All right. Good morning, everyone. I'm Krish Sankar from TD Cowen. I'm the analyst covering Seagate, where we are fortunate enough to have Gianluca Romano, the CFO, and also Shianne Hudson from the IR team here. Seagate, obviously, as you know, one of the leaders in hard drives. Gianluca Romano, thank you very much for your time. While I'm on it, I'll also tell every investor on this, like, you know, please do vote for TD Cowen in Extel and Institutional Investor. Anyways, with that, let's start. I think, Gianluca Romano, I think what is kinda interesting is that I think there seems to have been some confusion last week, where some folks assumed that there was capacity addition. I think, you know, can you just clarify what's going on?
I think what has been pretty, like, you know, consistent with what you have spoken in the past is mid-20% exabyte growth, probably not real unit capacity additions. Has anything changed or is still that the narrative today?
Thank you, Krish. Always nice to meet with you. Before we start, let me say that I will be making four forward-looking statements today, and you can learn more about the risks associated with those statements on our website. Short answer to your question is no. We are not adding any unit capacity. We think our technology roadmap is really strong, so we can generate the exabyte that we need through technology transition, you know, going from fourth-generation HAMR, second-generation HAMR. We already discussed at our earning release about third-generation HAMR that is 5 TB per disk, a 50-TB drive that we will start qualifying at the end of next calendar year. Our roadmap is really good. We think we generate about 25% increase in exabyte year-over-year through technology roadmap, we don't need more units.
Good. What about head capacity addition? Is that a slightly different story compared to units, or?
Not really. I would say, you know, we produce heads and media. The heads to read and write the data, and the media is the disk. We also buy a lot of other components externally. The technology is in those two components as the media. Depending how you assemble the drive, you know, if you have a 5-disk drive or 8-disk drive or a 10-disk drive, for the same number of units, you can need a little bit different number of heads or media. Depend how where the mix is going. Generally, I would say our focus is keeping the units very stable.
Of course, increase the exabyte that we generate through this changing mix, changing technology, and transition, you know, every 18 months, 24 months to a new generation of HAMR that give us, in this case, you know, from first-generation HAMR to second-generation HAMR, give us more than 30% increase in exabyte. From the second to the third, another 25%. So.
We can grow through technology. We don't need more units.
Got it. Also along that same path, you know, when you look at it, clearly, demand is strong. I think last time you spoke about kind of, if I remember right, kinda sold out for all of 2027. Is that still the case? What is your visibility into 2028 and how to think about some of the LTAs that you assigned?
Yeah. We have, two different kind of agreements, you know. For the next four to five quarters, we have orders. An order has a specific product, specific volume, specific price, and time to deliver the product. We always want to cover the next four to five quarters because this is a time we need to produce an hard disk drive. You know, to produce an HAMR hard disk drive, you need about three quarters. We want, you know, three, four, five quarters that are fully defined. When we start the product in our manufacturing, we know exactly who will buy it and at what price. After that, we have LTAs that are agreement-based on exabyte.
To you know, for our customers, it's very important to know what kind of storage they can get two years out in time, three years out in time, even longer because they need to plan their new data centers. The price is less important. The mix is less important. They don't even know what product they will be qualified, you know, three years out in time. They want to know how many exabytes we will allocate to them so they know, you know, how many data centers they can build. We have agreement on exabyte. When we arrive into that, you know, four to five quarters range, we translate the exabyte LTA into an order. At that point, they know what is the product that they are qualified. We know how much we can ramp up that product.
We have usually, we have a list of products that they buy, not only one. We define the price, and we define exactly when we ship.
Historically, if I remember right, the pricing was negotiated on a quarterly basis, but now has it changed to an annual basis?
Yeah. It's depending from the customer and the duration of the orders, it's, you know, three, four, five quarters.
Gotcha. you know, the other interesting thing, I remember, is that, you know, in the past, you mentioned that 70% of hard drive demand comes from new data center openings. How do you track that? Obviously, there are a lot of things that go into a new data center opening. Obviously, they have to deal with, like, you know, permitting process, power supply, things like that. How much visibility do you get? In that build-out process, where do you come in? Are you, like, a late-stage purchase from a data center, build-out standpoint, early, requalified?
Yeah. We have two major segments, you know, Data Center and Edge. Data Center today is probably 80% of our revenue. It's growing. It's grown a lot in the last two or three years. In particular, public cloud is the sub-subsegment inside Data Center that is growing the fastest. We are already 80% of revenue, more than that in terms of exabyte. We see this trend continuing. You know, Data Center is growing faster than any other segment. We also see a good demand on the Edge. Today and the Edge is where we compete with NAND. Low-capacity drives, you know, two TB, four TB, eight TB drive, small. Because NAND price is so high, you know, that it's a good opportunity for hard disk to increase price. We don't have purchase order in that segment.
We can be more, you know, we can increase price faster. Demand is also higher. Again, it's not the majority of our business, only 20% of our revenue. Data center, much more structure. We have, you know, those, purchase order in place. We have a price is defined. Price is increasing, but it's a different kind of, trajectory. In the data center, as you know, we don't really compete with NAND. The NAND price has no influence on our demand. NAND is used to run the application on the compute side. A hard disk is used for storage. Very different applications.
Gotcha. I mean, it kinda makes sense because I think hard drives is probably up to 2% of a data center CAPEX, while NAND plus DRAM is probably up to 40% now with the price increase. You're not the problem child, you know, they think that memory is the memory price increase, i.e., DRAM/NAND price increase, slowing down data center build-out?
So far, probably not much. I would say what we try to do is not to be the main problem.
Yeah.
We don't want to be the component that is limiting the development of the data center. You know, in the past, you have seen the power being, you know, the bottleneck, probably still the bottleneck, the GPUs, you know, that were a bottleneck. Possibly, DRAM is becoming the bottleneck. We are always number, you know, two, three, or four in the list. You know, it's a good place to be. You know, we still have our power in the negotiation, but we are not limiting their development. We are not a problem for our customers.
You know, the other thing that kinda noticed is compared to last year, I think last quarter, which is obvious from your numbers and your competitors' numbers, the pricing is getting better or, you know, a little more better than historically. Historically, it was, like, more, like, mid to high single digits. Now it looks like it will be low double digit. Obviously, it is still not the problem insurance. You could probably even increase pricing further and not be an issue. What is the thought process behind price increase, and how sustainable is this run rate over the next couple of years?
Well, we know it's sustainable because we already have PO in places. We as we said at our earnings release, we see not only this quarter, but, you know, we guided more precisely. We also discussed about the next four quarters. We said every quarter, you will see revenue increase, and you will see profitability increase. Of course, a good part of that improvement is coming from pricing. We are executing a very good strategy that is not being super aggressive with price but be very, very consistent. Every quarter, you see that price up. Every quarter, you see that exabyte up, not units, but exabyte up. Revenue is growing, very well, and profitability is growing extremely well.
Mm-hmm. The other thing that is kinda interesting, when you look at last year, I would say the inflection for hard drives really started when you started seeing videos, AI videos and things like that. Today, how do you see that? Is that still a big trend, or is it, like, more data retention? What is the driver today and into next year for the next leg of, like, hard drive demand?
Yeah. I would say there are different drivers, you know. I would say the positive part for hard disk is that on storage, on hard disk, you don't need a different mix, different kind of hard disk depending from what kind of data you want to store. It's a good simplification for us in terms of what we have to produce and what we have to generate. I would say video AI is huge. retention, you know, started probably two years ago. When companies and people were starting to use AI, the first things we all did is stop deleting data. If you want to have a good result from running AI, you need to have a lot of data.
The data retention already started two years ago. Then you started to have AI generating data itself as an application. The beauty for us in terms of data generation and data storage is AI is very quick. It generates a lot of more data than humans. You know, it works 24/7. Every day, no problem. It generates data that you need even as a step through the final result. Everything that AI generates gets stored and then gets used again to generate something else until you arrive to the final result. A lot of benefit from AI. It's not the only application. You know, of course, robotics is becoming more and more a data generation application. You know, AI robotics will be another one, that is just starting today. Autonomous driving.
You know, you go to San Francisco, you go to Phoenix, Arizona, you see a lot of cars that are autonomous driving. They generate a lot of data every day. All that data is stored every day for compliance and to learn how to better drive the car in the city. When you see that expanding through all the city in the U.S. and outside the U.S., that is an incredible volume of data that will get stored.
You know, today, like you said, the majority of your demand is coming from hyperscalers. Is there an argument to be made that hard drive really benefits only from hyperscaler or cloud demand? Let's just assume in the future, we go to an enterprise AI world with, like, inference. If there's gonna be more on-prem deployment, would hard drives benefit, or are you going to more a cloud play rather than an on-prem play?
As I said before, we see demand in both segments, in the data center and at the edge. I would say it depends how AI will continue to evolve. You know, AI today uses a lot of the public cloud. It's concentrating in the public cloud. The Neocloud are compute. They don't have storage. You know, they take the storage from a big public cloud, an hard disk. They move into an NAND, into the Neocloud data center, run the application, and store back in the hard disk in the public cloud. They basically outsource or complement the compute part of a public cloud. If they want to become more independent from the public cloud and run their compute for other customers, they will have to build storage. At that point, we will sell hard disk even to the Neocloud.
At the edge, it's the same, you know, or on-prem. If you have a company that doesn't want to use a public cloud for many reasons, you know, efficiency cost or security of the information, the structure is the same. It's just smaller. Storage is hard disk, and then when they run the compute, they move the data into an NAND, run it, and send it back. It's just smaller, but same concept. Now, when we go to the edge, that will be interesting. You know, it's probably not happening, you know, tomorrow or the day after. When you go longer, you will have for example, autonomous driving. We require a lot of edge storage and fast compute. For sure, you will have, you know, a, a good application for the DRAM and the NAND part of the business.
They will also start parking data for a while until they send to the public cloud. That can be hard disk. We will see. I think there are a lot of opportunities for hard disk even at the edge.
Would you consider your VIA China business? Is it kind of, like, similar to edge, or is it different?
VIA is a bit different. I would say there are two parts of VIA. The video and image application, first of all, there are, you know, the cameras that are connected to a storage. That storage is usually an hard disk. It can be depending how big is the security that you're having. It can be 4 TB, 8 TB, 10 TB. Usually, it's a mid-cap hard disk, not a huge. The company that are providing the surveillance, they also offer storage. If you are a building like this one, you know, you have all the camera, you record everything, and then you say, "Okay," or you delete or you store somewhere. You can store with the same company that is providing you the service of surveillance. In that case, they buy big drives.
They buy 20 TB, 30 TB, 40 TB because they are actually offering a simplified cloud. There are not a lot of applications. There are some applications related to the surveillance, but not a lot of applications, but a lot of storage because the data is kept until they run the application, and then they decide what to do with the data.
Gotcha. Interesting. I'm just about to see if anyone in the audience had a question. If not, I'll chug along. You know, the other thing, like, when you look at your business, I think historically, your cost downs are probably, like, high single digits, 10%. Now it's more, like, low double digits, mid-teens. The cost downs are actually better than historical. Is that purely a function of moving to HAMR, or do you think there is a lot more room for this cost decline to improve, or do you think there's, like, a once-in-a-lifetime step down because of HAMR? Now we should think about more, like, 10% longer term.
I would say if you look at the product, the product cost going from first-generation HAMR 30 TB drive to second-generation HAMR 40 TB drive, the unit cost is fairly similar. You have 10 TB more. You have actually a fairly huge decline in terms of cost per TB. When you look at the entire company, the entire P&L, it doesn't depend only from the last product. You know, we sell from 2 TB drive to 40 TB drive. The cost in the period depends how this mix moves up. It's not only the last product, you know, with the time, moving more and more into HAMR, especially at the beginning of the technology where you increase, you know, 33% going from first-generation to second. Is a good you know, it's a good improvement of our cost decline.
When you go from 40 TB to 50 TB, in percentage, that is 25%. It's still the same similar unit cost. You add the same 10 TB, but in percentage, it's a bit lower. Depending how you calculate. I would say, of course, HAMR is advantageous for the cost. Every period is different. Every quarter is different depending on how many PMR we are still selling and what is, you know, the segment in that specific quarter that is getting more volume. It's not so linear. I would say HAMR, longer term, will be, of course, the way to reduce cost in the industry. Otherwise, with the old technology, with PMR technology, you continue to add one disk and two add, one disk and two add until you have space in the box.
The cost per unit goes up because you need to increase the bill of material. With HAMR, you keep the 10 disks, 20 heads, from 30 TB drive, 40 TB drive, 50 TB drive. The bill of material remains consistent. Of course, the components are not the same, so there is an increase on some of those components. The unit cost is fairly similar.
Are there any pressures on that? Because, you know, one of the things is, like, for example, you get glass from Hoya or someone, but they also have increasing demand from the optics and photonics folks. Is there any disproportionate pricing where other sectors are actually willing to pay a premium causing, you know, your cost to go up because you had to probably match that to secure those components?
Well, no. Supply chain, of course, is very important. As I said before, we buy a lot of components. We buy electronics. We buy memories that, as you know, are fairly expensive today. We buy a lot of other mechanical components. You know, of course, every year is different. You know, there are years with higher inflation, years with lower inflation, years where there is a shortage of one component, so you need to spend more. You know, my discussion is keeping those at the same level. You have this clearly, you know, strong reduction in cost per TB. As I said before, every period is different. You know, right now, we have memories, you know, costing more for sure. You cannot compare. It's not because of the technology. It's because of the components.
You need to adjust your estimate based on how you see those components, cost evolving, yes.
Gotcha. I think, you know, clearly, the demand is very strong. You're doing these cost downs. You're already at, like, 50% plus gross margin. I think people always speculated that it's a rational duopoly, so you should be at 65% gross margin. It seems like there's a path to get there easily, maybe exceed, but I think realistically, get to mid-60% gross margin. From your view, I mean, I'm not looking for guidance, but is that a fair assumption given the trajectory of demand and trajectory of your cost downs?
I would say it's an assumption of continuing to improve is very fair. You know, we actually said that, just a few weeks ago. We said, you know, there are this quarter plus other four of improvement. You know, this quarter, we are guiding at the gross margin that is about 50%. If we continue to improve, we continue to improve. We will go but we don't have a gross margin target. It's not a number that when we achieve, we just say, "Okay, this is enough." We stop there. We like how we run this strategy. We have done it for 12 consecutive quarters. This is the 13th. We have other four that we know are coming, and we will continue to do that.
One day, we will say, Okay, this is the result that we have achieved. We don't see an end at this point yet.
Gotcha. You know, the other interesting thing is you know, you said early on you're not adding unit capacity. You're still growing exabytes 25%-30%. At that CAGR, your hyperscaler customers are not pressuring you to add more capacity. They are happy with the 25%-30%.
I don't know if they're happy. I think there is pressure to do more. Their unconstrained demand is for sure higher. Every year, there is some components that get short. You know, right now, it's DRAM probably is short, and, you know, power is still short. From that unconstrained demand, you need to go from what they can really build.
Right.
It's our assumption. It's not their assumption. Our assumption is if we increase exabyte by, you know, about 25%, we will not be the component that is getting.
Right.
The development of the data center. Well, we could be wrong, but we think 25% is good enough not to become the top of the list problem.
Gotcha. I mean, the reason I’m asking is it seems like, you know, you’re doing 25% or 30% exabyte growth, slowly increasing pricing more than before, but your customers are not pushing back, so they seem to be okay. In other words, if you go to, like, say, 35% exabyte growth, you’re not the bottleneck. You know, your hyperscaler customer will still be constrained. The marginal benefit is not much for you or them for you to go to, like, higher exabyte growth, right?
I would say so far, this strategy has worked very well. We more than double our revenue. We more than double our profitability. There's no reason to change. I think as this strategy is giving us a great result, our focus is to continue to execute the same strategy for a long time. You know, we have already done 12, 13 quarters. It's already fairly long. We don't see this ending. We have another four quarters coming. We will have more later because, you know, we see the exabyte that our customers are demanding for year two and year three and year four is actually way higher than what we you know, what we have in our plan.
Right. Also, I think your own forecast is that I think probably next month or so, I think 40% of exabytes would be from HAMR. Maybe a year from now, 70% of your exabytes would be on HAMR. Is that still the plan? Are we or do you think that HAMR percentage is going to be higher than expected?
Well, we are ramping HAMR well, especially because the second-generation HAMR was qualified a little bit earlier than what we were thinking, you know, with the two major cloud customers. Of course, the % depends also from how much we produce on the old technology. Not to achieve those %, it would be very easy just, you know, not producing a lot of PMR, and you have a lot of HAMR. Depends. You know, we are trying to optimize exabyte. To optimize exabyte, we produce a lot of both, you know, HAMR and PMR. Let's see. You know, it's I think, first priority is to achieve this 25% exabyte increase. You know, you as you said, in the past, we were able to do more. That's more is probably coming more from PMR than HAMR. Now we are ramping more HAMR.
I you know, HAMR is ramping very well. I think we will be, you know, around those percentages, but depends every quarter how we manage the two technologies.
Gotcha. you know, I think if you do the math, you're probably going to generate close to, like, $3 billion of free cash flow this year. I mean, after the dividend payment, you're about $4 billion in debt. How would you prioritize the free cash flow? Is it, like, to mainly pay it on debt? Would you consider repurchases at the stock price, or do you think you probably need more supply chain preparedness for the ramp that it probably makes sense to invest more in your own business?
I would say longer term, we have as we have done in the past, our free cash flow is always focused on shareholder return, you know, between dividend and share buyback. In the short term, we have reduced our debt, you know, going from about $6 billion to now a little bit less than $4 billion. Now we can go lower. You know, we still have a little bit of the convertible, to buyback. We did something recently. You know, we will do the remaining pro possibly next quarter and maybe reduce debt even lower. Now longer after we have reduced the debt, you know, the free cash flow mainly will go to shareholders.
Gotcha. Gotcha. Is there any like, do you see, like, you know, in the past, like, you mentioned that I think heads or media takes, like, a year to bring it online. Even if you decide to add capacity today, you're probably looking at least a year before that becomes useful exabytes into the marketplace. Is that still the same, or do you think there is, like, more tightening potential in the supply chain where lead times can come shorter than today?
Well, I would say the cycle time of a wafer where we produce heads is about, you know, nine months, let's say, six to nine months depending from which technology, which product. If you need to build capacity like a greenfield, well, that takes more than a year. I would say possibly at least two years because, you know, you need to build the factory and then, you know, get the tools and qualify the tools. It's long, long term, yeah.
I mean, if you decide to add capacity, would greenfield make the most sense, or it'd be more existing brownfield?
Well, right now, we have not looked into that because, you know, we don't think we need to add units. I would say if we arrive at that point, in time of a certain decision, we will see what is the best solution. Again, in our plan, we don't have an increase of the number of factories that we have.
One other thing, you know, you spoke about cost reduction. You know, when I look at your footprint, you know, in manufacturing between the heads, the wafers, the media, everything, you're in, like, Singapore, Ireland, U.S. Is there a consolidation angle in this to help get the, you know, cost reduction, or do you think the footprint stays the way it is?
I think it will stay because we do different type of manufacturing in different locations. Singapore, we produce the disk. We don't have any other site producing the disk. Malaysia, we produce the substrate. We don't have any other place where we produce a substrate. China, Thailand, we do assembly and final test. Again, you know, China is used for our Chinese customers. Thailand is used for everyone else in the world. Probably the only two factories that are similar are the one U.S. and Northern Ireland, where in both places, we produce heads. Because we need that level of volume, I don't see any reason to change that.
Gotcha. You're close to running out of time. Just want to see if anyone had any quick question. If not, I'll try to squeeze one more in. You know, double-digit exabyte like, you know, mid-20% exabyte growth, and, you know, clearly, like, pricing is pretty robust. I'm just wondering, like, you know, there is an expectation that, you know, we should see double-digit growth in revenue-wise even in 2027 and beyond. If the current demand scenario continues, let's just linearly extrapolate it, is it a fair assumption that the revenue growth should be double-digit because?
Yeah. No, at earnings release, Dave, the CEO, said we expect revenue to grow for the next several years at least 20%. He said at least, so probably will be more than that.
All right. I think, Gianluca Romano, thank you very much for your insights.
Thank you.
Always fun having you. Thank you.
Thank you very much.