AI Storage Need Fuels Hard Drive Demand

Key Takeaways

There are loads of people that are going to watch this program and be like, I've got a WD uh hard disk drive HDD somewhere in my home. >> I understand you have one >> and I have one, right? And and it it it protects uh some of my most valuable memories in terms of photos and videos. But the story is data center and probably it's under appreciated. So start with that where HDD fits in the data center story and then use that to contextualize what is crazy demand right now. >> Sure. I mean if you look at the AI value chain a lot of focus has been around the compute resources right GPUs uh memory uh that's that's a source of compute that's generating data but ultimately data needs to get stored that data that's generated that needs to be stored right and that's where hard drives really come to the four 80% of data that is stored in the cloud is stored on hard drives and as if you look at where things are going with AI what is happening is actually there's a structural change in how hyperscalers the market really looks at the value of AI and as a result of that as the value of data increases by association storage increases as well because you need to store more and more of that data and a simple analogy to use is power generation >> right >> if you think of um GPUs and memory as the power generators >> data is actually the fuel that powers those power plants right and so really that is is fundamental ble to enable AI going forward. So in short, if you don't have HDDs to be able to store that data that's generated that's being created. >> Can we can we extrapolate on that analogy a little bit? So when we think about a data center, we think it of it in terms of gigawatts. A gigawatt equates to um hundreds of thousands of GPUs depending on the on the stack. Is there an equivalent that our audience can understand when it comes to storage? So you know a like for like uh that that's that you can chew on uh maybe it's in pedabyte something like that at that scale >> there isn't really a rule of thumb that you could use say if you have a gigawatt of compute how much does that translate into storage because in a way there's a bifocation of compute from storage if you think about AI every time you run an AI workload whether it's training or it's inference those compute resources can be recycled every time you run a training model development or an inference workload but the data that's getting generated and as a result of that gets stored it doesn't get deleted >> right >> so storage is actually growing because of the value of data increasing over time as well >> how much is your supply growing how much can you ramp up to meet the face demand >> so um we've projected that exabyte growth which is the amount of storage that's needed is going to go at 25% keer over the next 5 years uh we feel very confident that with the technology enhancements that we're bringing to be able to deliver higher capacity drives >> uh to the market we're able to meet that demand going forward What's interesting is it's yourself and Seagate many would sort of see as you >> Yeah. the the place that people come to. How are you ensuring with this fierce demand and perhaps anxiety coming from cloud providers and the like that they're not double ordering that they're not ordering from you at the same time as Seagate to make sure that they get the fastest of the speediest? >> Yeah, that's something we we pay very close attention to, you know, looking at the inventory levels that our customers have. Uh working very closely with them to reassure them that we have a technology roadmap that's going to be able to meet their needs. Um, and I think what they also appreciate that storage has become a critical piece of the AI data center infrastructure stack. And as a result, uh, we've really shifted the partnership model with our customers. They're giving us really good insight into the applications and workloads they're bringing to market over the next few years. A lot of it's going to be multimodal in nature. And what I mean by that, lots of it's going to be video based because video is going to be a big driver of storage. And that's giving us a lot more confidence in terms of the demand uh that they're signaling to us and also ensuring that they're not creating an over supply in the market. >> I mean you talked about your innovation road map. Can we talk a minute about heated assisted magnetic recording for those who aren't in the lingo but basically this is something that Seagate's been pushing forward on and you too have been bringing that sort of innovation. Where are you in terms of neck and neck and you're already bringing this to customers? How is it testing out? What does it do? So we're already qualifying um our heat assisted magnetic recording or hammer professionally hammer >> as we call it with two customers but we've taken a slightly different approach from our peer uh where we have a dual track process right we we definitely are committed to hammer and that's required to get us to 100 terabytes per drive and beyond but in the interim we feel that with our current EPMR technology for the next 3 years we can actually get up to 60 terabytes >> and deliver it with more reliability more quality and more scalability uh to our customers than a hard transition to hammer. >> So, I've got to go back to something earlier. Why don't you just increase capacity? >> We don't need to increase unit capacity. What we're actually doing is to use technology to increase the capacity per drive. So, today we're shipping at the top end of our products a 32 terbte hard drive. In the second half of this year, we'll be shifting shipping a 40 terbte drive. And >> between now and 2026, 2028, we'll be shipping close to a 60 terbte hard drive. So just imagine that the capacity per drive is doubling >> in the next two years. So we don't actually have to increase the number of units we're producing. That's great for our customers because it means better total cost of ownership through higher rack density, lower power consumption, and that's something that also aids the lower power demands of data centers. >> We uh I'm going to have to compress this, but WD separated SanDisk, right? NAND flash. Um that has been an interesting financial transaction and hopefully we can get to it with time but the basic conviction is that HDD is the way to go focus but you know the the flash guys would argue in the context of the data center in the cloud that SSD is going to have a role um what is your latest thinking on on how that went and and the the project long term >> so I think that the spin has been a very successful one if you look at how both companies have performed since the spin close to a year ago >> astonishingly well >> both have done really well. I think what has given both companies is a singular focus on their specific technology road maps, specific segments of the market that they play in as well. Uh and specifically to your point look uh we are symbiotic, right? 80 I mentioned 80% of data that gets stored in a in the cloud are on hard drives. SSDs represent the next 15 20% of it as well. So it's a symbiotic uh relationship. There are some workloads that uh require lower latency that are very much designed for SSDs. when it comes to long-term storage, the the vast majority of metadata to support training and AI inference going forward, that's going to be leveraging HDS because of the superior economics that we deliver. We've also launched a month ago at our innovation day some new innovations coming out over the next 12 to 24 months. It's going to improve the uh data throughput and performance of hard drives that will also be be able to support the needs of AI workloads coming down the pike as well. >> You're a global company. We have a pretty global issue on our hands at the moment. How are you thinking about your supply chain risk? How are you thinking about ensuring that your customers get what they need and when they want when you have manufacturing as farflung as Thailand, as China, here in the United States as well? >> Yeah, we have a very uh global supply chain and quite a diversified one as well. And uh we constantly uh model uh risk scenarios like the conflict that we're unfortunately seeing in the Middle East right now. And uh from our current perspective, we're keeping a very close watch on it. Uh I'm pleased to say as of now uh none of our employees in that region are in fact impacted and there's no material impact to our business as