Good morning, everyone. Thank you for being here. I'm Samik Chatterjee. I cover the hardware and networking companies at JPMorgan. As you all get settled, I'll just introduce the next speaker here. Here for the next fireside chat is Dr. Dave Mosley, Chief Executive Officer of Seagate. Dave, thank you for being here. Thank you to the audience as well. Maybe, Dave, I'll start you off with a question that's more broader, more longer term. You've been at Seagate for nearly three decades, CEO since 2017. When you look at the storage industry today versus prior cycles, what is structurally different about the demand profile? I know everyone wants to say this time it's different. What do you think investors are still underappreciating about this current cycle that we're in?
Thanks, Samik. A couple things. Forward-looking statements, risk factors on our website. Get that out of the way. Yeah, it's been a long road for me and for our teams as well. I don't look at it as small cycles, 'cause I think every one is different. The last downside we went through was profoundly different, created by its own unique situations. I look at it more as a macro cycle. I mean, from my perspective, data's always growing underneath everything. Our industry went through a enormous build-out period through the client server days, up until peaking about 2012. For reference, we shipped 66 million drives in one quarter, they were all low capacity going into notebooks and desktops.
If you think about those times back then, the drives were not full. They were not being utilized very often, you know, just not turned on very much. What's different now is the drives are chock-full of those heads and media critical components that we'd built out the factories for back in those days. A lot fewer drives, but they're big beasts now. If you go into the data centers, they're working very hard, you know, 100% utilization and full. Very different profile for the industry as we've pivoted the supply chain for the last decade.
Okay. Okay. Maybe to get into that a bit further, you've talked about agentic AI generating much larger data sets than periodic queries, for example. Can you just go and sort of one level deeper into that, decompose what you're seeing in exabyte growth in calendar 2026, 2027, and how much is the broadening out of the use cases when it comes to training, inferencing? How much of that is becoming a driver of exabyte growth going forward?
Yeah, it's really hard to break out because the application space is growing so quickly. What some things are called, you know, inference one day, you know, may be actually developing into something that's actually learning more the next. I think we try to think about the First of all, the recovery that happened in the last three or four years was really driven by video properties first.
Yep.
It's not small text data sets. It's large unstructured data sets, which are usually video data sets. If you think about the training that happened early days, it was a lot of text for training large language models and being able to get utility out of them, and they are very useful, don't get me wrong. Now it's not about training on, you know, ages and ages worth of text that was stored somewhere. It's a lot more about understanding what's going on in the real world, and being up to the minute in your cycle and ingesting a lot of the large data sets and then being able to figure out how that is changing so that you can give the proper inference, and we're seeing this all the time.
If I look at the workloads that are across the drives in some of those applications, they're very diverse. There's not a predictable way of using them, and depending on the day or the user or whatever, it can actually change quite frequently, and that's what I would say characterizes the AI workloads more than anything.
Okay. Okay. You've mentioned moving to a build-to-order approach with your customers. How has that impacted visibility for the industry for yourself, and is that something that you're seeing across the industry being adopted as well?
It is. I mean, we're driven by very long lead times in our wafer operations, so we have recording head wafers that take a long time to build, you know, could be more than 9 months. The drives themselves take another, you know, quarter or something like that. Inside of that, we know what's coming out a year from now, and we've basically gone to the customers and said, "Look, if you wanna plan this really well, which you should be for your data centers, we know what's coming out. You can buy this stuff up to a certain period." We wanna keep that 4 or 5 quarters of visibility very, very solid for what's being built.
The demand is significantly higher than that. We know that if we can get a little bit more out of the factories, we're sensing the market. The market really wants the exabytes even more. We're now giving predictability into what we're going to be building in the next year or two years, three years based on technology transitions. The customer demand for that is still very strong. It's not really locked in. The build-to-orders are really only locked in for four or five quarters. Beyond there we have great visibility.
What's the financial implication of moving to this model? Like, for investors that have looked at Seagate over the cycles, how do you think about what changes on the financial side?
Well, what I really wanted out of it when we started was revenue predictability. Just, you know, we wanna make sure we're not building into a forecast that's not real. Now it appears that not only are the forecasts real, but the demand's significantly higher. You know, fundamentally, the financials will be driven by the demand. We're trying to answer that demand as aggressively as we can with our technology. You know, I think that's why the financials keep getting better. If we have predictability for four quarters, but then we start to roll through that period, and we go to re-up the next one, then we're seeing that the demand's even stronger, and so therefore the economics get better.
Okay. Okay. You mentioned this, lead times are now quite long. We've seen some of the press reports as well mentioning lead times are over 1 year in some cases for HDDs. However, when you take a step back, I think investors have always been concerned about the industry adding capacity right at the peak of demand, and by that time the capacity comes online, you're going through a moderation in the demand cycle. What's giving you the comfort now in terms of your capacity planning, and how are you thinking about your capacity roadmap?
There's a couple different kinds of capacity. If I go back to the industry 15, 20 years ago, the headway for the fabs, the lead time was only a few months. Very, very fast to turn. Now it's significantly longer with the technology being as complex it is as it is now. You know, that's really what we're gated by is those critical components that are long lead time. As far as the number of drives at the back end, we could be still fairly flexible on that. Because we're able to sell the capacity that's already coming out, we don't necessarily need to. How do I think about adding capacity now?
It's not just one machine you plug in, it has to be synchronized all the way across not only our production, but also the supply chain. It's really long lead time now and we're adding tools to make our capacity a little bit stronger inside of the manufacturing facilities that have those long lead times, but they tend to be the types of tools that will go through technology transitions rather than just pure capacity adds.
Maybe just to follow up on that, and this question comes up pretty often, so I'm sure you've taken this question before, is what fundamentally needs to happen for you to add unit capacity or floor space capacity rather than just depending on higher capacity drives to add exabytes?
Right. There is an answer to the question. I'll get to that in a second. What our customers are driving us for right now is more exabytes. We believe that the way to get the most exabytes is to take our talented teams and really go through these technology transitions. You know, we're targeting mid-20s% growth, which is enormous CAGR, the only way we're gonna get there is to be able to go through those technology transitions. If you will, to take a 3 terabyte per platter product to a 4 terabyte per platter to a 5 terabyte per platter year over year over year. That's really the way we're driving it. If we took the teams off and started building new factories or bringing up new machines, it would just take too long.
You'd end up more capacity, if you will, you'd slow the rate of growth on that technology. Back to your question directly, the wild card really is in unit capacity for disk drives, which we, again, could be fairly flexible with once we package those heads and media. That gets down to more customer diversification and edge AI and, you know, all those use cases, which I think could come someday. We would take the heads and media that we have planned and divert them somewhere else should those applications take hold.
Okay. Interesting. sovereign AI, neoclouds, the sort of next tier of cloud companies below the hyperscalers, what does that opportunity look like for Seagate?
Yeah. It's complex space. There's a lot of different terms being thrown around. The way I think about it is neoclouds generally tend to be compute, data centers, if you will, that are failover or being used by the major CSPs. Not entirely, you know, can be. There's not a whole lot of hard drives in there. Where do they get their data when they need it? They get it from major CSPs or the big installs that already exist in the world. Sovereign's a little bit different, because Sovereign could be more what I would call traditional data centers, not just compute data centers. In a lot of cases, people are saying, "I want my data right here next to me." That's the term Sovereign.
We are seeing some of those people build out, wanna have aspirations themselves. Some of them are still very beholden to the architectures of the CSPs, so they're asking the CSPs what to build, and they're just plugging that in. Some of them have their own architectures as well.
Is that the way I should interpret that, is you're not seeing them as direct customers, but more in terms of demand coming through your hyperscalers directly?
A few direct customers, but it's still very small. Yeah.
Okay.
I would say more of the Sovereign behavior has been, "Hey, I know somebody's gonna need this data center equipment. I'll build the shell, or I'll start plugging gear into the shell that I already have, and then, you know, I'll go find people that can use it at scale.
Got it. Okay. All investors today associate the Seagate investment bull case with HAMR. Maybe talk about HAMR, why that's a strong differentiation for Seagate, and do you think that differentiation is defensible against competitors?
HAMR is heat-assisted magnetic recording, in case you're not familiar with it. It's really our ability to use a different material set on our disk, iron platinum alloys. What I would say is that the problem with making the disks out of iron platinum is not just making them, but it's also being able to write them. We needed a large amount of photonics, integrated circuits, PICs, and lasers to be able to write spot sizes that are 20 or 30 nanometers in diameter. It's been a miraculous journey, really, frankly. We're now at the point where we've shipped millions of HAMR drives, not 1 or 2 million, many more than that too.
We've got qualifications, a little bit of new news here, is we've got qualifications against all planned CSPs now on Mozaic 3+. We've even got the 4 terabyte qualified at 2 places, which we've already announced.
Yeah.
The qualifications are going very well. We're very confident in the technology, and we can get it to 5 as well. This is our desire to bring more exabytes into the world. I can't really speak about competition right now because, you know, they'll probably figure out their own plans. I would say that, you know, as we keep working really hard to keep marching up the technology curve, I think this is the best way to answer what the world ultimately needs, which is more exabytes fast.
Okay. Great. Congratulations on the additional qualifications there. Maybe now thinking about the implications of HAMR from a financial perspective, how do you think about it leading to market share increases versus lower cost and driving more of the financial returns from a lower cost equation?
Yeah. I've said this many times. We don't really think about market share. I think about how many wafers I'm gonna start and whether I've sold them or not, and what the economics of that is. We're trying to get predictability not only for ourselves, but for, also for our customers, you know, for the next, say, 4 quarters or 5 quarters. To the extent that we go from 3 to 4 to 5, we're doing that without adding significant componentry. The cost leverage we get is really good, actually, and we get a lot more terabytes, exabytes out of that. So that's the way I think about the advantage to the extent that you're making a 40 terabyte or then a 50 terabyte, which we talked about in our earnings call we'd be doing by the end of next calendar year.
That's a huge value proposition for power, space into a data center if you're gonna be running that data center for five to 10 years. Like, that's the way our customers think about it. I think, you know, those products are gonna get very good looks from the customers, and we're gonna be able to book them predictably. There is another wild card in this, in that there's still a lot of capacity that's actually at 20 terabytes. If you think about how you're making a 20 terabyte drive today, how many critical heads and media it takes versus at 4 terabytes per platter, 20 terabytes necessarily in 5 disk. 5 terabytes a platter, it's 4 disk, so you're able to address those price bands with significantly better cost perspective.
That's the way we think about it, one of the reasons we're driving areal density so hard.
Got it. I have a follow-up on that, but I'll come back after this question because this will be related. Mozaic 3+ took years to qualify at the first hyperscaler. Mozaic 4+, you qualified it in much shorter duration, like PMR equivalent times at 2 of your largest CSPs. What specifically changed on the customer side? Are customers much more familiar with the platform? What's changing the timelines around the qualification?
Yeah. I think two things. The first is that customers, it's a new technology, so customers were maybe a little bit reluctant at the start. You know, anytime you're making big areal density jumps, people will look at it and say, "I have to make sure I check it out." They did. We had some early issues as well. We've got through a huge learning curve as we went up the volume ramp. I think the second thing is that there was a lot of noise about HAMR being different. The reality is, once I deliver the drive to you as a customer, it's not different. It's, you know. To the extent that you're not ready for a 30 terabyte drive in your, you know, architecture, the way your software runs.
Now I think people are. They're ready for 30, they're ready for 40, they're ready for 50. I think the adoption will be much quicker and we're trying to do this all very predictably. We'll start those head wafers when we know we have demand, and we know we can get through the qualifications quickly, and so on and so forth. That's what we're in the middle of right now.
Yeah. The expectation should be, which I had for you as a follow-up, is Mozaic 5 and 6 should be qualified in much quicker time with the same group of customers?
I think so, back to the traditional qualification cycle.
Okay. Going back to the last question when you said a lot of customers still are on 20 terabytes. Now, do you see them then taking the leap to 40 terabyte rise with Mozaic 4+, or still opting to go through that transition with Mozaic 3+, then to 4?
I think there are a lot of different use cases in the world where people really don't need that much capacity, depending on what's going on, especially in small data centers. Could be on the enterprise edge, which, you know, I know needs a little bit better definition out in the world. You know, our traditional enterprise customers may be running small, medium businesses or even large businesses, they'll still be able to use low capacity drives. Not everything will move to the highest capacity.
Okay. As you think about the improvements that you're delivering on cost per terabyte with the ramp on HAMR, you've talked about a 50% exabyte crossover to HAMR in the second half of calendar 2026. Once you get beyond that crossover point, what happens to the cost per terabyte curve? Like, what does it look like once you get over that 50% threshold?
Again, you know, the same number of components in the drive. You know, if we can keep pushing areal density, the cost stays relatively the same. The more terabytes we get out, the better. The demand for terabytes is so aggressive that I think some of our customers are willing to work with us on the, not only the new technology, but any new features that will get us there. With that fundamental, you know, cost floor, if you will, I think we're set up pretty well to continue margin expansion. That, you know, all comes down to exabyte demand though, right?
Got it. Maybe on the technology side, you're focused on HAMR, and that's one approach that we've seen. There are other approaches we've seen from some of your competitors in terms of looking at track layout being different. How do you think about what makes sense longer term? Do different technology approaches, including exploring alternative areas around track layouts, do those have a position in your roadmap as well long term?
Yeah, I mean, things like SMR get discussed a lot. We started shipping SMR, I think, 2013 when we went from 5 terabytes to 6 terabytes. And we've shipped literally hundreds of millions of SMR drives into desktop markets and things like that. There are other track layout tricks that, you know, may be useful over time, and we're already doing that for major CSPs as well, I think, just to be clear. It's HAMR Plus, if you will. One of the things that I'm most interested about the industry right now is that, yeah, you get through the transition of being able to use this new media type and getting the photonic circuits right.
There are fantastic new roadmaps now on PICs, on photonic circuits, and on lasers, and being able to use some of the other technologies that exist in the world and take pieces of that and apply them to areal density. There's a lot of ideas on the table now, and I don't think our areal density roadmap is gonna stall at all. I think it's gonna increase.
It will be largely building on top of the HAMR platform, but with additional innovation on top.
I think we'll be using iron-platinum for quite some time. It gives a much stronger signal to be read back at small bit sizes. The bits are about roughly 30 nanometers wide by 7 nanometers down track, so they're already pretty small. You have to make them even smaller, you know, the HAMR media gives you a great signal out of something that's that size. We have room to run.
Got it. Maybe just flesh out how you're thinking about density of the drive itself. Like, when you think about nearline being the clear driver of growth here, you've done a quite robust roadmap in terms of technology innovation. When you look past the roadmap beyond, like, Mozaic 6, for example, what are the big two or three technology bets you will have to take today to enable that?
Again, back to photonics, I think we're going to be pouring technology into making the world's smallest edge-emitting lasers and making them behave the way we want and as low cost as we can get them. Then there's a lot of technology to be brought to bear at the integrating the photonic circuits level. I think recording channels still wide open. A lot of progress being made in silicon. There's a lot of progress being made on things like GPUs and TPUs and learning so that we can actually apply that learning back to what's going on on each individual drive. That's one of the reasons I'm very excited about we have research efforts into all these.
Got it. When you say you will make those investments or efforts, is it all organic, or do you see that there's a need to do acquisitions to fill the portfolio on that front?
We're not afraid of doing acquisitions if we saw the right technology. We, you know, some of our technology has always been so bespoke to our application that we'll be doing it. It'll be organically led for sure.
Before prices of NAND went up where they are today, the popular bear thesis on hard disk drives was that you're ceding share to SSDs incrementally. How do you think about that sort of long-term roadmap, now, and is there overlap with SSDs when it comes to the warm tier of storage? Are you seeing that more and more becoming more prevalent? Or with the price increase in NAND, is that actually reducing in terms of how much you overlap with SSDs?
I think a lot of this comes back to what happened in the PC. You know, the PC forever, the capacity points in the PC were growing, and then they stopped, and they're still stopped, and it's still, the PC is still the same capacity point it ever was. When that happened, it transitioned from hard drive to flash and has been flash for a decade, right? That's the way I think about it. A lot of people, it's a very natural, emotional thing. You say, "Well, then the hard drives are going away." Not in the data tiers in the data center.
As a matter of fact, I think the data tiers there are not only stuck on hard drives, but they're stuck for a long way into the future, and that's because people know how to manage them really well, and the economics of replacing the amount of data that's in the data centers with flash, it doesn't make sense. I say this all the time. Flash is great technology. It has a lot of little niches to put product in. There's different kinds of flash. There's fast and there's more, I'll call it long-term flash. Can be tricky to manage and so on. Our customers in the data tiers in the data centers understand all this really well. They're some of the best storage architects on the planet.
They've already figured out exactly how they're gonna deal with hard drives and memory layers above. I don't see that changing. If anything, now that flash has gotten considerably more expensive, they're asking the hard drives, "How much more can you do at your tier?" The demand just keeps growing. You talk about zettabytes worth of storage capacity inside of disk drives. To replace that with flash would be ridiculous investment. It's not gonna happen.
Yeah. No, makes sense. Let's talk about pricing. Big shift for the industry, particularly now revenue per terabyte basis. I think you, your peer Western Digital is seeing the same thing. Just talk about the pricing power that you now see, particularly when you engage with hyperscalers in these pricing discussions. How has the balance of power in terms of pricing changed?
Yeah, I wouldn't say it's power. Fundamentally, it comes down to demand. If when we started out of the depths of the last downturn, we said, "Hey, we wanna get paid for what's coming out of our factories, we cannot just build it speculatively." Because of that, we locked in certain pricing for, you know, say, four quarters or five quarters back to the BTO discussion. As those roll forward, different customers are on different timelines, of course. You get to the end of that cycle you say, "Okay, now there's more demand, then let's up." We're still going for predictability, and that's what's driven that CSP pricing up.
The way we test this all the time is if there happen to be any swaps, anybody says, "I don't need product," or we get a little bit more out of our factory and can take the parts that we're already making and make a new different product out of them, then what happens is we go out and offer that to the markets, and we see where the new pricing is. That's the biggest reason for things to go up. We don't see that stopping.
Maybe let me take a step back and try to address the question what the more longer term focus is. When you look pre-COVID, most of your gross margins were in that sort of 20%-25% range for most years. Today, both you and Western Digital are progressing to 50% gross margin. You're almost there at this point. One part of it is the demand cycle, as you said. If the demand cycle were to change in a few years, is the company structurally a better margin company than what it was pre-COVID?
Yeah, I think given visibility, of course, you know, demand is the fundamental answer to everything, right? Given visibility that we have into the demand, We actually took a lot of supply offline by virtue of what we went through after the COVID downturn. Right now, lead times being what they are, that we can't really react on the upside to the demand. In some respects, if demand were to start going down, we have visibility into it. We would take necessary action, I think, again. I don't see that really right now. What's happening in the application layer is profound. The way I think about it, and I'm a hardware person, so you have to be careful with me.
You know, some of the software layers before that used to siphon money off the top are not there anymore. Now what's exposed is, hey, I've got this really fast application, it needs data, and I think that's more of what we're seeing in the world.
Okay. Interesting. What a lot of investors ask us about is hyperscalers have been increasing their capital spending plans every time they get on an earnings call, et cetera. With the kind of long lead time that you have, the build-to-order framework that you have, what's the typical duration between a hyperscaler saying, "Well, we'll spend more," to Seagate seeing that impact on the revenue line?
I think the hyperscalers understand what workloads are actually driving them, they are seeing these new application spaces. Say the video, for example. They know how much video is being created all the time at the extreme edge. They know how much video could be if it was enabled properly. They know the sources of data on the edge. You know, they know exactly how much people are uploading to their properties all day long. They get a pretty good sense of those businesses. There's new nascent spaces like people talking about physical AI and robotics that will need mass capacity. Is that gonna be serviced in the cloud or at the edge? Very good questions.
All these applications are being forecast into the future and we'll have the storage components as well. I think that's one of the reasons why they're giving us that 2, 3, 4-year visibility. Even though we haven't locked in the BTO orders, we are getting very good visibility and a nice, solid growth. That's another thing different about this demand cycle.
Got it. Got it. Okay. In the last 5 minutes, let's try and get through three questions here. A big talking point has been your 70% incremental margins that you've been delivering consistently. As you go through the ramp on HAMR and you continue to deliver more exabytes, how should we think about sustainability of these 70% incremental margins that you've been delivering?
Yeah. Again, fundamentally, demand will dictate everything. Our execution to the extent that we can, if we get through the 4 terabyte to 5 terabyte transition aggressively while the demand is still high, I think that'll serve us very well. We have all the assets in place that we need to do that. Our team is executing just fantastically. That's what it really comes down to is, you know, from an OpEx perspective. Our team's aligned on the right technology, knows exactly how we got to 3 and 4, now knows what we need to do to get to 5. We have all the tools in place. We've been planning this technology transition for quite a while. You know, I think it is sustainable.
Okay. Does pricing and how the pricing environment is give you a lever to even exceed the 70% on the incremental margin?
Yeah, pricing will always come down to demand. I do think that, you know, the difference between a 3 terabyte and a 4 terabyte and a 5 terabyte is pretty compelling if you're building a new data center or replacing the hundreds of millions of drives that are out in an existing data center. From a power perspective, the rest of the equipment you need to support that, the number of exabytes you're running, you know, I could see if the demand continues to remain strong, then, you know, we would have that kind of capability.
Okay. operating leverage. significant expansion on operating margins, I think 1,400 basis points year-over-year in the March quarter. You've been holding OpEx relatively flat. As the business now scales towards a $20 billion annual run rate by fiscal 2028, how do you think about the OpEx envelope that needs to support that overall revenue profile?
I mean, given the downturn we went through, our first priority was, you know, giving our people raises and they've deserved the bonus they've been getting, the variable comp and so, and so on. Again, we have the team, the technologists together to do what we need to do. We could spend more on OpEx over time, but, you know, I don't really see the need to right now. Hiring more people, great, but that would take a long time to train them. We are just really focused right now on using the existing team that we have to go get 5 terabytes of platter going, and then we'll see what happens in the future.
Does broadening out the customer base or investing in these new technologies that you're seeing, talking about photonics, does that impact the R&D pace or is that largely sort of similar to the run rate you've been on?
Far, we've been doing that inside of our existing footprint. And you know, we have thousands of thousands of engineers that are in tune with those technologies and markets. Again, we could if we wanted to. I still don't see the need right now. We love the footprint we have, we love the team we have, and we're trying to stay focused.
Okay. Let me end with capital allocation. Clearly, you'll be generating a lot of cash. How are you thinking about capital allocation, primarily in terms of how do you invest, where do you and how much do you invest in the business? How much do you allocate to M&A or buybacks? How are you thinking about those drivers?
Yeah, no real change. I mean, I think I said this on the earnings call that we just had, last year was all about patching up the supply chain, just getting working capital back. This year is about readdressing our debt, you know, getting in fighting shape in case there is some other downturn. We don't see that, you know, we just wanna be the best stewards that we can. Then, you know, we'll get back to being the Seagate that we want to, which is, you know, returning 75% of the cash to shareholders. We have a strong dividend, we always did. You know, we're gonna be reassessing exactly how that weighs off versus buybacks, we believe that it's the shareholders' money and we wanna give it back to them.
Inside of that, I think we have enough flexibility to do anything we need to for whether it's investing back in our own R&D or if we need to do small tuck-ins or something like that for technology access, we can do that. We're not really focused that way. We're more focused on just getting through this year, getting that 5 terabyte per platter launched and seeing what the next year holds.
Okay, great. I'll wrap it up there. Thank you for coming to the conference. Thank you to the audience as well.
Thank you.
Thank you, everyone. Thank you.