Please welcome Seagate's Senior Vice President of Investor Relations, Shanye Hudson.
Hello everyone. As you just heard, I'm Shanye Hudson. I have the great honor of leading Seagate's Investor Relations program, as well as the honor to be the first to welcome you to our 2025 Investor and Analyst event. I had the opportunity before coming up here to speak with many familiar faces and really appreciate you and the weather here coming today. Also, good to see some new faces in the audience. I'd like to acknowledge a few of our board members here today who are supporting us in New York City, and also for those of you listening in around the world on our webcast. I think we've got a great agenda planned for you, but if you'll bear with me for one minute, I'm literally going to read our Safe Harbor Statement. Okay.
During today's presentation and webcast, the company will be making forward-looking statements, which include any statements that are not historical facts. These forward-looking statements are based on expectations that involve risks and uncertainties that could cause actual results to differ materially. For further discussion of risks related to our business, please see our annual report on Form 10-K and subsequent filings with the SEC. Additionally, the company may refer to certain non-GAAP financial measures. You can find a reconciliation of these non-GAAP financial measures to the comparable GAAP numbers in the company's earnings release, filings with the SEC, and in the presentation that will be posted on our website later today. You can find that on www.seagate.com. Okay.
For the rest of the morning, you're going to hear from four members of our leadership team who are going to walk you through the tremendous opportunities that we foresee ahead for Seagate in today's data-driven world. Kicking things off, we've got our CEO, Dr. Dave Mosley. I think many of you know he's been with the company for over three decades now. He's been at the helm for nearly eight years, and I hope that when he's done, you'll understand why we believe Seagate is stronger now than ever in the company's history. He's followed by Dr. John Morris. He's our Chief Technology Officer. Of course, he's going to give you an update on HAMR technology. He's also going to talk about technologies that his team is working on to ensure that hard drives are relevant in the data center for years to come.
Following that, we've got our Chief Commercial Officer, B.S. Teh . B.S. is a veteran of the industry. He's a veteran with Seagate, and he's pivotal in creating the relationships that we have, deep relationships with customers both in the cloud and the edge. Finally, wrapping up the day, you'll hear from our CFO, Gianluca Romano. Gianluca will share the financial performance that you can expect from Seagate as we successfully execute on our strategy. We've got a brief break built in about halfway through the presentations, and I think we've left plenty of time at the end for Q&A. Without further ado, I think we can cue the video.
Please welcome Seagate's Chief Executive Officer, Dr. Dave Mosley.
Thanks. Good morning, everyone. Thanks, Shanye. She's the best, by the way. She put this on, and thanks to the whole team who did this as well. Appreciate you all here, local, brave in the weather this morning. I know it was tough. There are a few hundred people on the broadcast as well. Welcome, everyone. Thanks for sharing your precious time. I know it is precious. It's a crazy world. You're trying to figure it out. Hopefully, we can share our vantage point today. You'll get time for Q&A as well. Very happy to have you here. Let's get started. Today, you're going to hear a compelling story about the growth of data, enormous growth of data in data centers and elsewhere, and some new applications that are coming that we believe will continue to spur that.
From my perspective, that's the biggest part of the story, data growth. It's been a wild world. Data just keeps on growing. We're going to give you a good perspective of that. Underneath that, we're forming technology, innovating technology that gives us an advantage to be able to solution that data growth for the next decade and beyond. We'll share some details with that. John will share some details later about that. All of this translates into a sustainable, profitable business model that we're very excited about. Gianluca will round out the day and talk about that some. We're in the right markets, we have the right technology, and we have the right business model. That's our story. First, just a little background. Who is Seagate? We've been around for 45 years building hard drives.
Hard drives, the backbone of data in devices and in the cloud. From my perspective, a deep history in this industry and super proud of it. That will come out today. Proud of the technology, proud of what we do for the world. We continue to innovate. We've never really taken our eye off the ball. Seagate's always been at the forefront of this innovation in this industry. Many, many technical firsts along the way. Through this last period, it's been a tougher period, but we kept our eye on the ball and innovated, and we're going to talk about those innovations today. We're about 30,000 people. That's not all we represent, though.
There's hundreds of thousands of people upstream of us in our supply chain working very hard every day to continue to bring these products to the world so that we can all share in the benefits of what that data storage enables. In the past 12 months, we have shipped 550 EB. By the way, we will be jumping from terabytes to exabytes to zettabytes today. So 550 EB is about half a zettabyte. To put that into perspective, in the world's couple billion cell phones, that would store about two-thirds of the data that is on those cell phones. We shipped that in just one year. We are going to grow from here, given our technology. That is roughly, in a nutshell, who Seagate is. Okay. I will start with the overview of the markets. B.S. Teh is going to come up later, as Shanye talked about, and elaborate.
He's really a great person to listen to because he gets to talk to all these cloud customers and all kinds of other hyper, I'll say, edge cloud customers, our traditional customers. He knows the consumer markets. He knows the channels really well. Prep your questions for B.S. on that front. I will give you an overview of the market very quickly. I think it's instructive to just take a step back and think about what we've been through. We've all lived through transitions in our lives. Most of us remember the client server really well. Some of us built our own PCs back in those days. By about 2005, Seagate was cresting at about almost our first billion hard drives sold. The world had created just under a zettabyte of data. Remember, I just said we shipped about a half a zettabyte last year.
By 2005, the world was creating about a zettabyte of data. Since then, it's been a torrid pace. The technology transitions that we've all lived through are just monstrous. The mobile cloud transition, where we went from a lot of data that was contained in devices out at the extreme edge, in some cases just under our fingertips, on our PCs and in enterprises that were localized, translated into computers in your pocket and enormous data sets that were in cloud data centers. We've been living through that transition for the better part of 15 or 20 years. All kinds of different technologies enabling this, of course, right? Such that by 2020, the world was creating about 72 ZB a year. It's a fantastic number. Most of that data is not stored. Most of that data is not used.
It's created and it is dumped fairly quickly. That's actually interesting. Let's put a pin in that and we'll come back to it. With some new applications that are coming, we believe the data will become more valuable over time. That data is going to grow like crazy with these new applications. That's what we'll talk about today. Let's focus on just data centers for a second. We won't talk about the edge very much. Just focus on data centers. Data center growth for the last decade has been pretty staggering. Mid-20s CAGR. Right now, we're looking at the future and saying, with the applications that are coming that I referred to, mid-20s CAGR is about right. That's a huge CAGR. I mean, it's a huge growth year over year. To do that for a decade and then forecast that for another decade?
Right now, on this chart, we just look out for the next four years. It says that for us, nearline storage in data centers is going to be about 7.2 ZB of new data in the next four years. That's more storage capacity than the industry needed for the last decade. This shows just kind of smooth growth. If you know us really well, it hasn't been that smooth. I mean, there's been the pandemic went up and down. There was this undulation that started in supply and demand that got pretty profound at the bottom. All the while, if you average over it, this is the growth rate that we've seen. I'm going to make a case that says that those undulations are probably going to be dampened down a little bit in the past.
These are good questions that you can ask and we can talk about. The point is that that supply chain reverb is pretty much over, we believe. Part of the reason is we had, as the cloud was growing, too much capacity from client server. We have been fairly honest about that. Then we had to shrink our capacity yet again as part of the supply chain disturbance that we saw with the pandemic. The other thing is, as we move to new technology nodes, for example, in our wafer fab, the process content gets more and more. We are losing supply capability as part of that. All this stuff in balance, while data continues to grow, things became a little bit more strained from a supply and demand perspective. We were able to balance supply and demand for the last year.
We've been very, very happy with that. What that all means is, with our technology transitions coming, we believe we're at an inflection point right now. It's a good reason to do an analyst day because the revenue growth will now accelerate. We don't have way too much supply in our industry anymore. We've been fairly disciplined on making sure that supply and demand is healthy balance. We've been very thoughtful about this. Don't have that overcapacity of the past. Analysts have forecast that the revenue growth will translate from double digits in the past into mid-teens in the future. We actually agree with that conclusion. We need that money to continue the technology advancements, which I'll talk about in a second.
I think this is fundamentally reflecting the fact that the supply chains have rebalanced and we're a lot more disciplined coming out of the adventure that we just went through. Data continues to grow. Now, why? Why would it continue to grow at 20%, 25%, 30%, whatever the growth rate is in the nearline? Just to be convenient about it, for convenience's sake, let's call it all AI. New applications are coming in the data centers that will allow us to look through a lot more data, process that data, and come to conclusions about it, and then spin off more data to be stored later. The technology development that's happened on this front in compute and application layers has been really impressive in the last few years.
See, I always make the argument that 10 years ago, things were kind of boring from an application perspective. I think they're getting really exciting. There are some killer AI apps coming. When they latch, not all of them, but most of them will require a lot of data to feed them, a lot of snapshots of the data to make sure they did it right. They will generate a lot more data as well. Hard drives have a critical role in these applications. Ensuring trust and integrity is probably one of the biggest things, making sure that you have the right data to feed your models. It's kind of interesting because our customers say it's really hard to deconvolve what is AI and what's not.
In the last few years, the recovery that we saw out of the cloud was driven hardest by some of the video properties, some of the tools that we use every day on the internet, watching videos, video content that's created by creative professionals and then consumed later and stored forever, and then referenced back to what's best for your feed, your particular preferences. Those video properties grew quite nicely. A lot of our customers would say, "That's AI already." Maybe it is, maybe it isn't. It's really hard to deconvolve what exactly is AI and what's not. I believe some more of these applications are coming, especially video applications that are going to make things very interesting. From a storage perspective, we're really at the tip of the iceberg. I think that this mid-2020s CAGR is a good way to think about our business.
We'll be agile all the way through that, and we have a lot of opportunity. It is not all just about the data center, though. The data centers, they get a lot of attention. There are other trends that are happening, even in AI, that would dictate that mass capacity might be other places. I'll say local data centers. Think about that as cloud and edge. At the edge, you have smaller instances of data centers. I was meeting with a customer yesterday that they're exabyte scale. These people have to run their own data, keep control of it themselves. This data sovereignty issue around the world is a big deal. You can't all go to the big cloud service providers. The cloud service providers will grow, but so will the edge.
Out at the extreme edge, where data is actually created, things get very interesting. We know about surveillance cameras, but there are cameras in factories, for example, that are spinning off all kinds of interesting inference. Some of that data needs to be stored as well. There are opportunities everywhere. We are thinking about this now, rather than mass capacity and legacy, as opportunities in cloud and edge. We maintain some brands and some lines of business at the edge. For example, you could go to B&H Camera here in New York City and go see our LaCie line. That is keeping us connected to what is going on with creative professionals, video professionals. It is a great market. There are new applications coming. There is a lot of data there. Our FireCuda line in gaming, very proud of that.
It's small, but it keeps us in the mix of what's going on in gaming. Map sizes are getting bigger. I'm personally very excited about that. There are ways that us understanding the applications at the edge, the AI impact of the edge can actually benefit our business as well. Data sovereignty is everywhere as well. Look, this is not all about mass capacity storage. There's network stuff happening. There's compute. There's applications. There's memory. There's everything else as part of this ecosystem. It's driving a lot of innovation. We're very happy about it. We think it means a lot of data coming. All right. Markets are growing at a great rate. Mid-2020s is a great CAGR. We're going to be agile. Do we have the technology to keep answering the call? John's going to come up later to elaborate on this.
Let's dive into the technology. I think, as I said this at the start, I'm very proud of what we do for the world. We as a collective, it's not just us. It's our supplier. It's our entire industry. It's working with our customers. I believe our industry has built some of the most technically sophisticated products that man has ever built. They're really awesome devices. You'll get a flavor for that today. Let's start back. I talked about my history a little bit. When I started, we were working on drives that weren't even a gigabyte. That's a long time ago. By that 2005 timeframe that I talked about, where we started cresting about a billion drives shipped, our biggest drive was about a half a terabyte at the time.
We were working on a technology that actually we transitioned to in about 2005 called perpendicular magnetic recording. Big name. It really comes down to the alloys that we use on the disk. PMR has been great for us. We've learned a ton through that cycle. PMR fundamentally was running out of gains in the last few years. John will, like I said, elaborate more on this, such that about by 2020, our highest capacity shipping drive was 18 TB. By the way, we had a 14 and a 16 and an 18 and a 20 and a 22. Sometimes we're adding cost to get there. The gains were just getting smaller and smaller. All the while, we've been working on a change to that alloy on our disk. That change, we give it a big name called HAMR.
It really comes down to iron-platinum media on the disk. That change will allow us to jump to the next capacity point and the next capacity point. We are going to go from 30 to 40 to 50 fairly quickly compared to what we have done before. When 400 ZB is being created by humans in 2028, we will be shipping 50 TB hard drives. Okay? We talk about aerial density, and I am going to elaborate on aerial density intentionally for this reason. We talk about aerial density a lot in our earnings calls because it is the fundamental driver of our technology. We have just made breakthroughs, demonstrated the breakthroughs, and are showing a path to the future. It is intense R&D work.
HAMR's at the forefront of all of this, but in some sense, HAMR with the iron platinum media, it's kind of we're done, and we're very happy about it. Again, it's a good reason to communicate it. Our customers benefit tremendously from this because they can get efficiency. Let me walk you through that. First, what is aerial density? The hard drive itself is a very sophisticated box. If you zoom in on the disk and you look at this little mosaic of magnetic domains in this picture, you'll see that down at the atomic level, those things are about 7 nm in diameter. Those little columns that you see there, that's probably 25-30 atoms across. That's what we're engineering. It takes five to 10 of those things to make a bit. Okay?
We have to make them atomically flat, remember, because we're flying over them about 10 atoms off the deck. That is how impressive the technology is. Now you can see where we get words like Mozaic. That is our new product line for HAMR. It is really built on the mosaic of the tile floor, if you will. The aerial density is a couple hundred square nanometers then. A bit is a few of those columns, a couple hundred sq nm . That sounds kind of geeky. We do not talk about aerial density in a couple hundred sq nm . In order to enable this with iron platinum and everything else, we have had to go into silicon photonics and start shipping lasers. I mean, Seagate's been investing a lot in this. This technology transition has not been easy.
We're happy to say we're through it and on the other side. Rather than talking about square nanometers, let's talk about disks. That form factor, if you will, the size of that drive does not really change because there are hundreds of millions of slots out there for that drive to go into. We start with a disk, and we say, "How much can we fit on the disk?" In the PMR days, we were limited to about 2.4 TB per disk. Right now, we're on the cusp of 4 TB per disk. 4 TB per disk versus 2.4 TB per disk is one of the biggest jumps our industry's taken in the last decade. We are not done at 4 TB per disk. What's that mean in the drive level?
If you say roughly 10 disks per drive, that's a 24 TB drive, which is actually our highest volume shipping drive today. That's what we said on our last earnings call. Our highest revenue shipping drive, sorry, let me say it that way. 10 disks would get you to 40 TB. That's Mozaic 4 +. That gives better efficiencies in the data center. At the data center level, the fleet level, if you will, this is the way our customers think about it. They do not think about individual drives. They think about fleets. For a 1 EB deployment, like this person that I talked to might have, 42,000 units on 24 TB goes to 25,000 units at 40 TB. There is a very compelling reason in both power and space, which is huge for data center operators. Compelling reason to adopt the new technology.
When you're building your new data center, if you're going to run these things for five, seven, ten years in the data center, no pit stops, by the way. We've got to make sure we get all this right because at the fleet level, everything's important. That's a massive savings. It's actually about a 40% improvement on a 1 EB development between 24 TB, highest revenue shipping product we have today, and 40 coming real soon. It doesn't just matter in the data center, though. It matters at the edge as well. We can take this technology, and we've made reference to this in the past. We can take this technology and replace a lot of the prior programs that we had. For example, we had a very high volume running 20 TB program.
We can now replace the 20 TB program that's at the edge in the edge data center. In those markets that are really cost-sensitive or do not have all the technology to move to the highest capacity drive, we can plug and play right now into those markets and start addressing them in a much more cost-efficient way. This just highlights exactly what that looks like. If you had 10 disks in your 20 TB PMR drive, now you have 5 disks in your 20 TB Mozaic 4 + drive. Half the disks. That saves a significant amount of cost inside the box. We can incentivize that. It also allows us to run our factories much more efficiently, fewer disks going through our factories that have to go to meet that exabyte point. We can get rid of some of those old programs.
We do not have to support the old programs. We can transition to the new. This is what our industry has been doing for decades. We know how to do this really well. That is why we are so excited about it. All right, the tail of the tape here. On Mozaic 3 + today, we have three qualified customers out of the top eight CSPs in the world. Three. Our calls are going well enough that I can say within the next 12 months, we will have them all qualified. I am very excited about that. 3 TB per disk, is that 30 or is that 32? It could be as high as 35 or 36. Anyway, the reason is because there are so many different applications in the cloud that we do not just peg one capacity point. We have been talking about three.
In February this year, we shipped samples of our 4 TB per platter to customers. Next quarter, we're going to start qualification. It's in just a few months. We'll start qualification. We forecast that that qualification will be done by the end of the year or early next year, and we'll start the ramp. In the first half of calendar 2026, we will start the ramp of 4 TB per disk. A lot of leverage, a lot of learning. I mean, building all these drives has translated into a tremendous amount of learning, and we're very excited about what that means. We will drive the transition fairly hard such that by the back half of next calendar year, we'll be at about 50% of our exabyte being produced on Mozaic platforms. Okay? All right. We're playing in the right markets.
We've got the right technology that we can accelerate. How's our business model doing? This is what you all pay a lot of attention to. I would say since our last analyst day, we've gone to a point where we have about 12 months of visibility via these build-to-order model transitions that we've gone through. A lot of that happened because of the extreme disturbance in supply and demand, the undulations after the pandemic. We finally said, "Hey, if we're going to build it for you, we need to make sure you're going to take it from us." That instilled a lot of discipline in us and I think the industry as well. We have about 70% of our data center volume on build-to-order today. You might ask, "Why not 100%?" Maybe you can ask that in Q&A.
There's a reason to keep some back because some of our customers are never predictable. We're trying desperately not to overbuild, but still give them enough room to be agile themselves. We've crested a million shipped drives on HAMR now, on Mozaic 3 +. That's building with these qualifications that I just talked about, building quite nicely. Not only that, but we've learned a ton about the technology. We now know how we can get to 4 and 5 and beyond. As a matter of fact, we have visibility, and John will talk about this, but I don't want to steal his thunder too much, but we have visibility beyond 10 TB of disk with the HAMR technology.
It's not going to be easy, but I'm convinced that's going to keep us on a competitive cost trajectory that no other technology is going to supplant in the next decade, probably beyond. You can ask questions about that. Also, during this whole time, we've been able to stabilize and raise our gross margin to a much higher level than it was before, about 10% uplift in gross margin. Since the last analyst day, we spun off $4 billion of free cash flow all the while through that difficult period. What's that mean to shareholders? We've raised the dividend 11% since the last analyst day. We bought back 18% of our shares at a price of about $72 on average. We've returned $7 billion in value to the shareholders. This is who we want to be. This has always been our mission.
We believe in long-term investments in our industry. We're very confident about our industry going forward and our technology, very confident in the growth of the data, very confident in our ability to answer it. We fund ourselves, and we make sure we return value to shareholders. That's who we are. In that vein, staying true to that mission, we've announced this morning an additional authorization up to $5 billion of buyback when the time is right. This underscores our commitment that we've made to the long-term future that we have in this business. I've described these building blocks. We have the right markets that we're playing in. We have the right business plan. We have the right technology. It's getting funded. We're very excited about it.
Strong and resilient financial model that just got us through this place, a lot more discipline when the supply was finally reduced as much as it is, but still serving our customers with agility. A lot of that happens because of the team. You'll hear from some of the team members today. It's kind of the best part of my job is engaging with all these smart people, diverse talent on our team and in the suppliers and customers. It's wonderful. It's a blessing. Hearing from really smart people, what are the logical questions? I'm so proud of what we all do for the world. I hope that comes through. Thanks for taking your time to share with us today, to hear our story. Before John comes up, let's hear it straight from one of our customers. Thank you.
In the age of AI, hard drives are very important to us in the data center. We use them because we already have an established deployment of hard drives and deployments of other storage devices like SSDs and tape. Hard drives are still the most optimal, deliver the most optimal TCO. We see that still the TCO is about a factor of six or seven lower than solid-state drives. Higher capacity hard drives enables us to reach the desired capacity that we want in the data center with fewer devices. Fewer devices means lower power per TB. That means that we can either save energy or actually spend the energy in other parts of the data center. In the current systems right now before HAMR, aerial density increases were very minimal.
The HAMR provides us that needed ramp, this disruptive ramp to increase aerial density in a cost-effective manner. We will continue to deploy hard disk drives as the most cost-effective solution. I think that the future is bright. That innovation allows them to remain the most cost-effective solution for data center data storage.
Please welcome Seagate's Chief Technology Officer, Dr. John Morris.
Thank you. Welcome, everyone. Yeah, I'm real excited to be able to share a little bit more information on our technology, how it fits into our products, and how we're working every day to ensure that our products and solutions provide the most compelling storage choice for our customers. I'll double-click in a few areas that Dave brought up earlier. I'll be going through kind of three key areas in my presentation. First, I'll go through Mozaic 3.
Dave mentioned earlier that we finished qualification and ramp on Mozaic 3. We are in volume production now. It really forms a foundation for everything that we are going to do going forward. Next, I will go through Mozaic 4 and some of the other key technical elements in Mozaic 4 that allow us to grow capacity to 4 TB per disk. That includes the introduction of our vertically integrated laser that gives us a lot more efficiency and growth. I will talk a bit more about the future, Mozaic 5 at 5 TB per disk and kind of the path that we see going forward beyond 5 TB. One thing to keep in mind, the long-term fundamentals, and Dave talked about the averaging over time and how on average, mid-20% exabyte growth continues to persist.
As new applications emerge, we'll continue to see that growth for the foreseeable future. The long-term fundamentals are there driving demand for data, driving demand for storage as a result of that. We need to be mindful. How are we going to solution it? We looked at perpendicular over since roughly 2005, so it's been a 20-year technology for us. Periodically, the recording technology is kind of unable to continue to provide sustainable growth and offer that economy of scale that's necessary to be that compelling storage medium of choice. We get to these points where we have to find a transition. This is what our innovation pipeline is really focused on, is identification of these disruptive ideas that put us on a new curve for aerial density growth.
We're at a point now where we need something to follow up on perpendicular, and this is what we've developed with Mozaic and underlying that, the HAMR technology. Here you can see the last 15 years of perpendicular. 3 TB drives grew to 24 TB drives. The way that was done, there's an 8x capacity per drive increase over that 15-year period. In that 15-year period, we went from 5 disks to 10 disks. 2x of the 8x came from adding heads and disks. Fundamentally, that's adding cost. Now, it's effectively improving TCO, but we had to add that extra recording real estate. The other 4x came from intrinsic technology. The underlying recording technology provided that additional 4x. That's how we get to 8x. Especially in the last couple of product generations, it's really slowed down.
The capacity CAGR over this period was about 15%, but it's slowing down towards the end. We need something new. Fortunately, in our industry, we've been able to innovate and come up with new recording technologies when things run out of steam. This is exactly what Mozaic brings. You can see the next product offering after going through 2 TB per drive increments, the next jump is going from 24 TB to 30 TB with Mozaic 3 at 3 TB per disk. Let's talk a little bit about the building blocks for Mozaic. There are a few things to keep in mind. Dave brought up the transition in alloys to iron platinum, and it's an appropriate way to think about things that it is truly the foundation for Mozaic and the HAMR architecture is the underlying media. We use a granular iron platinum media.
The main point of transitioning to this new media architecture is it lets us shrink the grain size. We are able to shrink the bit size and translate that into aerial density. In order to use that, on the recording head side, we had to make some other fundamental changes. On the right side, you can see kind of a rendering of some of those key elements. We introduced a laser and a wave guide. If you follow the light down where the bright red spot is on the media, at the very bottom, there is a nearfield transducer, a very important element of the HAMR architecture. That nearfield transducer, essentially, converts light into an electric field that couples into the media. That exchange of energy is what provides the heating necessary to activate the HAMR recording process.
This whole process takes place on the order of a nanosecond, and it occurs at nanometer scale. There is a lot of expertise in many areas that become necessary to actually integrate these components into the drive. We have a highly integrated light path with a laser with optimized magnetic and optical components. Material science is instrumental in successfully pulling all of those pieces together, whether it is creating optical efficiency through the light path or the coupling into the NFT or materials that we use to provide the reliability of the interface and other elements in the head over the operating life of the drive. It is that level of expertise that lets us kind of pull all those parts together, manufacture in millions, and provide the kind of service life experience that our customers demand. There is a lot of complexity associated with all of these building blocks.
Ultimately, that does create competitive advantage that we believe could take years to replicate. I'm going to go into a little bit more detail now on Mozaic 4. Kind of the last part was really establishing that foundational technology that we can now leverage. One thing I'd like to highlight, if you zoom out from the inside of the recording head, you can see the next level of integration. Shown kind of in the center, there's a recording head, and on top of it in the circle is a light delivery unit that we use in our Mozaic 3 platform right now. We talk sometimes about recording heads as sliders. We fabricate the wafers that we use to make these recording heads at our locations in the United States and Northern Ireland. We put 20 of these assemblies in the drive.
One thing to keep in mind, just to get maybe a little bit of a grounding on the sizes, because these elements are quite small, in the circle, that light delivery unit is about the same size as a grain of salt. Our manufacturing process, in this case, requires us to sort of pick up a light delivery unit located on the top of the slider, and then we mount that slider onto a suspension. We have 20 of these assemblies in the drive, and we're building and shipping many millions of those every quarter. I want to go into a little bit more detail on the two different laser architectures that we're using on the next couple of slides.
Just to visualize that manufacturing process flow, there's an animation here, and it's showing a singulated slider, the recording head, and then dropping on top of that is a light delivery unit. One important comment I wanted to make is this process is done at the head level. The manufacturing process for this laser is done at the head level. We singulate wafers into recording heads or sliders. We have designed and built a high-volume piece of equipment that can attach the light delivery unit to the top of the slider, align the laser to the light input coupler, and do it at high volume and with outstanding precision. One thing that may come to mind here is this is a process that is operating at the individual head level, individual laser level. Can we do something more efficient than this? The answer is yes.
This is where our internally designed laser comes into play, where we have an architecture that is more capital efficient to do the same process than what I just described here. Last year, we announced that we do have a vertically integrated laser design. The previous configuration of laser we'll talk about is called a laser on slider configuration. With our internal laser, we refer to it as a laser in slider configuration. It does support wafer scale manufacturing. What I mean by that is we have a wafer with all the recording heads, and we have a wafer of lasers, and we can simultaneously pick up and attach hundreds of lasers onto the recording head wafer. It creates a far more efficient manufacturing process to do high-volume assembly of our HAMR heads. This is operational now.
We've already built millions of fully integrated heads with this laser architecture. As I said before, it gives us some measure of capital efficiency and also improved control of the other design elements, the other photonic elements in our design. Long term, it opens up some novel concepts in our technology pipeline. Just to kind of recap here, we have two laser sources, an external and an internal. Both are in volume production now. The laser on slider config, the one on the left, uses the external laser. The one on the right is our internal design. It is important to comment both are functionally equivalent. From the standpoint of our customer, they look functionally identical. One other important takeaway from this is we believe that high-volume precision photonic integrated circuits like this are the future of recording technology in our industry.
It is an area that we have built considerable expertise, not just in the design side of the technology, but also expertise in high-volume manufacturing at low cost. Okay. We are going to talk a little bit now about Mozaic 4. Dave highlighted Mozaic 3 in volume. Mozaic 3 is the foundation for our future platforms. There is enormous leverage as we look at the Mozaic 3 going into Mozaic 4 and 5 in the drive architecture, the electronics, the firmware, many of the design elements. Because of that, we get leverage as we go forward in time into Mozaic 4, delivering 4 TB per disc. Now, it is important to recognize that at 4 TB per disc, it actually creates a very compelling platform to refresh and replace industry standard capacity points, which sit conveniently at multiples of 4.
What that means then, effectively, as we begin to transition our customers from those legacy configurations to Mozaic 4, we get a multiplier on our supply capability. Every transition we take a 2 TB per disk product and move it to a 4 TB per disk product, we kind of double our internal efficiency. And that efficient supply of our exabyte demand is an important element of our strategy to service this 20+% EB CAGR in our industry. Dave also mentioned we have shipped limited 40 TB engineering samples to our customer already. We do plan to initiate calls next quarter, and we'll continue calls into 2026, where we'll be bringing over a wide portion of our customer base to the Mozaic 4 platform. All right. I want to talk a little bit now about extendability.
The next few slides are really going to go into why are we confident in the long-range viability of our products and technology. Here we are looking back about 10 years aerial density progress. In the figure, the green dots are experimental data points where we characterize kind of the intrinsic recording physics capability of heads and media. We do that on a fairly sophisticated piece of equipment that we refer to as a spin stand. It is basically an experimental setup where we can put a disc and a head outside of a drive, more or less, but we can characterize kind of the details of the recording capability of those components. It is a really good platform with a standardized set of rules, if you will, to characterize aerial density capability at the component level.
Those green dots are all spin stand experiments, and they follow a nice trajectory that we fit a line through. The dotted line just goes through all the green points. You can see from this figure the point in time that we demonstrated different capacity per disc densities and how long it took us to productize. It is about five years, give or take. Mozaic 3, our first demonstration, was about five years before we initiated the call that ultimately passed with our lead customer on Mozaic 3. Looking at the CQ3 timeframe for a Mozaic 4 qualification, that is just under five years from when a spin stand demonstration occurred. If I go forward and I look at 5 TB per disc, when was it done on a spin stand? I am just adding five years based on our track record.
That says we have an opportunity to initiate call for a 5 TB per disk product, Mozaic 5, roughly end of 2027 or early 2028. This is more or less how we build our roadmap, how we communicate our capability with our customers and kind of establish the cadence of qualification. If you follow the trend line, it gives us confidence that we can do a 10 TB per disk spin stand demonstration sometime in calendar 2028. It is not just the trend line that gives us confidence. It is a lot of underlying research capability that we have established over time, knowing roughly how things are going to scale, the geometries required, the materials we need to support the scaling, and so forth.
At the end of the day, we do have confidence that we can provide a path to 10 TB per disk in roughly this timeframe. Okay. We will take a look now at where we are positioning product intercepts, and I will repeat a few things here. I am going to quickly transition into where we see volume showing up on these products. The circles represent the point in time that we expect to initiate qualification. We have located 3 + about where that qualification started. 4 + is next quarter. 5 +, as I said, is end of 2027, early 2028 timeframe. Using a four or five-year rule for spin stand demo to product, roughly 2032 is when we would expect to initiate a Mozaic 10 qual.
Now, obviously, between 5 and 10, we do expect to introduce some capacity per disc improvements, and we'll work closely with our customers to understand what they actually need and what actually makes sense. We'll fill that out over time. Here I'm showing at the bottom, the chevrons at the bottom really highlighting when does volume occur for these capacity offerings. We're in the Mozaic 3 volume phase right now, and it'll continue. We expect to deliver up to 3.6 TB per disc on that platform, providing a 36 TB drive with volume in calendar 2026, so the leftmost product transition. If we look at where we're initiating quals with Mozaic 4 and about how long a typical qual takes, it says first half of calendar 2026, we should begin the volume transition.
It'll grow over time from a 40 TB config at 4 TB per disc up to 44 TB per drive in early 2027. Likewise, based on when we expect to initiate quals at Mozaic 5, 5 TB per disc, we would expect the second half of calendar 2028 to begin volume transitioning to that platform with our lead customers. We also expect to go from 5 TB per disc to 6 TB per disc over the service life of Mozaic 5 in roughly 18-24 months after the initial offering. We'll project volume expectations for products beyond Mozaic 5 as we get closer to the qualifications on those product lines. Okay. Now, why are we confident in our long-range capability? I'm going to highlight here a research result.
This is a less mature result than a spin stand experiment, but it is an important result in establishing proof of existence. The figure on the left just highlights recording three tracks of information at very high density. It is recorded at the density required for 10 TB per disc. It does not meet our normal requirements for a product or even a spin stand, but it gives us confidence that there is a scaling of technology that we can actually deliver this in a product. It will take time. There is still a lot of work in front of us to get there. This is just one other element in the work that we do to underpin our strategy. It feeds into the level of confidence that we have that we will get spin stand capability by 2028. All right. Some other elements of confidence.
First, we've already done 6.5 TB per disc in a spin stand at the end of last year. That gives us pretty high confidence in having line of sight to 65 TB drives. We've done a lot of investigation of the design space that we're operating in, this foundation of iron platinum media. We believe that granular iron platinum media is capable of scaling to 10 TB per disc. Even more than that, we believe that there's another level of extension of that granular iron platinum architecture that could theoretically get as high as 15 TB per disc, either with some new innovation in media architecture, the many layers of alloys and oxides that we put on the disc to create the recording structure, or possibly some form of ordering of the grain structure.
Beyond 15 TB per disc is going to require some level of disruptive innovation. We haven't ideated yet, but there are many ideas in the pipeline that are candidates for getting us there. Something like fully patterned media, as an example, could potentially open up that space above 15. The outlook is super promising. We do see long-range viability of the recording technology, a path to 10 TB per disc. We have high confidence in our product roadmap through Mozaic 5. Notably, the design space for granular iron platinum media that's in Mozaic 3 looks very viable to get us up to 10 TB per disc. In the course of pulling all these things together, we've developed world-leading expertise in magnetics, in material science, in recording physics, in photonics, all at nanometer length scales and nanosecond time scales.
We have harnessed and scaled all of that technology to be efficiently manufacturable at high volume. We see a pathway to get to 10 TB per disc. To summarize, we have been first to market and first to scale volume for our products with HAMR, with the Mozaic 3 launch. With our vertically integrated laser design and the manufacturing capability that we built to support it, we believe we have very capitally efficient, high-volume production at scale available to use and have underpinned Mozaic 5 at up to 5 TB per disc. This opens up another decade or more of product innovation and really solidifies the hard drive's role as the mass capacity storage medium of choice in data centers. I get to work with all of our customers on these products to ensure that they seamlessly integrate into their data centers.
That's the last part for me. B.S. Teh will be coming up after a break that we'll go into now, and he'll go into a lot more detail about our interaction with the customers and the markets. Thank you.
Thanks, John. I think we're going to take just a quick 10-minute break to stretch your legs, grab a cup of coffee, and then we'll start up at maybe 10 past the hour, if that's all right.
Just direct your feet to the sunny side of the street. Don't you hear the pitter-pat? And that happy tune is your stair. Life can be so sweet on the sunny side of the street. I used to walk in the shade with those blues on parade. But I'm not afraid. This rover crossed over. If I never had a cent, I'd be as rich as Rockefeller. Gold dust at my feet on the sunny side of the street.
Ladies and gentlemen, please take your seats.
Give me like 10 more seconds.
There is absolutely a revolution that's underway, and it's largely driven by artificial intelligence.
For many years, customers have been on a digital transformation journey. When you think about the gist of that journey, it's always been about the data. High-capacity drives are absolutely key. The ability to have high-performance storage with high-performance drives like those provided by Seagate are an incredibly important aspect of any generative AI system. We see four top use cases today with our customers: content creation, customer service, sales chatbot, and coding assistance. These are all hybrid workloads that require a combination of solid-state flash drives as well as hard drives, including HAMR-enabled mass-capacity drives.
AI success depends on data. It depends on good data, vast quantities of data for training. On the inference side, it creates a vast amount of data as well. AI requires a large, very large dataset to train your model and also a very large storage capacity for you to store the result of the model and also the result of your queries. Mass-capacity storage is essential to AI's success.
Please welcome Seagate's Chief Commercial Officer, B.S. Teh.
Good morning and welcome back. My name is B.S. Teh . I'm the Chief Commercial Officer at Seagate. It's indeed an honor and a privilege to have all of you here. Thank you very much for joining us, as well as all of you on a webcast throughout the world. I have the distinct honor and privilege to talk to our customers every single day, understanding their challenges, understanding their opportunities, and having discussions on how we can solve these problems together. You heard from Dave this morning about the vast opportunity, the immense growth in data. You heard from my colleague, John, about the incredible technology that we're investing in to really enable disk drives to be the mass-capacity storage of choice for our customers and for this application.
What I'd like to do for the next few minutes is to take you on a little journey from our customers' point of view. What are our customers seeing? What challenges do they want to solve? Why our solution is a compelling one for them? There are three parts that I'd like to walk you through. Firstly, AI, the buzzword off everybody's lips these days. You heard from Dave about the immense opportunity and the mass storage that's created. I'll talk about what our customers are seeing and why so much data is being created. I'll talk to you about why hard drives will be and continue to be the optimal mass-capacity storage platform of choice.
Finally, to wrap it up, to talk about our Mozaic platform and that it answers a lot of problems and challenges that our data center customers are facing, but also in the edge applications as well. You saw this chart from Dave. I mean, these are big, massive numbers. For 394 ZB. What even is that? Let's talk about applications and what our customers are really seeing. These are some of the vertical markets that we play in. These are markets that our customers operate in as well: financial services, video surveillance, smart factories, smart grids, hospitals, healthcare, and then, and then, and then. These are kegers of the data storage growth. Maybe I'll just give you a couple of examples. We operate factories throughout the world. Dave talked about that. We're not the largest. We're not the smallest either. We generate in our factory petabyte of data.
As we deploy machine learning and AI into our factory processes, the amount of data that's being generated and that needs to be stored is growing exponentially. Right now, our data is doubling every two years. Thank goodness we make disk drives to store all that. Right here in New York City, there are 20,000 surveillance cameras that are linked together in a DAS platform. The challenge is not about how many cameras and what it generates. By the way, these 20,000 cameras generate approximately 4 PB of storage of data that needs to be stored every single day. That's like about filling up 220 TB disk drives every single day. The challenge is not about how many cameras are being deployed. It's about how you process that. With AI, that ability with agentic AI, that ability increases exponentially.
Therefore, more and more cameras and more and more data can be deployed and stored. That only means an increase in data storage. We talk about all the different applications, but what is the underlying driver for the growth in data? Our customers have identified four key areas: richer content, higher resolution. More and more of the data has to be replicated and even transformed. More and more of this data has to be retained, whether it is because of regulation or different issues. Lastly, regulation, especially data sovereignty, is requiring data to be stored locally more and more. Let's walk through them very quickly. Just one example. In the last 150 years—yes, we were all not around then—fifteen billion images were created. With AI, that same number of images was created only in the last one and a half years. I mean, that is just mind-boggling.
Not only were those images created in the last one and a half years, the resolution of that is incredibly high as well. We're not talking about text and images anymore. We're now talking about video. A typical one-minute video—I think most of us on social media right now have evolved from social media tags to sharing images to now video reels, right? A typical one-minute video consumes 100 times more data than one high-definition image. Think about that: 100 times more data than one single image. Replication and transformation. Everybody wants everything instantaneously. Because of that, the data has to be replicated and deployed at locations closest to where it's being consumed or processed, right? The more the data is replicated, the more storage is going to be required. Not only that, each data can now be transformed easily because of generative AI.
I'll be sharing with you a use case from one of our partners in a little video after this to show you exactly what replication and transformation means. Regulations are requiring that data is being stored longer and longer. We've all heard about SOX, HIPAA, GDPR, right? All of those things are requiring data to be stored longer and longer. In AI, there's also another use case why data should be stored longer, right? It's because we have to create trust in the AI models. We need to store the model checkpoints. We need to basically store the inference data to really make the learning models a more efficient one and a more accurate one as well. We talk about data sovereignty and regulation. Over 150 countries in the world today have some form of digital protection laws governing the data.
Many are making it stricter as we go on as well. Data sovereignty is front and center of those digital protection laws. Now, what does data sovereignty mean? Very simply, right? Countries want their data to be stored in-country within their jurisdiction. That is all. If you look at where data is stored today, there are about 10,000 data centers worldwide. Over 80% of those data centers are only concentrated in 10 countries. Think about that. One hundred fifty countries say, "Hey, I want the data in my country," but it is all concentrated in 10 countries today. There is going to be a massive build-out of all these data centers to ensure that all this data can be stored within the borders of those countries that have those laws as well.
As I mentioned earlier, one of our partners, Adobe, has a very, very interesting and exciting use case. That is all about transformation of data from kilobytes to terabytes to exabytes. Please enjoy the video.
Firefly is a way for creators just to write a simple prompt. It really allows content creators at any level to create a generative AI. If they can dream it, they can create it. I think the biggest thing that we're seeing around Firefly is just the amount of content that's being created. We've had over 18 billion images created. We just added text to video to this latest model, and it's just taken off. The way people interact with Firefly is really unique. You could have an image that's, say, 1 MB in size. By the time I prompt video into that, I could easily get to 100 MB in size. You just basically type what you'd like to see. I'd like to see a brown labradoodle puppy playing in the snow.
Now, all of a sudden, I prompt that into video, and I've got multiple videos that I'm putting together. I've taken this from a very small image of just, say, a megabyte all the way up to maybe 100 MB or 500 MB by the time I'm finished with this campaign. If you now take it and you multiply it times 12 languages and then multiple frame sizes, you could easily see where you'd get up to over a terabyte in data. By the time you create 10 of these images and 10 of these videos for a storyboard, now I've given you something like a three-minute movie.
I mean, if you have not really had the opportunity to experience Firefly, please do check it out because it is really very, very fascinating and quite entertaining as well. Moving on, why are hard drives optimal? That is the question that people, that I get asked by many customers as well. Very simply, we have to start with looking at the architecture and how that is evolving because the architecture itself plays a significant role in deciding where the storage is going to be residing. Moving back several years ago, we were in a client-server era. The architecture then in the client-server era is what we call an aggregated model, a converged model where the networking, the compute, the storage, everything is all together. Call it in a single box, right? An example of that would be just like a PC.
You have your networking card. You've got your drive. You've got your CPU and all that, right? You bring it along everywhere. It's connected. As we move towards a mobile cloud era, it's still pretty much a converged architecture that's moving a little bit into what hyper-converged software defines. Largely, the architecture is such that basically all of the hardware building blocks are really together, right? As we move into the AI era, there are several challenges that are really dictating why this architecture is changing and evolving. Number one is scale. Number two is cost. Number three is technology advancements. All of this, each of this is really exploding much, much faster than we've ever seen it before. The beauty of having an aggregated storage is that you're able to scale the different parts, the different components of the architecture differently.
Therefore, you're able to take advantage of the technology advancements. Separately, you don't have to change the whole box. You can do it in a modular fashion. This is only just beginning. I think some of you may have seen the announcement by Dell just two days ago talking about the disaggregated platform that they're now launching. This is where the market is going. What does that have to do with Seagate? Hard drives are front and center of that. You've got storage at the edge. We've talked about all the different storage, all the different compute infrastructure that's being built. That's great. The more compute infrastructure that's being built and generated, the more storage that's going to be required. In a disaggregated model, there's a massive data lake that's out there somewhere that is being connected to all of these compute infrastructures.
These compute infrastructures cannot operate without this massive storage lake. As more and more of these compute data centers get set up, the amount of storage that's going to be required goes up increasingly as well. The optimal storage device and the storage technology for these data lakes are hard drives. The reason for that is simply because it provides the most optimal cost and performance and scalability. Let's take a quick walk into a typical application that many of us perhaps are very familiar with. This will give you a perspective of exactly when you see an application, where is that content being stored? Video reels. I talk about that a little bit. Many of us maybe pop up in the morning and look at our social media. The first thing that jumps up is, oh, hey, look, the video reel today.
What's the latest and greatest? I mean, the social media platforms want to make sure that when they pop up the video reel, you get it instantly, right? That data likely resides on the edge, on a local SSD, or a local hard drive nearest to you. Now, when that video reel is no longer viral, or when you upload lots and lots of your photos to be stored in the social media platform, where does it reside? Not on the local SSD. It gets pushed out to the massive data lake out there where exabytes and exabytes of data sit. This is where data which is, well, not frequently accessed, but infrequent and intermediate data is being accessed. That's really where all of this data is residing.
Now, in a world where advertising, targeted advertising, is extremely important, and many people make money off that, right? How do they get targeted advertising? By going into the information that each and every one of us has. That resides in the data lake. All of the AI and machine learning algorithms that these social media companies have basically draw on information and data that sits on the data lakes and then transforming that into the targeted advertising that becomes a product for many companies. I've shared with you the evolution of the different architecture, right, from an aggregated model to a disaggregated or modular model. I've talked about and alluded to the different storage platforms of choice. Dave talked about this earlier as well, right?
Why do we believe, and why do our customers continue to tell us that hard drive is a mass storage platform of choice? Very simply, a few key points. Number one, cost. When you're in the business of selling storage, as many of our customers are, right, the lower acquisition cost, the better it is, right? Today, the cost differential between NAND and hard drive, and just using QLC, not even TLC or other higher order of NAND, is at least six times. At least six times. This drive is six times cheaper than SSD. This separation will continue to exist over the next five years, right? You saw from the technology demonstration that John demonstrated. Increasing the aerial density will allow us to make that separation continue. Secondly, our customers are asking for scale, exabytes and more exabytes every single day.
If you look at disk drives versus NAND today, the scale is completely different, very, very different, right? In order for the NAND industry, and let's take a walk to 2028, IDC forecast, right? The amount of data that is going to be sitting on hard drive is 2.4 ZB, right? 2.4 ZB is needed by 2028. Now, in order for the NAND industry to then have enough capacity to fill and service the 2.4 ZB, it's going to require an investment over the next few years, $240 billion, $240 billion over the next few years just to have enough capacity to service that. By the way, the cost of the $2.4 billion that our customers are paying is probably between $20 billion-$30 billion. So economics may not really make a lot of sense.
IDC is projecting that 87% of the data in enterprise, 87% of the data in enterprise today sits on hard drives, 87%, and only 13% on NAND and DRAM. This is the equation that's likely to continue on simply because of the economics and the scalability that I talked about earlier. Our customers, when they think about the choice of storage and what they want to deploy in their fleet, there are a few things that they have to consider. Cost, very clearly, is a big one. It's not just about cost. It's about scalability. It's about, is there enough capacity of the devices out there to support what we need in the growth? It's not just about people saying, oh, yeah, I will have enough capacity. Do the economics favor the ability to support that scale?
If you look at the CapEx efficiency for disk drives, in order to scale to the mid-20%, the mid-20s growth in exabyte that Dave talked about, our CapEx investment is about 4%-6%. The CapEx investments that the DRAM and the SSD industry has to undergo is about at least 35% or more. Not only that, one other big equation comes into mind, which is sustainability. This is increasingly important because GPUs are sucking up a lot of power. You've heard from one of our customers, if we're more efficient, they can deploy the power elsewhere. Disk drive is the most sustainable storage device out there as far as the embodied carbon emissions overall, which includes manufacturing and application. It is because of all these factors that our customers continue to choose hard drive as the optimal mass capacity storage of choice.
Is data center really the only market that our Mozaic platform will go after? Absolutely not. You saw this chart from Dave earlier. We address markets from cloud to edge. The applications are a little bit different. Let's talk about cloud first. What is important to cloud? We've talked about power is important. Space is important. Sustainability is important. The Mozaic platform answers all of these big challenges that our customers are facing. If you deploy the Mozaic 4 TB per disk platform, compared to the average capacity that's being deployed today in the data centers, you get a 3.3 times savings in terms of the floor space. You get a 3.5 times savings in power efficiency. You get a 3.7 times savings in carbon emission.
These stats are real and nothing to sneeze at because these are implications in the billions of dollars that the data center industry is investing today. As I mentioned, it's not just about data center. It's about the edge. It's about the purpose-built server and storage platforms. It's about the network-attached storage for the different applications, whether it's in video surveillance, right, or the industry-standard servers that are being deployed for years and years for the on-prem solutions. The Mozaic platform addresses those because we have lower capacity devices that suit these applications as well, down to 20 TB at a 10%-15% lower cost that Dave talked about. Why? Because of the fewer parts that we have to put into the drive itself.
Even right down to the Edge IoT, right, which is the video surveillance drives, and that goes all the way down, or even the consumer drives, which goes all the way down to 10 TB, will be available on the Mozaic platform. In conclusion, the world continues to innovate. And so does Seagate. I've shared with you the immense opportunity that's ahead of us because of the growth in data generation and data storage. I've shared with you why hard drives have been and will continue to be the optimal storage platform of choice. I've shared with you the diversity of applications and markets that our Mozaic platform gets into. With that, I hope that this gives you a better understanding of the opportunity that we have and the work that we do with our customers.
With that, there's only one thing left for me to do, and that is to invite and welcome on stage my friend and partner, our CFO, Gianluca Romano. Thank you.
Good morning, everyone. Thank you for being here today. Thank you also to the people that are watching us on the webcast. I strategically chose to be after B.S. because I knew I could get the best introduction ever. It was very strategic. I will start my presentation explaining the impact to our financials of what has already been introduced by Dave and John and B.S. In particular, I will discuss our improved and new business model, and also the impact of the transition from the old technology PMR to the new technology HAMR, and how that will result in better profitability, better free cash flow, and finally, an enhanced return to our shareholders.
Let's start with the new business model. What is the base of the new business model? It's mainly the build-to-order. We discussed a lot in the last two years about build-to-order. Build-to-order has a lot of value for us. It's not only the predictability and the visibility on the timing, the volume, and the mix. There are also other important factors linked to the build-to-order. For example, we know what to build. And when we know what to build, we can right-size our capacity. We can right-size our internal resources. We can optimize our external cost. We do not produce product that at quarter-end has no demand or does not have an order attached. Then we need to move it out and impact pricing. There are a lot of benefits from the build-to-order.
This is why we really pushed, starting about two years ago, on really implementing this new business model. There is also an impact on pricing. In the past, several times, we had to discuss pricing with our customers at quarter-end for the next quarter. Obviously, not the best situation for us. Now we discuss pricing two, three, four quarters before we ship the volume. Much better situation. This also helped us with our pricing strategy that was a fairly small increase but very consistent through the various quarters. Every time we renegotiate the contract, the same product has a little bit more price attached. A lot of benefit from the build-to-order. This helped us to not only achieve but actually even beat our prior targets in terms of gross margin and in terms of operating margin.
We are today, in both cases, above the prior target. This is the past. It's a very good result, but it's the past. We always focus on what's next, what is the future, what will drive us to do even better. Data growth is a primary driver for this business, for Seagate, for the industry in general. The main driver for this industry to continue to grow and to expand in terms of revenue, in terms of profitability, is actually data growth and the exabyte that will continue to increase. Dave talked about our expectation of data center to grow about 25% CAGR. If you look at the last many years, like 10 years, in terms of exabyte volume, we only had one down cycle. That happened after three years of a strong increase in a period of time that was not really normal.
It was more the COVID period. There was also a reason why in that period the data center exabyte grew a little bit faster than what was before. They had to readjust. You can say, exabyte is one thing and revenue is another thing. Why revenue didn't have the same trend? The main reason is what is driving the business is the green part of the bar is the data center. When you go back 10 years ago, the vast majority of the business was on what we call legacy part of the business. For that part of the business, we had a lot of supply. The industry was way oversupplied. This is why revenue couldn't grow at the same pace of the volume. A few years ago, in 2020, there was a crossover.
The data center volume was equal to the legacy volume. Starting that time, it continued to grow. This year is more than 80% of the exabyte. We expect that to continue to grow. That is the main driver of the business. On top of taking benefit of this volume increase and because of what we discussed before on the build-to-order, we are now also able to capture more value from all the exabyte that we sell. As we said before, we have more visibility. That means we can have a better cost structure. We can have a better price negotiation. Very importantly, we can move the mix to higher capacity drives. The better pricing, the lower cost, the mix moving to higher capacity drives is helping us to get more profit from each exabyte that we sell.
We expect that to continue in the future. Until now, the impact of the HAMR transition has been still very small. We did not sell a lot of volume in a single quarter. We just started basically last quarter to have a fairly good volume. Before, not a lot. Not so much that could really impact our P&L. Part of that improvement that you see in the slide going into the future is actually related to HAMR and how HAMR will improve our financials providing higher capacity drive. That means lower cost per terabyte. In Seagate, we all love HAMR for different reasons. John likes the technology. He likes the laser. He likes the media. He likes the heat and the cold and the atomic size of the media and how we increase the aerial density. He is a technology genius. It is good.
I appreciate that. I like HAMR for a different reason. And the reason why I like HAMR is this one: cost reduction. This indicates how the cost will decline starting from the PMR product, our last PMR product, going to the first 3 + TB HAMR product, but more importantly to what happened later: the 4 TB, the 5 TB per disk, the 6 TB per disk. Imagine the millions of units that today are sold with a 24 TB PMR when we will sell those in 40 TB HAMR drive. Look at the difference in cost. It is enormous. That is our opportunity for the next two or three years. What happened to our gross margin until now? In the last two years, without the help of HAMR, before the big help that we will get from HAMR, our gross margin almost doubled.
We went from a 19% two years ago to 36% last quarter. We already have line of sight to 40% gross margin in the next few quarters. We ramp more HAMR, and we achieve that result. Of course, we need to also improve a little, increase a little bit our revenue. We see that coming. We discussed that at our last earnings release. We have a plan to increase revenue and to increase profitability through the rest of calendar 2025 and to achieve a very good gross margin level. Until now, in the last two years, mix has been a very good driver of this improvement. The mix was still based on PMR. Going from 16 TB to 18 TB to 20, 22, 24 was a good help, but still not comparable to what HAMR will do for us.
With the build-to-order, we have the opportunity to reduce the cost, improve the pricing situation, and also eliminate or almost eliminate the underutilization charges, having better visibility on what we have to produce. HAMR has been a small part of this improvement, basically only in the last quarter, but will be a big part in the future. I think I can stop here. Thank you. What is happening to HAMR volume? That is a big question. Dave already talked a little bit about when we will crossover. Even before that, by the end of next fiscal year, we will have 40% of the nearline exabytes sold with HAMR drives. We will exit fiscal year 2026 with 40% of the exabytes sold through HAMR products. We will crossover in the fourth part of our fiscal 2027, that is our second part of calendar 2026.
By the end of fiscal 2027, we will be at about 70% of exabytes sold through HAMR products. Of course, that ramp will start to decline. When we go into fiscal 2028, we will go to 80%, so another increase. The big part of the ramp will happen in the next eight quarters. Very, very soon, starting today, eight quarters from now, we go up to 70% of the exabytes sold in the nearline space through HAMR products. On top of focusing on our improvement in the product margin, we also focus a lot on OpEx and CapEx efficiencies. For the OpEx side, we have decreased a lot our OpEx from $1.4 billion to the current run rate of $1.1 billion. Huge decline.
We expect for the next two or three years to be at about OpEx at about 10% of revenue range. We will still improve that ratio. The CapEx, I think we are already in a very good range. Now, 4%-6% is all focused on technology transition, on moving from PMR technology to HAMR technology. We do not need more heads. We do not need more disk. We need to have more terabyte on each disk and in each unit that we produce. This is how we increase the exabyte sold without increasing the unit sold and without increasing the cost to produce the heads and the media. The next slide is the model. You remember the 40%. I said we will be at 40% very soon, 40% gross margin.
I also said we have a lot of opportunity to capture the volume, the exabyte volume into our revenue through the build-to-order. Our expectation for the revenue is to grow at a low to mid teens as a CAGR between fiscal 2024 and fiscal 2028. We also expect our gross margin starting at 40%, that we will be there soon, to continue to grow. The incremental margin will be 50% starting from a revenue of $2.6 billion. Let me be very clear. We will be at 40% gross margin fairly soon. At a revenue that will be approximately $2.6 billion, which we discussed already at our last earning release. Starting from that point, our incremental margin based on the increased volume from HAMR will be 50%. We discussed already about OpEx at about 10% of revenue.
We discussed about CapEx, 4%-6% of revenue. In terms of our capital return, we will return at least 75% of the free cash flow to our shareholders. Let's talk a little bit about capital allocation. In the last five years, we invested almost $4 billion in R&D. Very well spent. We have HAMR. We have this great opportunity in front of us. We also spent about $1.6 billion in CapEx, again, to facilitate the transition in manufacturing from PMR technology to HAMR technology. We returned to our shareholder more than $7 billion, about three in dividend, a little bit more than four through share buyback. Dave said we repurchased 18% of the shares outstanding at an average price of $72. Our board authorized an increase of the current share buyback authorization up to $5 billion.
We will start to do share buyback only in the next few quarters. As I said before, we are focusing right now in reducing our debt. We have already taken big steps in the last few months. We are almost there where we want to be. Fairly soon, we will restart our share buyback program. We can do all this because we have a fairly strong balance sheet. We have $2.1 billion in liquidity. We have reduced our debt at about $5 billion. That was our initial target. We have now a net leverage of 2.1 times. We see this leverage continue to improve in the next several quarters. To conclude my presentation, I just want to remind what is really important to our business and to our financials. First of all, data growth. That is the primary factor for this industry.
Now, through the build-to-order and through HAMR, we can efficiently capture more value from all the exabyte that we sell. Those new products based on HAMR, they will provide a better TCO to our customers, higher capacity, better TCO, and will provide a better profitability, a better return to us. This means we can generate higher profitability. We can generate higher free cash flow. All this, finally, will enhance our shareholder return. Thank you. The Q&A session.
About to go. I think give them a chance to get up to the front or wherever, wherever you want to start.
Good morning, guys. Thank you for having us. Erik Woodring from Morgan Stanley. I'll be annoying and ask a multi-part question. First, Gianluca, just clarification on the low -to -mid teens revenue CAGR.
Just clarification kind of on the base that we're thinking off of because calendar 2024 versus fiscal 2025 obviously make drastic differences. So just one clarification there. And then Dave, for you, obviously a compelling case for aerial density gains, mixed shift to larger capacities. That also probably means the market will need fewer disk drives versus the, call it, $120 million annual run rate the industry is at today. Can you maybe just help us better understand how you plan to manage that unit capacity such that you can maintain favorable supply demand? And also making sure, obviously, that that doesn't mean you run into any underutilization costs or risk associated with that. Thank you.
Yeah, the model is based on fiscal year 2024 as a starting point.
All the information that has been included in my presentation are starting from fiscal 2024 and are modeling up to the end of fiscal 2028.
Eric, to your question, it's difficult math to translate exabytes into units. And different customers think about this very differently. When we talk about a build-to-order model, we do talk units and exabytes with them. Obviously, there's a pull to make, since they're going to be running this stuff in data centers for a long time, there's a pull to make those as efficient as possible. They want the new platforms. If we can't get them there because we don't have enough capacity in the pipeline, we may choose to fall back to the old as well. We are driving aggressively through this transition.
It's not clear to me yet that the units is going to drop off too much as an industry. It may stay fairly stable depending on what the ultimate demand is. We'll have to play that by ear. Our desire is to drive to the higher capacity points and make sure we satisfy the exabytes that way. You got the mic.
Cool. I guess I'll ask a very quick follow-up, which is just John, Luca. I'll keep going too. Pass me your questions. Just John, Luca, very quick on any more color you can share on kind of free cash flow margins, free cash conversion. Obviously, a really compelling case for a lot more free cash generation, buyback authorization much larger than you've announced in history.
Just any more detail we can better understand for what that means for, again, margins or free cash conversion, however you would want to characterize it.
Yeah, free cash flow has always been a focus of Seagate even during the down cycle. Now we were able to generate enough free cash flow to protect our dividends. I think you will see starting next fiscal year, so basically next quarter, a much better correlation between profitability and free cash flow. We will use part of that free cash flow also to restart share buyback at a certain point in time during the fiscal year.
Hi, it's Krish Sankar from TD Cowen. I have two questions. First one for Gianluca, or maybe John. Historically, your cost declined to the per terabyte basis at even the low teens.
How should we think about them heading into the next four or five years with HAMR? Is it a similar range, or is it better? I have a follow-up.
I would say it depends on the mix. We have shown a fairly huge decline in cost per terabyte, going from 3 TB per disk to 4 TB per disk to five and six. Of course, depending on a specific period of time, how much mix we move from one platform to the next, you can have a significant cost decline. It depends also when we qualify our customers on the different platforms. It is not so easy to give you an answer on an average. I would say it is just a matter of time. We will qualify, as Dave said. We are starting to qualify customers on the 4 TB per disk next quarter, a few months from now.
The more and more we go there, you will start to see the first major declining cost. From that on, every platform is really giving us a good upside.
Got it. A quick follow-up for John. Historically, higher capacity hard drives have had lower IOPS. People have used dual actuators, but it looks like hyperscalers do not like it. I am kind of curious, do you think that changes? Also, with HAMR, when you were qualifying HAMR, I think hyperscalers wanted similar failure rate as PMR. Would that be the same as you go from 3 TB- 4 TB per disk, or would that requirement change dramatically? Thank you.
Yeah, two parts. Performance. Maybe just a little clarification. IOPS per drive is pretty constant. It is more a function of just the geometry and the mechanics in the drive.
IOPS per terabyte does decrease as drive capacity goes up. In some architectures, that presents some challenge in integrating the highest capacity drive. Generally speaking, we work with our customers to help them identify ways to integrate the highest capacity drive in their infrastructure. Like I said, it varies by customer, whether they have to change anything in their architecture or if it is fairly straightforward to drop in. By and large, across our customer base, there is quite good traction to take the highest capacity drive and integrate it in their architecture. I would say, generally speaking, we expect good traction as we transition from 30 to 40 to 50 TB drives. Having said that, we try to provide solutions that help our customers with that integration. We did introduce dual actuator drive as one potential solution. It is still available if they require it.
At the end of the day, they'll optimize their architecture for lowest cost. They have been able to architect up to now without widespread adoption of a dual actuator config for performance. Like I said, if the time comes that they see that platform as their lowest cost path for high cap drives, we'll make it available again. Second part of the question on reliability. Generally speaking, we're going to operate with a similar five-year reliability profile as perpendicular drives and working with our customers to make sure that their experience with Mozaic adoption is comparable to what it has been historically with drives up to now.
It's coming. Perfect.
Thanks a lot. Amit Daryanani, Evercore, thanks a lot for the presentation. Really helpful. I guess, Dave, you talked upfront sort of the expectation that you think volatility will dampen in the industry going forward.
I'd love for you to just talk a little bit more about why do you think that's going to transpire, especially because as we look at the aerial density improvements, the cost saving improvements you get on HAMR, why wouldn't your cloud customers come to you and say, "We've got to share this a little bit," versus you having it all flow to your bottom line? Just talk about that narrative a little bit. Maybe my second question, if this all plays out the way you folks are talking about, then I think Seagate will see exabyte market share, gross margin expansion faster than your peer. What do you do if Western Digital starts adding capacity, assuming they're behind on HAMR? Do you have to react to that, or do you remain disciplined?
I think first we will run our play.
I mean, I believe we have a great portfolio. It's really hard to get through these technology transitions. And there's a compelling reason to go through the technology transitions for the customers. I wouldn't say we're keeping it all. They're seeing an incentive as well. I mean, to go from a 30 TB to a 40 TB to a 50 TB, there's a big incentive. And being predictable, when you build your new data center, you get the benefit of all that TCO. Plus, there will be some dollar per terabyte erosion there. We have to fund ourselves. We'll do everything in balance. I think we'll do that ramp according to the balance that Gianluca showed in the financial model. Did you want to ask for the second part?
I would say in general, we are not really focusing on market share, as you know.
We ramp our product. Our product right now is showing a good opportunity for us to transition from a PMR 24 TB capacity up to a 30 TB, out to a 40 TB. That will help us in getting a certain level of exabyte. Now, we do not know what competition will sell. We do not really focus on that part.
Yeah, I have been saying lately, and this is the way we grew up in our industry, we learned the lesson many times. Market share is an outcome of running your play very well. If you do it well and somebody else trips and falls, you might pick up market share. Other than that, I think especially the lessons that we have learned lately, we want to make sure we get as predictable as we can in what we are building. There are such long lead times in that.
It doesn't make any sense right now, I think, to build too much capacity for the industry.
Thank you.
Yeah, thank you. Aaron Rakers with Wells Fargo. I guess one on the technology side, you talked, I thought it was a very interesting presentation around the laser technology. I'm curious of when do we expect to see the internal laser kind of represent the full volume or the crossover. And is there a cost differential that we should be thinking about in that? Then I'll ask the follow-up question. I guess we're all going to be left to kind of think about some of the math, Gianluca, on gross margin given to CAGR and everything else. To me, it sounds like what you're trying to message is that, look, we're pretty solidly on pace to do the 40% +.
Hey, if we progress this out, 45 is not out of the realm of reason. One, is that true? Second, how do we think about the non-HDD business gross margin embedded in that view? Thank you.
John, you want to answer the laser part first?
Yeah, I mean, our products are configured for dual laser optionality. We will begin phasing in the vertically integrated laser configuration into and through calendar 2026. Product qualifications are underway. As they complete, we would start seeing that volume in 2026 and beyond. The mix is going to be a function of a bunch of other factors, not just internal consideration, just overall market size and product mix and product adoption. I do not have a specific comment to make on these ratios.
I think it is important to have dual sources anyway.
Now we have our internal solution, and we have a very good vendor that is very valuable to us. We will continue to buy good volume also from them. On the gross margin, we said we have line of sight to the 40% at a certain level of revenue that I indicated. Above that, we can grow at 50% incremental. This is mainly coming from more volume in HAMR that is giving that kind of cost reduction for us. You need to define a certain level of revenue. We will have an associated level of gross margin.
Yeah, I think obviously staying on our plan, we see very good demand for higher capacities, 40 TB and 50 TB. Those drives allow us to be a lot more efficient across our entire portfolio.
There's that one-time speed bump where you can replace a lot of the legacy products, if you will, the past platforms that you've had with these lower-cost platforms. From my perspective, going beyond the 2.6 that we mentioned, that's all about demand. If the demand is strong and we're disciplined on supply, we should be able to continue to grow at that 50% way beyond where we expect to be soon with 40% gross margins.
Yeah, thank you for taking the question, CJ Muse with Cantor Fitzgerald. I guess first question to clarify on pricing. You talked about like-for-like kind of ratable increases. I was under the impression it was more value-based pricing where you share in the TCO. Was that thinking kind of wrong? The second question on gross margins, you talked about incremental 50% when you hit the 40%.
Is there kind of a point in time where we have to see the HAMR volumes where you can achieve that 50%? And if so, what quarter would that be? Thanks so much.
Like-for-like, I think so when we talk about that, let's just pick on 20 TB or 24 TB. At the time when the demand was so volatile and we had precious supply that we had to be very, very careful, make sure we're going to get paid for, that's what we're talking about. Like-for-like, we were raising price at the time. I think to your second point, when we get into product transitions, we can go from 24 to 32 to 40 TB, then we can renegotiate. Usually there's some in that for the customer, and there's some in that for us as part of that transition.
The customer is seeing tremendous TCO benefit anyway, so they want to go through the transition that way, so.
Yes, on the gross margin, now we have indicated a certain level of revenue and how to think about incremental margin when we go above that level of revenue. And we said the majority of that benefit is coming from our new products that are based on HAMR. So we expect a good cost reduction from HAMR. We expect to price that in a competitive way and to continue to improve our profitability.
Hey, guys. Yeah, thanks. Ananda Baruah with Loop Capital. Thanks for all the great info. I think one for each Dave and Gianluca.
Dave, this might be a bit of a clarification, but on the customer TCO conversation with regards to HAMR and the refreshing of lower-capacity drives, is that something that you think customers could embrace refreshing more quickly than they traditionally have as a TCO such that they could look to fast forward that? Does that impact demand in any discernible context over the next few years? I'll leave it there. I have a quick follow-up for Gianluca.
Yeah, I think, Ananda, as I look at edge data centers where there's those more sensitive price points at the lower capacities, the answer to your question is yes. There will be a big move to get off of the old platform where we had too many components inside and reduce the cost of that platform. There will be an incentive.
We'll have to incentivize the customer to do that as well. They'll see that, and they'll be able to pass that value on to their end customer. Yeah.
I got it. It's at the lower cap.
That's right.
Got it. Great. The follow-up for Gianluca is, Gianluca, you mentioned in your prepared remarks that you're good on supply chain. Actually, I think the way you said it is you're good on heads and disks from a supply perspective. Maybe it's a bit of a clarification. If that's accurate, is that because just at HAMR you require fewer heads and disks and that normalized yields free up some of the head and disk capacity? The supply chain tightness from the component chain that we're seeing today looks different as you get to HAMR volume and normalized yields.
No, what I said is, at least right now, for what we can see in the next couple of years, we do not see a need to increase the number of heads and disks that we produce because from those heads and disks, we will generate an increased exabyte because of the change in technology. Going from PMR to HAMR and then from the 3 TB per disk to 4 TB per disk and 5 TB per disk, we think it will be enough to keep up with the increase in demand. No, we think it is close enough. If it is not, we will look at the right time.
Totally got it. Thanks, guys.
Thank you.
Thanks, Ananda.
Yeah.
I think it is me. Matt Bryson with Wedbush. First question is just, again, on the kind of incremental improvement above $2.6 billion to 50% gross margins.
I can understand that in the near term where it seems like that additional revenue will tend to be HAMR-oriented where you're getting higher gross margins. I guess my question is, over time, since you don't have a really high cost of capital, if you will, shouldn't we eventually expect that maybe you can move the entirety of your revenue closer to that, whatever that margin level is, as you move more and more to HAMR?
We will move more and more to HAMR, yeah, for sure. Implicit in this is we move from 4 TB per platter to 5 TB per platter, and there's more incentive for customers to go there as well. It is a journey that doesn't stop at one point in time, to your point.
I guess just the other question is, Gianluca, you showed that graph where costs keep on coming down. Are those cost declines, are they mostly just incremental in the sense you're moving from 30+ TB to 40+ TB all on 10 disk? Or I'm assuming that also incrementally there's some technology improvement in there as well that means that in a sense, you've got a moving target for anyone trying to catch up to you on the HAMR side of things.
This is based on the visibility we have today on those new platforms. I don't exclude that with time, John will find solutions to do even better and to continue to improve our efficiency through the different platforms. They are based on the 30, 40, 50, and 60 TB per unit. Yes.
Yeah, we don't have to add additional componentry.
We have a pretty good line of sight during these transitions of what the cost is going to be of all the internal components. That is why we can say it with confidence.
Thanks. Hi. You said that you run by your own playbook from a competitive point of view. I just wanted, first of all, a clarification that the mid-teens growth rate does not assume any market share gain. In terms of trying to just assess that a little more, maybe a question to John rather than placing it financially, how far behind do you think your competitors are in HAMR?
Yeah, I'd say we definitely think that as exabytes grow, if exabyte share growth is all at the top end, we'll be picking up exabyte share. That is not what we're going after. We're not going after share.
We're going after predictability of the transitions that we have in the market. There may be more exabytes spread across because probably every customer isn't going to go to the highest capacity point as well. It is kind of hard to forecast exabyte share. That's not really our objective. Our objective is to be predictable on the transitions. John, do you want to talk about it? We don't really know everything, but we hear rumors. John can talk about it.
Yeah, I mean, I don't want to speculate. What I can say is that the process that we went through was a multi-year process, and we initiated that some time ago. I would expect any entity that would want to develop the technology and ramp it in volume at the level of reliability that our customers require would also have to go through that same multi-year process.
Now, the part that would require speculation is when did it start for our competition? I do not think I can speculate on that.
I would say one other point is we have been helping the external heads and media vendors along on the process because we believe at some point we may need their technology to be able to further our industry. Because of that, we get visibility into how easy it is to follow what we are doing. It is not easy to copy what we have done. When you talk about atoms on top of atoms, and then you have to have the right capital equipment, it is just not easy to copy it. There is a lot of learning about additional integration pieces because you cannot just build this, build this, and slap it together. There is a lot of integration technology that goes in.
We believe that that's been a tough journey for us. We believe it's a tough journey for anybody. It's going to require a lot of experimentation, which requires a lot of money and a lot of time to learn too.
[audio distortion]
Sorry?
We don't really look at market share. We don't care. No, we serve the demand that is there. That's all.
Yeah, I can't really tell you how the cloud service providers are thinking about this relative to how much they're going to give to somebody else too.
Hi. Hi. This is Yang Pu for Karl Ackerman at BNP Paribas. Thank you very much for the question. I have two questions, and the first one will be a two-part question about your HAMR ramp.
You said your 4 TB disk HAMR is in sampling right now and will ramp in volume from first half, calendar 2026. You also indicated that you will ramp 36 TB HAMR in calendar 2026 and 40 TB HAMR in calendar 2027. The first part of the question is, does this mean that the 36 TB HAMR drive in calendar 2026 would be primarily 9 disks of Mozaic 4 in one drive? If so, why would you not just go straight to 10 disks so you could be ahead of your peers' alternative solution?
Right. It is interesting because we can do variants, 10 disks, 9 disks, 8 disks, just to satisfy what people want. Here is what I would say. The qualification cycle takes quite a long time. Sometimes we will qualify the 36 TB, and then somebody will buy that in volume for a year.
That same person will be qualifying the 40 TB, but take quite a bit of time and feather that in over time. One of the reasons we want to get to the 40 TB is that if somebody wants, and there are people that do this, 32 TB or 24 TB off the same platform to keep the qualifications very simple, then we can satisfy it all with the same platform.
Yeah. Thank you. The second part of the question would just be, you said the HAMR mix will increase to 50% of the mix in the second half of calendar 2026. Do you expect it to be primarily Mozaic 4-based HAMR?
It'll probably be a blend of Mozaic 3 and Mozaic 4, again, based on who's qualified at the time.
Okay.
Yeah.
Thanks. The second question, sorry for the long questions.
The second question on your capacity, I understand that you and your peers are not going to spend much on HDD unit capacity, but you are going to increase exabyte shipment through technology transition. Does this mean the industry units will be kept at like 100-110 million level? Is technology transition by itself enough to support 20-25% CAGR of EB growth? If not, how quickly can Seagate get ahead in the media capacity to fulfill the demand increase?
Right. We focus on CapEx investment for the technology transition. That is the way we satisfy the exabytes. If there was more demand, I think is your question. If there was more demand even beyond that, how fast could we add heads and media? It would take a long time. That would be fairly expensive from a capital perspective too.
We're kind of loathe to do that. We want to keep adding more exabytes by driving through the technology transitions and incentivizing people to get through those technology transitions as well. As far as box count, like somebody else asked before, is it 110 million or 120 million? I don't really get hung up on that too much. It's more dealing with our customers and being as predictable as we can on what their exabyte demand is. Yeah.
Thanks. Steve Fox with Fox Advisors. Dave, you've talked about a lot of interesting technology that you've developed over the last 10 years at the nanometer level, your own lasers, your own proprietary assembly, and photonics. It seems like there's a lot of other applications for this besides just hard disk drives.
Do you guys have the bandwidth, one, to address those, or is that on another roadmap to be talked about in the future?
That's something interesting we think about. I think over the many years in the hard drive business, this is not just us. We've all developed fantastic technology and applied it to hard drives. We probably missed some windows where we could have applied it somewhere else. I mean, on these photonics integrated circuits, the lasers, there are things that we're doing in those technology moves that are far better than anyone else in any industry. There are other things that it's fairly standard for what other industries need. We've looked at linear combinations of what that means and who do we partner with. It's something interesting to think about going forward. We do have some people thinking about that.
I personally believe that silicon photonics is going to be a big, big business someday. We, out of necessity, developed the technology to be able to do that at scale, nanometer scale, and huge volume as well. I think that's probably worth something, but we haven't figured out exactly what to do with it yet.
That's helpful. Just as a quick follow-up, it seems based on all the trends you talked about at the edge data center and IoT that at some point you'd see more of a percentage increase in the edge demand relative to just traditional data center and cloud data center. Is there a point where that smaller piece starts to take up a bigger portion of your business again?
Yeah, we're very excited about the edge.
We see growth opportunities in both.
I think, firstly, I think the edge is going to be a combination of the hyperscale edge as well as an on-prem edge. All of our customers are really, most of the end customers are adopting a hybrid model. We do see opportunities for growth in both.
I do not think you'll see the return to desktop-like volumes that I lived through way back in the day. I do think since a lot of data is created at the edge and the inference should be made at the edge, we're as closest to that. Sometimes because of data gravity, you can't get it to the big cloud anyway. I think there's opportunity for mass capacity at the edge. We're all waiting for those killer apps that are coming out of AI, just like, and I think those applications will make everyone more efficient.
There'll be a big demand for it at the time. Stay tuned.
Vijay from Mizuho. Just a quick question, Gianluca. I think you mentioned 40% gross margin pretty soon. What is the mix of HAMR by then? Are you expecting?
Sorry, what's making HAMR?
What will be the mix of HAMR revenues when you hit that 40% gross margin?
It was on one of the slides, I think.
Yeah. I think, no, HAMR is part of that improvement in the margin. Of course, every quarter we will have more volume. As Dave said, we have now three big customers qualified. We need to ramp the volume for those three customers. It will be a good contributor to the 40%.
Got it. On the 3 TB per square inch to 4 to 5, what's the approximate cost reduction?
I was also wondering, John, if there's an improvement in IOPS going from 3 to 4 to 5. Thanks.
I mean, qualitatively, cost per terabyte scales inversely with the capacity per disk. That is a rough qualitative indicator. In terms of IOPS, as I tried to say earlier, the IOPS is more a function of the mechanics of the drive and less a function of the capacity per disk. IOPS, in an absolute sense, is fairly invariant as we go platform to platform. There is a slight increase in throughput. Megabyte per second does slightly improve as the linear density improves as we go up in capacity. That is an important consideration because as our customers adopt higher and higher capacity drives, their workloads are actually becoming more and more sequential and less and less random. They will benefit from that throughput improvement.
Yeah, most of the scale architects have really gotten rid of the random small block stuff, and it's really streaming a lot of data. The way they think, somebody said earlier, do they not like dual actuator? They do, but remember they have many, many, many actuators in their fleet, right? When they talk about streaming data, it's a different equation than just looking at what one box can do. I think oftentimes people get trapped in this thinking about what's the performance of one box, and these guys are thinking at the fleet level.
Thanks.
Hi, Joseph Leeman filling in for Wamsi Mohan at Bank of America. I have a two-part question. Your low teens CAGR will get you to about 10.7 in 2028, but the incremental is only kicking at 2.6. We won't see the high incremental until the end of the forecast period.
And then second question on overall CapEx trends. If hyperscaler spending slows in F2027 or even declines, can you still meet your targets?
Yeah, I'll take the second one. I'm not sure I understood.
Yeah, I don't want to say the first one. What is the first question?
The incremental margin, you said, is at $2.6 billion run rate.
So yeah, I said starting from $2.6 billion of revenue, the incremental margin with additional revenue will be 50%.
You should expect that at the end of the forecast period?
We need to model.
Yeah, and on the hyperscaler CapEx, it's kind of interesting because hard drives and storage is a relatively small portion of that. It could still vary quite a bit if they were, say, making huge investments in compute one day and then pulling back on that scale.
Their CapEx might come down, but the hard drive and the storage piece of that, the data continues to grow every day, right? The storage part of that might be decoupled to that.
Thank you. Excuse me. Mehdi Hosseini Susquehanna International two follow-ups. You mentioned that you're now build-to-order and you have pretty good visibility. Can you either talk about your backlog and how do you see that trending over the next couple of quarters? If customers are giving you a longer-term forecast, what's the contingency? Is there like a down payment? What gives you confidence that they actually honor their commitment? One last one, how should we think about manufacturing cycle time for you as you go from 3 to 4 to 5 to 6 TB per disk?
To answer the question on the customer engagement, the conversations with customers have increasingly transformed into a longer and longer range. Many of us, Dave had said earlier, I think most of our customers are on a 12-month rolling commitment. I think to date, we've not seen anybody backing out. In fact, it's probably more on the contrary. The terms and conditions for the contracts vary by customer. I'm not privileged to go into the details of that, but we are confident that the demand that's shown to us are fairly robust.
Many of your question on manufacturing efficiencies as we go from platform, over the last few years going through the first HAMR Mozaic transition, we gained a lot of knowledge. Now we believe that we can get from one generation to the next generation without too much disruption. It's stress on every subsystem.
We talk a lot about aerial density in the heads and media subsystems. There is servo mechanical, there is electronics and firmware, there are all these other subsystems and tests. There is a lot of stress, right? We have to be able to deal with that stress, but we have been planning for that as well. We are pretty confident in our transitions.
I think this was the last question.
Yeah, thanks very much. I appreciate y'all coming live and everyone participating. We have a fantastic opportunity in front of us at Seagate. Thanks for coming and listening. Happy to talk to you after the meeting or at the next earnings call. Travel safe in this crazy weather out here as well. Thank you.
Thank you.