash balances all of those.

Sumit Sadana: Yes, there is certainly interest in some of these agreements. We have certainly experimented with different types of long-term agreements in the past and innovated with some ideas on how to do some of these things. I do think that we are going to be very thoughtful about what we do here and what term and lifetime of these agreements we sign up for. We also have U.S. manufacturing that we are going to be bringing online. We also have changes that are likely to happen on the landscape due to tariffs, and we will have to respond to that when they are announced, Section 232 semiconductor tariffs. There are all kinds of reasons to be thoughtful about what happens to pricing, what happens to how we create value for customers, whether it is through U.S. manufacturing or product capabilities.

Flip that line.

Speaker #10: Great. Thanks for taking my question. I just had a couple of follow-ups. So, first on HBM, Sanjay Mehrotra on the call really went out of his way to kind of say, "Hey, we can support 2.8 terabytes per second and 11 gigabits per second." It seems like these performance specs are greater than the original JEDEC standards that were set, and it looks like customers are asking for more.

Great. Thank you.

Thank you.

Our next question.

Yeah.

Comes from the line of <unk>.

Kevin Cassidy of Rosenblatt Securities. Please go ahead Kevin.

Yes, thanks for taking my question.

The supply demand.

<unk>.

Favoring the demand side is there any discussions yet for long term agreements with all of your clients and your customers.

Speaker #10: I'm just curious why customers are increasing the original specifications for HBM4 and then asking for higher performance? And then just a quick housekeeping question.

Yes that is.

Speaker #10: For Mark, just looking at your guidance, given the midpoint of all the ranges that you provided, it looks like to get to $375 in EPS, it implies that interest income is positive in the quarter.

Certainly interest in <unk>.

Some of these agreements.

But we have.

Certainly experimented with different types of long term agreements in the past.

And innovated with some ideas.

Speaker #10: Is that the case of what's implied there?

How to do some of these things.

I do think that.

Sumit Sadana: There is also, obviously, this very significant shift towards data center that I had been describing, and you have been watching that's been happening in the mix of the overall industry as well as for Micron Technology Inc. We are being thoughtful about all of these different factors, and we will definitely be leveraging these in the discussions with customers. What that comes out on the other end of these, I would rather not speculate at this time.

Speaker #4: Okay. I'll just answer the first part of your question and then hand it over to Mark. In terms of the HBM4 specs and its evolution, no doubt that the JEDEC specs for definitely exceeded by a significant margin by some of the numbers that we have quoted to you today.

B.

We are going to be very thoughtful about.

What we do here and what Tom and lifetime of these agreements we signed up for we also have U S manufacturing.

We are going to be bringing online.

We also have changes that are likely to happen on the landscape due to Paris, and we will have to respond to that when they become.

Speaker #4: With greater than 11 gigabits per second pin speed and 2.8 terabytes per second of bandwidth, we have sampled these products customers obviously are looking for as much improvement as they can get in the ROI that they can offer to their end customers that typically means that if there is a way to really drive higher bandwidth, you can increase the tokens you can reduce the time to first token you can get to a lower cost per token kind of a metric.

And when they are announced section 232 semiconductor that up so there's all kinds of.

Reasons to be thoughtful about.

Operator: Yes, understood. Maybe just a question on your decision to drop the managed NAND for the smartphone market. Is there a lower-cost supplier out there, a new supplier that is taking that type of business?

What happens to pricing what happens to how we create value for customers, whether it is through U S manufacturing or our product capabilities.

And there's also obviously, there's very significant shift.

Towards data center that I had been describing and you have been watching that's been happening in the mix of the overall industry as well as for Micron.

Sumit Sadana: It is not about a lower-cost supplier. It is just about, you know, for us, the ROI in the business. The NAND business ROI is certainly, has been for the past several years, lower than that of DRAM. Our approach has been that as we have strengthened our portfolio in NAND, particularly with the growth of our data center SSD business, we wanted to leverage that to improve the overall portfolio mix. We determined that the pricing expectations and the level of competitiveness in the mobile NAND market did not lend itself for robust ROI over time. We determined that it was better to exit that and concentrate our resources in other places where we can drive better ROI. One of the things that you have seen us do over time is, you know, dramatically improve our product portfolio.

So we are being thoughtful about all of these different factors.

Speaker #4: And that becomes a more attractive solution for their end customers. So that's really what is happening. But certainly the HBM4 specs that we have been able to produce will create a new bar for performance that would be needed to really enable these types of high-performing systems.

And.

And we will definitely be leveraging these in the discussions with customers.

It comes out on the other end of these.

I would rather not speculate at this time.

Yes understood.

Maybe just a question on.

The decision to drop the managed NAND for the smartphone market.

Is it is there a lower cost supplier out there a new supplier that is.

Speaker #4: So that's really what has happened on that front. And I'll pass it on to Mark to talk about the second part of your question.

Taking that type of business.

Speaker #6: Yeah, John, we do expect

Well, it's not about the lower cost suppliers, just about you know for us the ROI in the business the man business our ROI is.

Speaker #5: That interest to flip to income. We have greater capitalized interest. We have lower debt. We have higher cash balances; all those will flip that line.

Certainly has been for the past several years lower than that of DRAM.

Speaker #6: Great. Thank you.

And our approach has been that as we have strengthened our portfolio and then particularly with the growth of our data center SSD business, we wanted to leverage that to improve the overall portfolio mix and we determine that the.

Sumit Sadana: As our portfolio has improved, it has given us the flexibility that we are now leveraging to mix into those parts of the portfolio which offer the highest profitability so we can have a bigger share of the industry's profit pool. That has been our strategy now. We can drive it more aggressively with the improved portfolio that we now have. We have been investing to get these capabilities for several years now, and we are in a good place to be able to leverage it now.

Speaker #1: Thank you. Our next question comes from the line of Kevin Cassidy of Rosenblatt Securities. Please go ahead, Kevin.

Speaker #10: Yes, thanks for taking my question. You know, with supply and demand favoring the demand side, are there any discussions yet for long-term agreements with all of your clients or customers?

Pricing expectations and the level of competitiveness.

In the mobile NAND market did not lend itself for robust ROI over time and <unk>.

Determined that it was better to exit that and concentrate our resources in.

Mark Murphy: To add to build on Sumit's comments, we don't report or operate the business by technology, but we do view that we've completed a second year of positive free cash flow in NAND.

Speaker #4: Yes, there is. You know there is certainly interest in some of these agreements. But we have, you know, certainly experimented with different types of long-term agreements in the past.

Other places, where we can drive better rois. So one of the things that you have seen us do over time.

<unk> is dramatically improve our product portfolio and as our portfolio has improved it has given us.

Speaker #4: And innovated with some ideas on how to do some of these things. I do think that we are going to be very thoughtful about what we do here and the term and lifetime of these agreements we sign up for.

Operator: Okay, great. If I'm allowed to ask one more on that, just your smartphone customers didn't have a problem with that, but you know, doesn't affect your DRAM sales to them?

The flexibility that we are now leveraging to mix into those parts of the portfolio, which offer the highest profitability. So we can.

Have a bigger share of the industry's profit pool and that's been our strategy now we can drive it more aggressively with the improved portfolio that we now have we have been investing to get these capabilities for seven years now and we are in a good place to be able to leverage it now.

Sumit Sadana: No, I mean, customers are never happy when one important supplier is leaving a segment. I wouldn't say that they're happy to see us exit that. We have a very strong relationship with all of these customers, and the DRAM business is super critical to them, and they have far fewer choices on the DRAM side compared to what they do on the NAND side. We have worked with them to transition their products to other suppliers. We have supported them through that transition on the NAND side, and we continue to remain very robust suppliers to them on the DRAM side.

Speaker #4: We also have U.S. manufacturing that we are going to be bringing online. We also have changes that are likely to happen on the landscape due to tariffs, and we'll have to respond to that when they are announced: Section 232 semiconductor tariffs.

And maybe just to add to build on <unk> comments, we don't report.

Operate the business by technology, but we do view that we've completed our second year of positive free cash flow and debt.

Speaker #4: So there's all kinds of reasons to be thoughtful about, you know, what happens to pricing, what happens to how we create value for customers, whether it is through U.S. manufacturing or product capabilities.

Okay great.

Sure.

Yes.

Just.

Speaker #4: And there's also obviously this very significant shift towards data centers that I had been describing, and you have been watching, that's been happening in the mix of the overall industry as well as for Micron.

Your smartphone customers.

It didn't have a problem with that.

Operator: Okay, great. Thank you.

It doesn't affect your DRAM sales to them.

No I mean.

Sumit Sadana: Welcome.

Operator: Thank you. That is all the time we have for our Q&A session. This concludes today's conference call. Thank you for participating. You may now disconnect.

Customers are never happy when you're.

Speaker #4: So we are being thoughtful about all of these different factors, and we will definitely be leveraging these in the discussions with customers. What that comes out on the other end of these, I would rather not speculate at this time.

One important suppliers to being a segment so I wouldn't say that they're happy to see us exit that but we have a very strong relationship with all of these customers and the DRAM business is super critical to them and they have far fewer choices on the DRAM side compared to what they do on the NAND side. So we have work.

Speaker #6: Yes, understood. Maybe just a question on the decision to drop the managed NAND for the smartphone market. Is there a lower-cost supplier out there, or a new supplier that is taking that type of business?

With them to transition there.

There.

<unk> to other suppliers. So we have supported them through that transition on the manned side and we continue to remain very robust.

Suppliers to them on the DRAM side.

Great. Thank you.

Yeah.

Hello.

Speaker #4: Well, it's not about a lower-cost supplier. It's just about, you know, for us, the ROI in the business. The NAND business ROI has certainly been, for the past several years, lower than that of DRAM.

Thank you that is all the time, we have for Q&A session. This concludes today's conference call. Thank you for participating you may now disconnect.

Speaker #4: And our approach has been that as we have strengthened our portfolio in NAND, particularly with the growth of our data center SSD business, we wanted to leverage that to improve the overall portfolio mix.

Speaker #4: And we determined that the pricing expectations and the level of competitiveness in the mobile NAND market did not lend themselves to robust ROI over time.

Speaker #4: And we determined that it was better to exit that and concentrate our resources in other places where we can drive better ROI. So one of the things that you have seen us do over time is, you know, dramatically improve our product portfolio. As our portfolio has improved, it has given us the flexibility that we are now leveraging to mix into those parts of the portfolio which offer the highest profitability.

Speaker #4: So we can have a bigger share of the industry's profit pool, and that's been a strategy now. We can drive it more aggressively with the improved portfolio that we now have.

Speaker #4: We have been investing to get these capabilities for several years now, and we are in a good place to be able to leverage it now.

Speaker #5: And maybe just to add to build on Sumit's comments, we don't report or operate the business by technology, but we do view that we've completed a second year of positive free cash flow in NAND.

Speaker #10: Okay, great. If I'm allowed to ask one more on that: just your smartphone customers didn't have a problem with that? You know, doesn't affect your DRAM sales to them?

Speaker #4: No, I mean, you know customers are never happy when, you know, one important supplier is leaving a segment. So I wouldn't say that they're happy to see us exit that.

Speaker #4: But we have a very strong relationship with all of these customers, and the DRAM business is super critical to them. They have far fewer choices on the DRAM side compared to what they do on the NAND side.

Speaker #4: So we have worked with them to transition their products to other suppliers. We have supported them through that transition on the NAND side.

Speaker #4: And we continue to remain, you know, very robust suppliers to them on the DRAM side.

Speaker #10: Okay, great. Thank you.

Speaker #4: You're welcome.

Speaker #1: Disconnect. Thank you for standing by and welcome to Micron Technology's post-earnings analyst conference call. At this time, all participants are in a listen-only mode. After the speaker presentation, there will be a question-and-answer session. To ask a question during the session,

Speaker #1: You will need to press *11 on your telephone to remove yourself from the queue. You may press *11 again. I would now like to hand the call over to Satya Kumar, Investor Relations.

Speaker #1: Please go ahead.

Speaker #2: Thank you, and welcome to Micron Technology's fiscal fourth quarter 2025 post-earnings analyst call. On the call with me today are Sumit Sadana, Micron's Chief Business Officer; Manish Bhatia, EVP of Global Operations; and Mark Murphy, our CFO.

Speaker #2: As a reminder, the matters we're discussing today include forward-looking statements regarding market demand and supply, market trends and drivers, and our expected results and guidance in other matters.

Speaker #2: 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 documents that we have filed with the SEC, including our most recent Form 10-Q and upcoming 10-K for a discussion of risks that may affect our results.

Speaker #2: Although we believe that the expectations reflected in the forward-looking statements are reasonable, we cannot guarantee future results, levels of activity, performance, and achievements. We are under no duty to update any of the forward-looking statements to confirm these statements to actual results.

Speaker #2: With that, we can now open up for Q&A.

Speaker #1: Thank you. As a reminder to ask a question, you will need to press *11 on your telephone. Please limit yourself to one question and one follow-up to allow everyone the opportunity to participate.

Speaker #1: Please stand by while we compile the Q&A roster. Our first question comes from the line of Tom O'Malley of Barclays. Please go ahead, Tom.

Speaker #3: Hey guys, thanks for taking my question. My first one is just on the state of the NAND industry. So, over the last couple of weeks, you've heard a lot of reports about hyperscalers stepping in, you've heard about tightness, and pricing increases. If you look at your current quarter that you just reported, you had bids down in the quarter. I've tried a couple of different ways here; tell me if I'm crazy, but it seems like you need to have bids down again in your guide to kind of get there.

Speaker #3: So, largely driven by ASCs. Is that the right way to think about the world right now? And, like, some of your peers are talking about some big growth going into the back half, so is that something that you're going to look to ease up with potentially some capacity additions?

Speaker #3: Just anything that you can comment on regarding the state of the world right now. Thanks.

Speaker #4: Yeah, hi. This is Sumit. I just have a question. So in terms of the NAND industry, I won't read too much into the bids for FQ4; it's really just noise based on just a few things in different segments.

Speaker #4: Moving around in our mix, so it's really just some mix-driven noise for the various segments. In terms of the demand profile and what is shaping up looking ahead, certainly the large hyperscalers are looking like they will be needing significantly more storage capability for their AI server deployment.

Speaker #4: because they have shortages on that side from the HTD segment of the market. So, the deployment of NAND SSDs in servers and data centers more broadly is going to increase in calendar '26.

Speaker #4: So, we expect that driven by that tailwind, the NAND industry conditions will start improving. Certainly, our position in the data center SSD market, where we have been hitting record share after record share for several years now, gives us a really strong competitive position in the market. We expect to benefit from that data center SSD position that we have.

Speaker #4: And we have announced a lot of new products. We have a very exciting portfolio, so we expect to be able to leverage that going forward.

Speaker #4: So, I think the NAND industry will improve with the improvement in the DRAM business, which is definitely ahead from a time perspective. In terms of being tight, NAND is improving and getting tighter, but DRAM is tight already today and getting even tighter.

Speaker #4: going forward.

Speaker #5: Yeah, Tom, it's Mark. I would just add on the supply side, continue to, you know, we structurally brought down wafer outs and NAND, continued to slow node transitions there and, and, and pace, new node, new node ramps.

Speaker #5: We still have low levels of CAPEX, and so we're working down inventories there.

Speaker #6: Yeah, I was going to add, we're working down inventories, and now with our decision to exit managed NAND, you know, there'll be more supply availability for us to focus on the data center market, as Sumit was mentioning.

Speaker #3: Helpful. And then just as a quick follow-up, if I look at the CAPEX guidance, you're saying kind of annualized Q1, you get to $18 billion, and I just want to make sure you're saying that's a net number, so obviously something higher in aggregate.

Speaker #3: And in your commentary in the slides, it seems like there is very little additional NAND spend and more on the DRAM side. So just on a very high level, I know you're not guiding specifically. Should we be taking away from this that you're moving from the 13-something to close to the 20-something, and that the incremental spend is largely a function of additional DRAM?

Speaker #3: Thank you.

Speaker #4: Yeah, we're

Speaker #5: Going to give a net number, Tom. So we're saying that we're going to go from $13 billion to $8 billion net in '25 to, you know, a net approximately $18 billion in '26. The vast majority of that is for DRAM construction and equipment.

Speaker #3: Thank you, Mark. I appreciate it, guys.

Speaker #1: Thank you. Our next question comes from the line of Joe Moore of Morgan Stanley. Please go ahead, Joe.

Speaker #7: Great. Thank you. Sanjay, early on in the call, talked about taking market share in HBM in calendar '26, if I heard him right. But I know that would sort of happen mechanically because you got to your target share towards the end of this year.

Speaker #7: So you would gain share. Can you just talk about, you know, what your share gain aspirations are in HBM, if any? Are you content with holding your DRAM share?

Speaker #7: Does that need to go higher?

Speaker #4: Yeah, in terms of our HBM share, you're right. I mean, we have ramped our share through calendar 2026. And we expect to be in the vicinity of our DRAM supply share in calendar Q3 2025.

Speaker #4: So in 2026, for the full calendar year, we expect to have a higher share in HBM compared to calendar 2025. HBM has now become a very robust part of our portfolio.

Speaker #4: And we treat it like any other part of our portfolio, where we focus on ROI of the portfolio, we focus on disciplined investments, we're focused on getting to be the best from a product perspective, and creating that value with our customers.

Speaker #4: And then ensuring that we are taking full advantage of the competitive landscape as it evolves over time. So those are some of the things we do.

Speaker #4: So we're not going to provide any more specifics beyond that, but that's how we are thinking about it.

Speaker #7: Okay, that's helpful. Thank you. And then the gross margins in HBM, will that continue to be higher than sort of non-HBM DRAM? And, you know, given that you're locked in through most of 2025 and DDR5 is moving up, is it possible that that changes?

Speaker #7: Is that because that lock-in actually hurt you if DDR5 prices rise faster?

Speaker #4: Yeah, I mean, I don't think about the lock-in as "hurting us" because the two businesses, the HBM portion of it and the non-HBM DRAM, are, you know, just different types of businesses.

Speaker #4: The HBM business—we are working with customers with long lead times, long order visibility, lock-in of volumes, and lock-in of pricing well ahead of time.

Speaker #4: A lot more stable ROI over the years. So that kind of profile is helpful because, no matter what the rest of the business does, our expectation is that the HBM business will have higher through-cycle ROI.

Speaker #4: So, we manage that business in a different way. It creates a lot of value for our customers, and we have that expectation of a higher ROI for us in our portfolio.

Speaker #4: Now, will there be times when the rest of the DRAM portfolio will become very profitable and can challenge or even exceed at times the profitability of the HBM business?

Speaker #4: Of course, that is possible. We don't think of that as a bad thing; we think of that as a positive thing that the whole business has become dramatically better.

Speaker #4: And so, you know, that's how we think of it. Those outcomes are a good place to be because that means the whole industry is extraordinarily healthy from a supply-demand balance perspective, and that helps us drive really robust financial outcomes.

Speaker #7: Very helpful, thank you.

Speaker #1: Thank you. Our next question comes from the line of Jim Schneider of Goldman Sachs. Please go ahead, Jim.

Speaker #8: Good evening. Thanks for taking my question. Two quick ones: one follow-up on Joe's question, which is, you know, just in general, given all the performance benefits you talked about relative to your position on the HBM4, would you expect your HBM4 share to be higher than it was for HBM3 in a sustainable way?

Speaker #8: Going forward?

Speaker #4: I mean, we do feel we have very strong capabilities entering the HBM4 business. I mean, we certainly believe that we have the best HBM3e product in the world, with, you know, the highest performance and 30% lower power consumption than any other competitor.

Speaker #4: What we did not have at the start of the HBM3e business ramp, though, is that we didn't have that installed capacity and that base capability.

Speaker #4: We were ramping from very low levels. And now, with the very strong performance from our operations team, we have been able to get our capacity, our yields, our performance, and our quality to extremely good levels.

Speaker #4: We have met our market share goal that we articulated many quarters ago to you. As a result, we are starting off our HBM4 in a very different position in terms of being able to meet the expectations and requirements of our customers from a volume support perspective.

Speaker #4: So we feel that, from a competitive positioning, that aspect of the business is in much better shape. Now with HBM4, you know, there has been a lot of chatter in the market and the media about, you know, what is our HBM4 performance compared to those who are doing the base tie and the foundry, etc.

Speaker #4: And we just wanted to mention and lay some of those doubts to rest that, you know, we today believe that our HBM4 has the highest performance amongst any competing product out there.

Speaker #4: And we don't believe that others can match this performance with this level of capability, quality, power consumption, and so on. We believe that there will be tremendous customer preference for our products.

Speaker #4: Of course, we have increases in our HBM supply that we have created through our investments. And we have confidence that we will sell out that supply for calendar '26. There may also be some HBM share shifts happening.

Speaker #4: On the competitive landscape, it is our belief that most of that is likely going to be between our two competitors rather than impacting our share in any material way.

Speaker #8: Thank you. And then as a follow-up on the CAPEX side, could you maybe give us some color on the split of WFE versus facility CAPEX?

Speaker #8: You saw in fiscal '25, and then whether we should expect any significant change in that mix heading into next year. Thank you.

Speaker #5: Jim, it's Mark. Yeah, we're not going to provide that breakout, just that the vast majority of this spend is for DRAM construction equipment.

Speaker #8: And any significant change in the facility piece of that for next year, though?

Speaker #5: We're not going to comment on that. We're, you know, we're, as you know, we're expanding in a number of sites, including, we talked about, you know, our first Sab in Idaho and we're also equipping those fabs as well.

Speaker #8: Okay, thank you very much.

Speaker #1: Thank you.

Speaker #5: Or I should say, equipping the fab and then also doing tech transitions.

Speaker #6: You mean sorry, just to be clear, you're saying in Idaho we're in construction, and you expect production there in the second half of...

Speaker #5: Second half, sorry.

Speaker #6: 27. But we are investing in technology transitions and optimizing production across our footprint in Japan and Taiwan, where we have available DNA space.

Speaker #1: Thank you. Our next question comes from the line of Chris Dainley of City. Please go ahead, Chris.

Speaker #9: Hey, thanks, guys. Can you hear me okay? So.

Speaker #5: Yeah.

Speaker #9: So far, thank you. Thank God, I've been having phone problems all day. So, a quick clarification. As far as the DRAM reps go, you guys gave us the HBM breakout.

Speaker #9: Can we assume roughly the same ratio for DDR4, roughly high single digits, and then the rest DDR5? And any commentary on the relative contributions to incremental gross margin between, say, DDR5 and HBM?

Speaker #4: Yeah, so we had said earlier to you that the DDR4 plus LP4 we had provided, you know, some guidance earlier. We are pretty close to those levels.

Speaker #4: I'm not going to say that it's going to sustain throughout the fiscal year because, you know, the mix changes over time. But DDR4, just by itself—not counting LP4—has actually been a low single-digit percentage of the business.

Speaker #4: So it's been relatively small. But we have UL this; we have extended some of our UL timeline based on extreme shortages at customers. And of course, the margins have become dramatically better as this shortage has continued.

Speaker #4: But yeah, I think that's that. I think in terms of the relative margin, if I understood you correct, question correctly, relative margin between DDR and HBM, we are not going to comment on that because while the HBM, as I mentioned earlier, you know, they run like two different business models.

Speaker #4: So while the HBM pricing is set ahead of time for the upcoming calendar year, the pricing on the rest of the DRAM portfolio, and of course the NAND portfolio, is also moving around based on quarterly trends.

Speaker #4: And we have mentioned to you that the DRAM industry is tight, and as we get into 2026, we expect it to further tighten. Due to all of the robust demand that we are seeing and the limits on supply growth caused by a number of factors that were discussed, we believe that this tightening is going to enable improving pricing and margins on the non-HBM portion of the portfolio.

Speaker #4: So that will move around, so it's not really practical or possible to give you sort of a relationship between the two.

Speaker #9: That's great. That's still super helpful. And then for my follow-up, a question for Mark on gross margin. So, Mark, remember we had lunch earlier this year, and I asked you about, you know, could you guys get back to 50% gross margin?

Speaker #9: You said it's possible. Who knows? It depends on a lot of things. It's going to be tough. And here we are. And now we're going past that.

Speaker #9: So, can you just maybe talk about what changed or what was the key to that? And then, you know, what are the limiting factors on gross margin going forward?

Speaker #9: Is it just essentially a function of how many more quarters pricing keeps going up, or is there something else going on as far as mix goes?

Speaker #9: Thanks.

Speaker #5: So, Chris, since you know through the fourth quarter, the market conditions continue to improve. And, you know, as you know, we updated in mid-August the 11th.

Speaker #5: And you know we reported a margin 120 basis points higher than that. Update, so you know prices continue to improve; the market conditions are very good.

Speaker #5: We are very tight on DRAM, and they've improved in NAND and continue to improve. Yeah, the margin we reported is now above where it was in mid-fiscal '22.

Speaker #5: And DRAM is margins are higher than that mid '22 period. The operating margins are the highest that it's been since November '18. You know from here, you know the NAND business can continue to improve.

Speaker #5: There's been a good supply response, you know, by us. And, you know, we've focused on higher-value SSD products in that business. So while that business is below 2022 levels, it can continue to improve, and that will help overall margins.

Speaker #5: And then furthermore, as we talked about, DRAM is very tight now. And incremental supply is meaningful incremental supply is difficult to bring on. You have a number of a number of factors you know that are structural here.

Speaker #5: Our inventory levels are low, below targets. We've got this extended life on DDR4 and LP4, which is constraining the ability to do techno transitions. You have, obviously, the silicon intensity of HBM that continues to grow quickly.

Speaker #5: And then incremental capacity, as you know, to bring on a fab takes a long time. For example, our new fab in Idaho will come on, and meaningful wafer production is expected to start in the second half of '27.

Speaker #5: So that constrained supply focus on diverting bits to the best products, and also our continued good cost improvements, all offer the ability to expand margins. We've indicated that we expect second quarter gross margin to be up relative to first.

Speaker #9: Great. Thanks to your explanation, Mark.

Speaker #1: Thank you. I'm sorry. Thank you. Our next question comes from the line of Vijay Rakesh of Mizuho. Please go ahead, Vijay.

Speaker #7: Yeah, hi, Mark. Sumit and Manish. Just a quick question on the DRAM side. I know you mentioned one gamma is starting now, but if you look at your one beta, it looks like most of the HBM4 will be on one beta as well.

Speaker #7: Any thoughts on what the mix will be for you, one gamma versus one beta, now versus exiting the year fiscal '26, let's say?

Speaker #6: Yeah, sure. Thanks, Vijay. Manish, we are very pleased with one gamma ramp. We were able to achieve both mature yield this quarter as well as first revenue shipments into hyperscaler as well as other applications.

Speaker #6: So, we're really pleased with how that ramp is going. We expect to be able to have one gamma be the primary area where we're going to be that's going to be providing us pit growth for fiscal '26.

Speaker #6: In terms of supply, we're going to be qualifying across multiple different product areas as we go through the year. We are already at a point where the one gamma production plus the one beta production are the significant majority of our bid output.

Speaker #6: And that mix will continue to grow towards one gamma as we go through the calendar year fiscal year '26.

Speaker #7: Got it.

Speaker #5: Yeah, I would just I would just maybe add one thing. We talked quite a bit about supply constraints on the last question. And a very important part of our supply solution and in '26 is this ramping of one gamma, which is Manish said is going very well.

Speaker #5: And then, Chris, just one additional note on your margin question is, you know, is this just improved mix in our business? The data center and the higher performance requirements of that market and the favorable effect of that on the business?

Speaker #7: Got it. And then on HBM4, given your higher speeds and lower power, I guess it dives bigger to your own logic dive. Would you expect the margins on HBM4 to be better versus HBM3E, I guess, equal high?

Speaker #4: Yeah, I mean, we are very confident in the capabilities of our product, and certainly the HBM4 cost is higher than the HBM3 cost. The HBM4 price will be meaningfully higher than the HBM3e price.

Speaker #4: We don't really talk about margins by product line. Or, you know, within a product line, you know, which product has what kind of relative margin.

Speaker #4: But overall, for HBM, we certainly expect that we'll have really good ROI capability for many years for the company. Not just driven by the huge complexity of HBM, but also driven by the dramatic value it creates for our customers and our intent to, you know, ensure that we are benefiting from providing that value.

Speaker #7: Got it. Thanks.

Speaker #1: Thank you. Our next question comes from the line of Aaron Rakers of Wells Fargo. Your line is open, Aaron.

Speaker #8: Yeah, thanks for taking the question. I'll ask too as well. I guess kind of building on some of the prior questions, I'm curious. I know Mark, you had talked about your ability to execute.

Speaker #8: I think it was high single-digit cost down with HBM in DRAM in fiscal '25, and then I think it was teens or something.

Speaker #8: Low teens on NAND. I'm curious if, as we make these process node transitions, as well as G9 and NAND, how do you think about the cost-down curve as these process nodes materialize through this next fiscal year?

Speaker #5: Just to be clear, I think on cost downs, we guided for fiscal '25 or slightly better than high single digits for DRAM front end x HBM.

Speaker #5: And mid-teens for NAND. You know, I think as for DRAM, with HBM down low single-digit percentage and NAND cost reductions in the low teens.

Speaker #5: So these are consistent with the commentary we've given in the past that, you know, at that or better.

Speaker #7: Yep. Okay. And then I'm curious, just like structurally, the NAND market. You know, there's a lot of discussion around these high-cap enterprise SSDs pushing 250 terabytes and plus.

Speaker #7: You know, as we look at that market, I'm curious how you guys see the average capacity trending on these AI servers for SSDs and whether we should really be expecting that to inflect materially higher as we move through fiscal '26 and some of these high-cap drives really hit true volume.

Speaker #4: Yes, I think the average capacities are going to continue to escalate rapidly. We had 60 terabyte drives, and then 102 terabyte drives, and we have ourselves announced at our last SMS event 122 terabyte drives to 45 terabyte drives.

Speaker #4: And so we are very excited about the demand that we see for these drives. We do think there will be meaningful uptake of these high-capacity drives, and they will drive the average capacities higher.

Speaker #4: And I also mentioned that there is expected to be a shortage of hard drive storage for these hyperscalers, and so more meaningful usage of NAND is anticipated, even beyond previous plans.

Speaker #4: Should drive a positive outcome here, as well as an added tailwind.

Speaker #7: Where do you think the average capacity is today for servers? Just curious.

Speaker #4: It's moving around quite a bit because, in any given quarter, the mix of general-purpose servers and AI servers keeps changing. AI servers generally tend to use the highest capacity that is available in very meaningful quantities.

Speaker #4: So AI servers have been there are, you know, AI servers that are focused on just ensuring huge amount of capacity, but then with different workloads of inference that are also increasing more fragmentation is happening in the configurations that are getting shipped.

Speaker #4: But generally speaking, the highest capacity drives that the industry has been able to produce do get used by AI servers. So, they are going from 100 terabytes, they're going from 60 terabytes to 120 terabytes, and actively using a lot of 245 terabytes going forward.

Speaker #7: Yep. Thank you.

Speaker #4: Welcome.

Speaker #1: Thank you. Our next question comes from the line of Brian Chen of Stifel. Please go ahead, Brian.

Speaker #5: Hi there. Thanks for letting us ask a few questions. I know there are a couple of questions on cost down or cost reduction, but I'm just curious. You've obviously talked about, in the past, HBM3e and some efficiencies you've gained as you've kind of moved into relative maturity from where you started.

Speaker #5: And I know you're not, you don't want to provide too much detail around pricing in 2026 for 3E. While your tone does suggest you should have sufficient leverage in terms of those negotiations.

Speaker #5: But I'm wondering, is there more efficiency in terms of improvement in assembly and packaging yields for 3E that maybe could also help you out a little bit in terms of margins?

Speaker #5: Margin profile next year.

Speaker #6: Thanks, Brian. So, you know, we are very happy with how the HBM3e 12 high ramp has gone. You know, we talked about being able to get our yields to maturity on 12 high.

Speaker #6: Significantly faster than we did on 8 High, which was really our first major HBM volume node. And, you know, we've now gotten production capacity up to meaningful volumes, as we kind of gave you.

Speaker #6: So, you know, there will be continued opportunity, but I would say that, you know, really the biggest part of that benefit in ramp has happened as we've gone through the 12 high.

Speaker #6: You know, over the last couple of quarters.

Speaker #5: Got it, and I appreciate that. Maybe a quick follow-up. I know there's been some discussion here about sort of, you know, your guys, you know, reattending this 50% plus blended gross margin, you know, for the first time since 2017.

Speaker #5: 2018 timeframe. I know you wouldn't recast or you can't recast cloud and core enterprise server revenue back that far. You did give us sort of the, you know, the sequential and the year-over-year comparison.

Speaker #5: Today, but, you know, those segments are roughly mid-50% of sales now. What roughly did they represent back then? It must have been a much smaller proportion.

Speaker #5: And, you know, that's obviously a structural shift in the business period to period.

Speaker #4: Yeah, I mean, without giving you specifics for any one particular timeline, I'll just mention to you that, you know, most segments that are part of the TAM over the different time horizons in the past, prior to generative AI coming on the scene, had peaked around a third of the overall market in terms of segment size, right?

Speaker #4: So, if you think about the data center segment, or you think about the mobile segment, when they would go into this big spurt, it would peak around, roughly speaking, give or take, you know, a third of the overall market.

Speaker #4: And this time around, of course, you know that the pseudo ceiling of a third is no longer in place because data center has become more than half and continues to outgrow the rest of the market.

Speaker #4: And very robust profitability levels. So we do think that, you know, AI has radically changed the landscape. Not only is the mix due to AI growth in the data center cost, but the data center has become a much larger part of the TAM; it is also a higher value as well as a higher margin opportunity.

Speaker #4: We did mention that, you know, HBM and high-capacity DIMMs and the LP that goes into the data center, these three categories alone were $10 billion in fiscal '25.

Speaker #4: So a very large part of our fiscal '25 revenue. And these are all high-value add portions of the portfolio. This doesn't even account for regular DDR5 that is in 64 gigabytes and below type of DIMMs that are going into the data center.

Speaker #4: It doesn't account for data center SSDs that are part of the data center portfolio. So there is a lot of that shift that is going on, but it's also a shift that, you know, has a lot of legs to continue and that's also the demand coming from the data center that's creating a lot of this tightness in the industry and causing pricing to and enabling us to be able to increase pricing across all of the segments of the market, not just the data center.

Speaker #4: So, it's helping the profitability of the entire portfolio of the company. Even though it's driven largely from the data center.

Speaker #6: Great. Appreciate that.

Speaker #1: Thank you. Our next question comes from the line of Quinn Bolton of Needham & Company. Please go ahead, Quinn.

Speaker #10: Hey Mark, maybe I just missed your answer. I just want to clarify your answer to Aaron's question on the cost downs. You did low single digits in DRAM, inclusive of HBM, and low teens in NAND.

Speaker #10: Did you say that those are good ranges to think about in '26, or did you not give a cost down target for fiscal '26?

Speaker #5: We didn't give a we just gave $25.

Speaker #10: 25. Okay, thank you for that clarification. And then I guess maybe just kind of a bigger picture question. As you look at the inferencing market, NVIDIA recently introduced their CPX GPU that uses GDDR7 instead of HBM memory. And to the extent that that architecture takes off...

Speaker #10: Do you think that that, you know, lower HBM content on inferencing platforms, would that have any impact on your broader outlook for the HBM market over the next few years or do you think that the, you know, that that CPX architecture, you know, represents sort of only a small portion of the inference market?

Speaker #4: Yeah, as the AI market continues to grow and evolve, there are sure to be a lot of different use cases that come up, as well as many different types of workloads for which optimized architectures will need to be developed.

Speaker #4: And this is just another step in that direction. The AI market is going to become very, very big over time, and it is going to have needs that are not going to be one size fits all.

Speaker #4: And that optimization will be necessary to scale out the capabilities in a cost-effective way. So certainly, this is just one example; there will be more in the future.

Speaker #4: One important thing to keep in mind as all of this happens is that a lot of the inferencing workloads tend to be memory-bound workloads.

Speaker #4: And consequently, as inference becomes a larger and larger part of the AI market, it will likely account for 80% of the AI market over the years.

Speaker #4: A lot of that being memory bound means customers will be looking for different architectures that increase memory capacity and memory bandwidth. Depending on the latency requirements and time to first token, and those types of different KPIs that these architectures have to hit, different types of solutions will be appropriate for various workloads.

Speaker #4: And we are implementing a strategy to ensure that we are very strong in HBM, as we have discussed at length. We have an industry-leading product with our GDDR7.

Speaker #4: We are the pioneers of LP DRAM in the data center, which is going to be used more and more over time. Not just because of its energy efficiency, but because more industry players are going to, from an ecosystem perspective and from a CPU enablement perspective, start supporting LP over time.

Speaker #4: So all of these capabilities that we have developed, which are industry-leading, will serve us really well. Our intent is to create those high-value solutions across the rich frontier of evolving architectures that our customers will be driving.

Speaker #7: Excellent. Thank you.

Speaker #1: Thank you. Our next question comes from the line of John Vinn of KeyBank Capital Markets. Please go ahead, John.

Speaker #11: Great. Thanks for taking my question. I just had a couple of follow-ups. So, first on HBM, Sanjay [Mehrotra] on the call really went out of his way to say, "Hey, we can support 2.8 terabytes per second and 11 gigabits per second." It seems like these performance specs are greater than the original JEDEC standards that were set, and it looks like customers are asking for more.

Speaker #11: I'm just curious why customers are increasing the original specifications for HBM4 and asking for higher performance? And then just a quick housekeeping question.

Speaker #11: For Mark, just looking at your guidance, given the midpoint of all the ranges that you provided, it looks like to get to $375 in EPS, it implies that interest income is positive in the quarter.

Speaker #11: Is that the case of what's implied there?

Speaker #4: Okay. I'll just answer the first part of your question and then hand it over to Mark. In terms of the HBM4 specs and its evolution, no doubt that the JEDEC specs were definitely exceeded by a significant margin by some of the numbers that we have quoted to you today.

Speaker #4: With greater than 11 gigabits per second pin speed and 2.8 terabytes per second of bandwidth, we have sampled these products. Customers are obviously looking for as much improvement as they can get in the ROI that they can offer to their end customers.

Speaker #4: That typically means that if there is a way to really drive higher bandwidth, you can increase the tokens, you can reduce the time to first token, and you can get to a lower cost per token kind of a metric.

Speaker #4: And that becomes a more attractive solution for their end customers. So that's really what is happening. But certainly the HBM4 specs that we have been able to produce will create a new bar for performance that would be needed to really enable these types of high-performing systems.

Speaker #4: So that's really what has happened on that front. And I'll pass it on to Mark to talk about, you know, the second part of your question.

Speaker #7: Yeah, John, we do expect

Speaker #5: That interest to flip to income. We have greater capitalized interest. We have lower debt. We have higher cash balances. All those will flip that line.

Speaker #11: Great. Thank you.

Speaker #1: Thank you. Our next question comes from the line of Kevin Cassidy of Rosenblatt Securities. Please go ahead, Kevin.

Speaker #7: Yes. Thanks for taking my question. You know, with the supply demand going favoring the demand side, is there any discussions yet for, you know, long-term agreements with all of your clients or your customers?

Speaker #4: Yes, there is certainly interest in some of these agreements. However, we have certainly experimented with different types of long-term agreements in the past.

Speaker #4: And innovated with some ideas on how to do some of these things. I do think that we are going to be very thoughtful about what we do here and the term and lifetime of these agreements we sign up for.

Speaker #4: We also have U.S. manufacturing that we are going to be bringing online. We also have changes that are likely to happen on the landscape due to tariffs, and we'll have to respond to that when they are announced, specifically the Section 232 semiconductor tariffs.

Speaker #4: So there's all kinds of reasons to be thoughtful about, you know, what happens to pricing, what happens to how we create value for customers, whether it is through U.S. manufacturing or product capabilities.

Speaker #4: And there's also obviously this very significant shift towards data centers that I had been describing and you have been watching. That's been happening in the mix of the overall industry as well as for Micron.

Speaker #4: So we are being thoughtful about all of these different factors, and we will definitely be leveraging these in the discussions with customers. What that comes out on the other end of these, I would rather not speculate at this time.

Speaker #7: Yes, understood. Maybe just a question on, you know, the decision to drop the managed NAND for the smartphone market. Is there a lower-cost supplier out there, a new supplier that is taking that type of business?

Speaker #4: Well, it's not about a lower-cost supplier. It's just about, you know, for us the ROI in the business. The NAND business ROI has certainly been, for the past several years, lower than that of DRAM.

Speaker #4: And our approach has been that, as we have strengthened our portfolio in NAND, particularly with the growth of our data center SSD business, we wanted to leverage that to improve the overall portfolio mix. We determined that the pricing expectations and the level of competitiveness in the mobile NAND market did not lend itself to robust ROI over time.

Speaker #4: And we determined that it was better to exit that and concentrate our resources in other places where we can drive better ROI. So one of the things that you have seen us do over time is dramatically improve our product portfolio.

Speaker #4: And as our portfolio has improved, it has given us the flexibility that we are now leveraging to mix into those parts of the portfolio which offer the highest profitability, so we can have a bigger share of the industry's profit pool. That's been our strategy now.

Speaker #4: We can drive it more aggressively with the improved portfolio that we now have. We have been investing to get these capabilities for several years now, and we are in a good place to be able to leverage it.

Speaker #5: And maybe, just to add to build on Sumit's comments, we don't report or operate the business by technology, but we do view that we've completed a second year of positive free cash flow in NAND.

Speaker #7: Okay, great. If I'm allowed to ask one more on that. Just your smartphone customers didn't have a problem with that? You know, it doesn't affect your DRAM sales to them?

Speaker #4: No, I mean, you know, customers are never happy when one important supplier is giving a segment. So I wouldn't say that they're happy to see us exit that.

Speaker #4: But we have a very strong relationship with all of these customers, and the DRAM business is super critical to them. They have far fewer choices on the DRAM side compared to what they do on the NAND side.

Speaker #4: So we have worked with them to transition their products to other suppliers. We have supported them through that transition on the NAND side.

Speaker #4: And we continue to remain very robust suppliers to them on the DRAM side.

Speaker #7: Okay, great. Thank you.

Speaker #4: Welcome.