Seagate at the Center of AI Storage and Data

[MUSIC] Hello and welcome to the Tech Distruppers Podcast host by Bloomberg Intelligence. In this podcast series we speak with sea level company executives and management teams about their views and disruption and how it's driving their decision making strategy. Bloomberg Intelligence is Bloomberg's research arm. We cover roughly 2000 companies globally across multiple asset classes, backed by Bloomberg and third party data supported by nearly 500 research professionals. My name is Wujen Ho, analyst at Bloomberg Intelligence. And this is actually quite a surprise for me and I'm very pleased to have the CEO, C.E.O. C.G. David mostly to the podcast, did it? Thanks Wujen. Now today we're talking storage and those who know storage know C.E.G.E.T. and look, the conversation is going to center around why it's become a much more part of the data infrastructure story than people realize. When we talk about AI and cloud, most people talk about compute, GPUs, networking, but quite frankly, none of that really works without the data and you really need storage. Now C.E.G.T. sits in the middle of the stack and the business has seen pretty sharp upswing. If we go back to 2023, revenue was about 6.5 billion, that's on a calendar year. Now last year, calendar 2025, they recognized $10 billion in revenue and consensus points to roughly over $20 billion over the next couple of years. Now that's a tremendous upswing when we talk about exponential growth. So with that, for the listeners, we have not followed the hard disk drive story or the storage story closely. What's the 30 second pitch on why C.E.G.T. matters today? Well C.E.G. has been around 47 years. We've mattered for a long time to your point. Before the pandemic, we were at $11 billion of revenues. We went down really hard with the pandemic and some of that was game from home, worked from home. So C.E.G.T. is making hard drives. Those hard drives are no longer right next to you or in your PC. Now they're in the cloud and data centers need data, obviously, with the proponderance of the new applications that we've seen. Call it AI, if you will. Lots of different kinds of applications. They're pulling from the data sources and in some cases snapshotting and putting the data back, growing the data all the time, especially with video data that you see in the world that we've become a video first experience on the internet or on our phone or whatever. So I think this is where the data layers are very important in C.E.G.T. It just so happens that we've been investing in our technologies. So we're hitting a stride in technology development at the same time that all this is happening. So I think that's why you see the growth that you do. Got it. So a lot of people in this audience, they don't understand where a hard drives sit, because most of us own client devices such as PCs. Could you just talk about, you mentioned a hard drives sitting in the cloud. Why is hard disk drive, why is hard disk storage pertinent in the cloud, whereas where it hasn't been for quite some time now on client device like a PC? Yeah, it's actually an interesting journey. If you go back 10 or 15 years, there were data centers then, but they were usually on-prem, managed by corporations. And there were hard drives in there. There were also hard drives in your PC and your Xbox and your laptops and even iPods once upon a time. So hard drives have been on a lot of these devices, but a lot of the data was very small data. It was not text-based data. Now what we're seeing is not only text-based data, but video data and it's just growing leaps and bounds with some of the new applications. A reporter said to me a couple of years ago, they said, "I finally get it." You know, there's more data that I'm creating and it's being created about me, but it's no longer right next to me underneath my fingertips and my laptop. Now it's all up in these data centers. And the drive types are very different. So what we've gone through is a supply chain, remarkable change over the last 10 years from the PC client server to what's going on in the cloud right now. But we've reworked it all now and the data just keeps coming. Okay. So just sticking in with this flash drive versus hard drive, I believe the metrics I've heard was 70% of all storage stored in the cloud is hard drive based 780% Right. Give or take 5% right. Is there a pricing dynamic that gives you an advantage? Is there a speed dynamic? What gives you an advantage here? Why this is a why hard disk storage is preferred? Yeah, I don't think about it like that at all. And and I don't think 70 or 80 is a good number either. I mean, there's some people who were great storage architects that use a lot more. The way I think about it more is we're talking about the data layers now not necessarily compute layers and so on. In compute, there might be a lot of DRAM and NAN different types of memory, but no hard drives in the data layers. There's a lot of hard drive bits and two things partly economics, but partly also the way that the storage architects have architected the tearing. So hard drives are really good at ripping up a lot of data, keeping large scale records together, but they're not very good at small block random rights and reads. That's what memory is good for. And so you come up with the right tearing structure for for those two different data types. So such as you promote stuff to the tier when you need it and you demoted to the hard drive tier when you don't. Hard drives are still working really hard. So a lot of people say is that cold storage. No, it's not cold at all. They're working 24/7 in the data centers, but they tend to be ripping up big data and feeding it to the tiers above them. And and that architecture is pretty set. It hasn't really changed very much. That's why depending on how you count in the data tier, it's much higher. It's probably 85 to 90%. Got it. Okay. So, but it sounds as if it's a co-existent story and it should be the case for quite some time now. That's right. You asked about economics is there's a there's a difference going on right now because everybody's storage and memory is tight because of all these new applications. So the the architects and the data centers are looking at all the tools that they have right now and saying this is the architecture I'm going to pick going forward. That's why we feel so bullish on our future. Got it. Okay. A.I. has been a principal driver. Could you just give us an overview on what really has ramped up the demand the storage demand. What within AI has really ramped up the storage demand. Yeah, this is a really difficult question because I think we don't do a great job in tech of explain all the different kinds of applications that are growing. We just call it AI, right. I would say coming up out of the pandemic trough that we were in the first thing was video and all the video properties that we all use every day. I consume way too much of myself. Yeah, some of that's learning purposes some of it's just entertainment. Sometimes these videos are going multiple times on devices in front of you at the same time. All that video really the business models that have been built on top of it are fairly profitable ad based in a lot of cases. And then again, undergirded by a lot of diverse storage. But what's what I think is also happening is with the rise of AI at least the way I think about it. There were these first training models LLM's text base to be able to really set the hook on the way you might address with either code or with queries or something like that the way you might address a data lake. And now you're actually using that for inference to actually go back and forth into and out of the data lake and you're drawing big data sets out to the extent that those are video data sets. I think it's tremendously beneficial for us. And what I would say is that in order to keep current in the world, you have to be learning what's going on out of the video data sets. But it just keeps compounding. But not every application needs big data. Some of it still needs very small data. It could be very valuable application to you. So I think there's a whole different host of applications that are embedded inside of AI. Some of them are big data. Some of them are not. But I think all of them pulling from the same lake. I think that's what's growing us. So if we go back to the start of the internet, it was text-based email bits and bytes. And then the multiplier effect happened in volume and then video, as you just mentioned, is there another multiplier effect from from AI as well that we should take into consideration? We're certainly watching very carefully what I'll call physical AI. So the text-based LLM is very useful. But ultimately they'll run their course. But then you get out into the real world where you want some of these great compute engines and the code that goes along with them to be able to look at pictures and start to or video, if you will, and start to pull information out of there that maybe wasn't accessible before. I think the computer engines are awesome. And I think we're gonna be making leaps and bounds of progress on that. That physical AI may have, I would give probably autonomous vehicles the first victory in that world, but I think there's a lot of other victories to come in some of the different applications. And it goes from how do you serve customers better? How do we run our factories better? I think there's a lot of new information that can come from some of these platforms, called robotics, if you will. Some of these platforms that actually will ingest a bunch of data and we can learn a lot from it. - Okay, I was gonna say, if physically you mean robots. So. - I do, you have to be a little bit careful because we've been, for our company, the way I think about us is we're robots building robots. So we've automated our factory, and we've been doing it a long time ago, and then we've made three and a half billion hard drives. And if you look inside, you'll see things moving around, and many, many actuators together. So I'm not really that impressed by robotic actuators. I mean, we've been doing that for a long time. What I am impressed by is the sensor fusion that's coming off of the device, and then the ability to process the data. In much the same way, humans we tend to look in people's eyes, and we tend to look at things moving, but really pay attention to the data and the data plane that's behind, where's the data going, who's got it, is it well-managed, is it private, there's a lot of questions behind there, and I think that's where a lot of the interesting applications will come. - Got it, okay, so there's another layer of volume-driven data that's gonna help the hard drive business, the storage business. Now, one of the things that you guys have been talking about for quite some time now, and I was rummaging through my storage closet in preparation. For this, and I found one of my old hard drives, three terabytes. Three terabytes. You're pressing up against three terabytes with some of this new technology that you've introduced, and you're talking closer to 40 and 50 terabyte drives over the next couple of years, and 100 terabytes possible exiting the end of the decade. Talk about the hammer technology, how different it's hammer from the other technologies that you that is mainstream today, and where can you actually take that technology? - Yeah, it's been an interesting journey. When I started working in the industry, we were working on things that were sub-one gigabytes. - Yes. - I'm kinda old. The last, we started in 2005, shipping what I would call perpendicular magnetic recording, PMR. - PMR, right? - Yep. - I was getting really confusing, and I'm not responsible for propagating them, but maybe not creating them. PMR served us really well for a long time, and there's a lot of other inventions that were going on at the same time that it would get the credit for as this S curve, we call it, right? But we're really reaching the kind of waning days of PMR. And fundamentally on our disks, that meant we were using certain cobalt alloys for the media material, and they have limitations as far as the signal strength of the smaller and smaller the bits get. Now, for a long time, we've known that these iron platinum alloys exist, and they're much stronger, and they give better signals, but we didn't know how to write them. And so what Hammer does is it uses iron platinum alloys on the disk, but in order to write them at the size of a bit today, you have to bring in a laser, hit a photonics integrated circuit, get a wickedly strong electric field down through a peg that's about 30 nanometers in diameter into the media, and as you're flying over the top of this bit, you're heating it up, and subsequently cooling it very quickly, two nanoseconds in a field, and setting the bit. Remember the bits are really, really small there, say, 30 nanometers wide, seven minute nanometers down track. So this is a really incredible device physics that's going on, but we've perfected it now. So to your point, we're shipping 40 terabytes today, and we said on the earnings call that we would be shipping customer test units for about the end of 27 for big cloud service providers at 50 terabytes. - Right, and given the essential demand for storage to help power these GoogleWat scale data centers, is 30 terabyte drives, 40 terabyte drives enough for them? I mean, it seems as if the demand for this for the hammer of drives are white hot. - Yeah, I think what's interesting is that we'll make one type of product. There's subtle differences on interfaces and things like that for customers and some features. But generally speaking, the core technology we make the same kind of, and then all these different applications have to reference back to our core technologies through the layers of memory like we talked about before. So all we have to do is focus on making more capacity. There are some discussions now, and I was asked some questions on the earnings call about can we get a little bit more performance out of the drive? Technically, there's one actuator that's addressing all the disks in one form factor. We won't change the form factor because there's a lot of supply chain implications on the downstream of this. But we may actually start putting more actuators in which we've done before, should somebody need more performance. Other than that, though, to your point, all the different applications through the compute and memory hierarchy address the same device and so we're just gonna keep driving aerial density. - And then going forward, I know that you've been talking about the hammer drive for quite some time now before you started qualifying, and even the qualification process seemed to have its own road bumps, but now that you had hammer qualified, and even the 40-terabyte drives qualified, is it a lot easier on the adoption path because hammer as a technology has been qualified by leading cloud providers? - Right, once we know how to do it, and once they have faith in it, so there was a lot of testing as you can imagine on the early side of this, and I think things have gone well, so we're now very happy with this. But pushing the technology envelope is hard. I mean, I've said this many times, through 2016, '17, '18, we were gaining, we were going from like 12 to 14 to 16 to 18 terabytes. Now we're making much bigger jumps, and that puts stress on every other subsystem. So hammer gets all the attention, but the read subsystem, the servomchanical subsystem, the electronics and the interface, all the data flows that's going on, all the controls, that's all being stressed equally. So, but we have confidence, we can solve this, we have a lot of smart people in our industry working on these problems. - Yep, there are three drive manufacturers, right? You, Western Digital, and the Tishiba. You're the only one with hammer, but Western Digital is talking about hammer themselves. I mean, how much of a lead time do you have over your competition, and do you welcome the competition? - I think our competitors are smart people, so they're gonna figure something out. There may be different types of solutions coming, depending on what you actually have traction for. Maybe one subsystem you can race ahead of where we are, maybe we're ahead on a different part, so there may be different types of solutions coming. And in the current demand environment, that doesn't worry me too much. I think the customers can deal with some variants, and I think they probably will. That's the way I think about it. If you're still in our industry, I'm pretty impressed by the technology that we've developed. I'm pretty impressed by the technology that industry's developed too. So, I think everybody's got a lot of work in front of them, and pretty smart people, they'll be up to some kind of challenge. - All right, and then, you know, I kind of leaned into it, and I'm sure the audience figured out that it's a tight supply environment with this high demand. You've said on the earnings calls, the high capacity supplies effectively sold out into 2027, right? Historically, how unusual is this? I mean, it's got to be a different feeling. - Very, very unusual. Again, you know, we've made three and a half billion hard drives in our lives, so, but most of that time we grew up, you know, we were coupled to people like PC builders who, as soon as we shipped it, they put it in their product and they shipped it, and they revenued it. The cloud service providers are very different, right? They have huge install bases. The fleet, they think at a fleet level, actually, which is very different than thinking at an individual device level. So, they know where the technology's going. They know what they're gonna need for their data center where there's a refresh of the old data center or build out of the new data center. They know exactly what they wanna be doing for power efficiency, space efficiency, so on. They know what their applications are driving, but because we had these issues a few years ago with the supply demand undulations after the pandemic, you know, we've kind of said, and sorry, our lead times are very long now as well as an industry. We've kind of said, "Hey, let's be more predictable. Let's tell me what you need." You know, two, three, four quarters out. They're telling us even further out than that right now, and you know, that's getting old too far ahead of ourselves, But. But it's driving us to a much more predictable model. We call it the build order model. - Yep, and I found that to be very successful 'cause it give you visibility into what customers want as well as the lead times and your production capacity. But the question I always had asked by investors, why is it this time different? Because the memory industry as a whole, whether it's a flash drive in or even the hard drive driven memory industry, they've always said, look, we're going to do build order, we're going to do LTMs. But why is this time different and why is this going to be durable? - I'll tell you how I think about it. And some people might not agree with me on this, but it's fine. We picked client server in 2011, 2012. And that built out enormous supply chain infrastructure that changed dramatically when we went to cloud. We did 66 million drives in one quarter, C-gate hit. We're not anywhere close to that now, but they're also totally different kind of drive. Those were very thin notebook drive, one disk to add. We had to get them in the notebooks so that everybody could get notebooks to get online for the first time, probably it was going on. And the devices now that are going to the cloud are big beasts, they have 10 disks. From our product line. And so when you think about that, all the while the data's been growing, but we went through these immense shifts in supply and demand. And so a lot of the cloud growth happened at a time where we were actually still coming down in total capacity from a box perspective in the industry. The heads of media capacity, which are the critical components that are controlled by us, a little bit different. And so the heads of media capacity was, it was full by 2018, 2019 again. But then obviously damaged after the pandemic again. But I think all the while, data just keeps growing, keeps growing, keeps growing. So if some people want to see these three or four year cycles, I see a much bigger cycle here is what's happening from a decentralized model to a centralized model now in the data center. You know, I think there's one other interesting thing that's going on is the application spaces change so much. In the past, there was a lot of software on top of the stack. The data could be relatively small and the software was kind of constraining. How the data was used. Now in the cloud models, it's not. It's actually pulling data up much faster. And so I think all of these things contribute to a remarkable set of trends that are going forward here. And with better visibility, we'll know exactly how much capacity we're installed so we don't overinstall. - Right, so going back to your customers in terms of AI spend $650 billion, our high-speed scale cap ex, we're in unprecedented times, right? I still see that trajectory going up to 2028, possibly to 2030. But we talk about structural versus cyclical one, given the low production capacity or the availability it drives today, how are you sure that you're not seeing much double ordering? And how do you protect yourself that just in case there is a CapEx downturn, do you guys do not see another 2023? - Yeah, well, I mean 2023 and the pandemic was different for a lot of reasons. But I also think the application spaces changed dramatically, cold AI since then. But I would still say so against the total CapEx of the major cloud service providers, the storage layers are actually fairly small. They're not even 10% of the entire CapEx, right? - Data matters that really does. - Yeah. - Now, we've seen this in the history of the IT industry is that what drives the initial demand is the application layer. And then the compute comes right after that because usually when you want an application, you want it to run as fast as productive as you can. So I think racing forward on GPUs and TPUs, that's been awesome to watch. And a lot of different kinds of applications coming. But all while undergirded by the data layers, that I think that's what's adding a little bit of resiliency right now. So it may, you may see things change periodically forward to the compute, depending on whether or not it's profitable, but I think the data will just keep growing. That's what's always happened in IT. You buy the application, you buy the compute, and the data just keeps on growing. - Yep, the data produced by the existing compute keeps on growing. This is one of the more impressive things that I think you've done, not only as a company, but as an industry as a whole. You've been very disciplined on how you're managing CapEx and production. You know, it's almost counter-attuitive in the sense that, well, if people want more, they want to make more. But would you, you as well as your peers, have put your foot down, it's like, we're not gonna make more drives, but we're gonna put more bits on drive, hammer technology included. I mean, why take that stance? - There's a couple of just realities. Some people would assume, hey, just buy more machines and plug them in. - Yeah, well, the machines have long lead times and it's not one machine. It's an army, a fleet of machines. Not even just one manufacturing process steps, just making our recording heads as thousands of steps, and takes a long time to make them with a lot of different machines. And then you have all of our suppliers. So if you stare at a hard drive, you say, oh, okay, now I get it. It's a very complex supply chain coming together. So to your point, what we've done with our talented people inside of our company, and I think the industry's actually answering the same way, is said rather than calling time out and going and getting new factories, which long lead time new machines, getting them up, running properly, let's jump forward in the technology because that's the best way to bring the most bits to the world. And I think that's gonna be true for the next few years. So we have enough confidence in our density roadmap to actually add bits into the individual boxes that we can grow much faster, exabytes that way, than trying to build new factories. - Perfect. So I'm gonna see if gears, I started off with the financial story in terms of revenue growth, expectations by consensus, and how far you've come from. But the margin story is probably one of the more phenomenal stories I've heard in quite some time now. We're talking about mid 30% growth margins, and not so long ago. And now I think you touched 50% this past quarter and over forecasting. - We didn't, but who? - You're getting there, right? And consensus has our asking about, can you how much further can you go beyond 50% growth margins? - I think that all comes down to what's the true demand, especially for the highest capacity products that we make. - We run wafer fabs like other people were on wafer fabs, and our margins would have been historically pretty low, but these are becoming very, very critical parts out in the world that we're all constrained by. So I think the margins will go up. We're still running the business for predictability right now. So we talked to our customers about a year, not five years, a year of visibility. What is gonna be qualified, what's running in our factories, and we price it. We talked about this on our call, but we're always able to out just a little bit more, and we can sense the market. And what the market keeps telling us is demand is a lot higher. So as we continue to race for, and we can predict what we're coming out of our factories, we could see some more margin uplift if the demand's high enough. - Dr. Asai from demand, once we get to a steady state business model, what is a steady state growth margin profile that we should be thinking about, given the revenue projections? - Yeah, we haven't even gone there yet. I think we're pretty, still pretty far away from what a steady state demand look like. The caveat I would always have is that, we're running the business great right now. - Yep. - But it is a crazy world. And some of what happened back in the pandemic days was because of how we had to impact our supply chain during those times when people started pulling hand breaks. So we're always watching, we're always watching what's going on in the world. I think we're a lot healthier in our supply chain now. We spent large part of last year working on working capital. - Yep. - I talked about that on our call. Now we're working on our debt. So we're still recovering from the past. And then I think in the future, we can start investing again in our technology and returning value shareholders like we've always done. But we're always watching to make sure that the world doesn't go through another hiccup. - Yeah, I mean, I think if you, given what you've done in 23, you made the financial moves hard once quite frankly, to get you in a very good operating shape. And if the revenues and the margin projections go where it's supposed to go, you should be strongly free cash flow positive and it should benefit shareholders, whether it's gonna be buybacks [BLANK_AUDIO] also reducing that that's the way you're thinking about capital allocation set the right way of thinking about that's right That's right. Yeah, and you know the last thing I would say on demand I think this is very important is there are Zedabytes of data being created by man only a few Zedabytes one or two are actually translating into what's being stored today and used you know effectively and I think some of these new applications are and the computer infrastructure around it making that way more efficient than the past so we'll see we'll see if you know This is truly a demand super cycle or not, you know, I tend to think that it is again with with the caveat that it's a crazy world All right, a couple of questions before we wrap I'm thinking about 2030 right five years from now or four years from now What does a well-executed C gate look like I'm assuming you're still gonna be at the helm? We we are pushing aerial density really hard on what I would say is that you know 10 years ago 15 years ago It was pretty hard to see the future always you know Hammer served its purpose which was hey, that's let's get to that new material set and we can we know we get more density out of that but Yeah, some of this is you know Sussing out what's coming out of academia or or you know other competitive industries that you know are doing their own developments Now we look around and there's a lot of really interesting technology development happening material science all the way to compute and And That's what we do in our industries. We look at those things We say if I use that and that and that I can turn this into density so right now I'm pretty Optimistic about the hard drive business being around for a long long time because of some of those opportunities I see we've got to figure out exactly how we're gonna invest in them So we've talked a lot about photonics moving on some of the investments in photonics are different For the hard drive applications than they are for what everybody else is talking about and in networking But you know to the extent that the the fundamental materials may be the same we can use that right? So we'll turn that into density and and I'm actually pretty excited about the next decade for that. Oh perfect and I know you interface with investors all the time what's The one last thing and the storage industry that you think that is still widely misunderstood Maybe a different answer in the last year because a lot more people are paying attention now But I I think always people didn't appreciate how sophisticated the technology was and as they start to look in they understand Why the supply chain is so complex and and why the parts are so hard to make and you know Not just investors but customers and and some of our partners are you know have a tough time with that but I think people are awakening to this right now and And it helps us run predictable business inside I think that's what we're focused on perfect That was great. Dave. Thanks for joining and joining us and sharing your insights on on C gate Well, Jim my pleasure, thanks and thank you everyone for joining us We have a great lineup of future disruptors similar to David Mosley at C gate So hit the subscribe button and keep up to date with the tech to shoppers podcast and not miss an episode I'm Woojin Ho. Thanks for listening to tech to shoppers. We'll see you next time