All set?
Good.
Thank you very much for joining us this afternoon. I am Aaron Rakers. I'm the IT hardware analyst and semiconductor analyst here at Wells Fargo. With us, we've got the Chief Technology Officer of Pure Storage, Rob Lee. Rob and I have known each other for a while now, and we're gonna get into some details. You had some news last night. But before we do that, I'm going to do my best to read this safe harbor statement. Statements made in these discussions, which are not statements of historical fact, are forward-looking statements based upon current expectations. Actual results could differ materially from those projected due to a number of factors, including those referenced in Pure Storage's most recent SEC filings on forms 10-Q, 10-K, and 8-K. I think I did all right.
Pretty good job.
Perfect. Thanks, Rob. So first of all, thanks for joining us. I had a list of questions I mentioned before we got up on stage. Last night's news is probably gonna change, a little bit of that, a little bit of the path we were gonna go down. But you guys announced hyperscale win last night. I wanna start at the high level for everybody that's not that familiar with Pure. Why? What makes Pure unique in their ability to even get this opportunity to participate in this market? We've been talking about it for a while, but I think it's great to start there and maybe unpack.
Yeah, absolutely. So if we go back to the beginning of Pure, we have always had, you know, the vision of replacing all the disk in the data center. And we started our journey in the enterprise. And, you know, if we look back over the last couple of decades, the enterprise really has led the way in terms of deployment of flash at scale in mainstream workloads. Hyperscalers to this date have lagged in that mainstream deployment for a number of reasons. And those reasons generally come down to they have been able to do the engineering. They've been able to get the TCO benefits out of disk that they previously have not been able to get from flash.
Flash has all these great properties, but at the end of the day, hyperscalers have been able to make their disk-based systems perform well enough at just a cost profile that is significantly advantageous. What we have done, and really what led to the design win, is frankly broken those barriers down, right? And let me unpack that a little bit, right? You know, we've been engaged, you know, really over the last year and a half, in discussions with a variety of hyperscalers, touting the benefits of our technology, the same technology that drives our enterprise systems, to take the best benefits of flash, deliver unparalleled reliability at scale, deliver much higher performance levels, to deliver significant power, space, energy savings. You total all that up, and it translates to significant TCO savings.
And what's happened really over the last several years is a couple things. One is breakthroughs in the technology, right? As we've talked about our direct-to-flash approach, really extending that to much higher densities, capacity levels that really start making the TCO equation for hyperscalers in particular look very compelling. So that's number one. Number two is a laser focus from the hyperscaler set on power availability, right? It's no secret. You look at the news. All of these hyperscalers are scrambling to get power. If they look in their data centers today, a large part of where they're spending their power sits on storage. A large part of that sits on nearline disk storage. We can go reduce that significantly. That's become a key critical element. And so when you net all of that out, those are the key drivers.
It's still a big hill to climb, and we've been plotting our way up that hill for the last year, working hand in hand with well, multiple hyperscalers, but our lead customer in particular, you know, moving through early assessment of the technology, really deep understanding of how that would fit into their designs, what that would mean in terms of TCO, reliability, operational considerations, and then now with the design win announced, becoming part of that plan of record in terms of how they're planning to deploy their next generation of storage architecture.
Yeah. So fantastic announcement. I mean, I think that it's important to kinda note, like, first of all, I've never seen an enterprise. There's no other enterprise traditional system vendor that's participated in a hyperscale cloud. So when you think about it, you've got how do you think about the size of this opportunity that it presents? And I've heard you mention it a couple times, multiple, right? It's not you got one lead customer, but this lead customer begets maybe the opportunity to participate more quickly at others. Is that fair?
Yeah, absolutely. And so a couple pieces to that. So number one, yes, this is very unique. Certainly, I would say it's first and only of its kind in storage. I would say the next nearest comparable would be what Arista has done in the networking space, as they've you know moved some of their technology into the hyperscalers. Now, if we look you know if we look at what we're doing in terms of sizing the opportunity, a couple of data points, a couple of things to think about, right? Number one is you know just taking a look at the nearline hard disk estate overall. I think Charlie shared on the call last night.
We size the annual deployment of nearline hard disks in the hyperscaler environments to be somewhere in the order of 6-700 exabytes a year total, all in, right? And you know, so that gives you a sense of just the enormity, right, of disks that are out there that we believe we now have, you know, a play for. You know, if we look at how does that play out over time, you know, that set of disk deployments is very concentrated, right? If you look at, really the most of that sits in the top 10, right? And so, you know, the same conversations, the same technology that we have developed for our lead customer, we believe is highly applicable.
We were talking to multiple other hyperscaler prospects about the same technology, different, you know, integrations to their environments. You know, we see a lot of commonality in terms of what they're looking for, in this space. And so, to the last part of your question, yes, we do believe, you know, knock on wood, we're not gonna make any predictions, you know, but, you know, I do believe that, being able to come out, announce the design win, certainly the partnership with our suppliers, as well, what we're doing with Kioxia, will serve to accelerate, our path with other hyperscalers. You know, to some degree, there, I think, has been some doubt in the industry of, "Hey, is this real? Is now the time? Can this actually happen?" and being able to, you know, existence is worth a thousand words.
And being able to go and prove that out in one hyperscaler, you know, will absolutely help us in the other engagements. Because remember, at the end of the day, you know, storage technology is not an area where these firms are competitive with one another, right?
Yep.
And so, we believe that will be a tailwind.
Yeah. I'm going to put a little bit of context around what you just said. I mean, six- 700 exabytes of hard disk drive capacity ship. The company last night alluded to, like, double-digit exabytes by 2025, calendar 2026, fiscal 2027 for you guys. I mean, if my math's right, I think you guys, you know, even that double digits is significantly more than what you're shipping annualized today, maybe in the three. I think I put in my note, three to five or three and a half to five to four exabytes today. I'm guessing that's the right kinda sizing of this. This opportunity is very large, but at the grand scheme of things, it's just scratching the surface on the size of the.
Yeah, I think that's a good way to think about it, you know, and a couple items to add to that. So, the, you know, double-digit exabytes that we're talking about in fiscal 2027, absolutely is significantly larger, right, than what we're shipping today. It's also, I think, the first year of scaled ramp that we're expecting, right? So, you know, if we step back and we look at, "Hey, where are we in this process?" For the last year, right, we've been engaged in really technology assessment, working our way to that design win, getting to part of that Plan of Record.
Really over the next year, right, the way we expect things to play out is, as this hyperscaler customer, you know, moves closer to production, they've got a set of stages to go through, moving from, you know, advanced prototyping to completing their integration and implementation of other parts of their data center design of that generation, doing scaled, kind of field testing, if you will, more detailed field testing, testing of smaller scale production deployments all the way to scaled ramp. We expect that to largely play out over the next year and for that scaled ramp to start beginning to occur in that fiscal 2027.
Okay. Perfect. And I think the context of the call last night too was, like, it wraps around a design that they've done for their next generation of data centers, right? You're not naming the customer, but actually these customers, these type of customers lay out an architectural design of the data center, and then they kinda, you know, that's the blueprint they follow in multiple data centers. That's how you'd characterize this?
Yeah, that's a very good way to characterize it. Again, all of these firms, they derive incredible economies of, well, hyperscale.
Yep.
By standardizing on a small set of blueprints that they can go and stamp out multiple times. So when they think about storage as an example, they absolutely do not think about storage on a workload-by-workload or project-by-project basis. They might have, you know, slightly different designs for their high performance, their mid-performance, their low performance, but they've got a design for that horizontal tier. That design is going to generally last for some period of time. During that period of time, storage deployments that are put out, whether it's brand new data centers, whether it's, you know, completely refreshing older data centers that are kind of coming due, are generally gonna be that blueprint, just, you know, just stamped out.
That's perfect. So, maybe I should've went down this path a little bit earlier, but, you know, talk a little bit about I'm going back to the why, right? The why being the architectural attributes of Pure, right? What you do at Purity OS, how that kinda segues itself into what you can do as far as the DFMs and the scaling of the DFMs. Talk a little bit about that. You know, if I compare you guys versus some of the traditional vendors, what makes Pure different?
Yeah, absolutely. So, so if we look at the hyperscalers, one of the questions we've gotten is, "Hey, so great design win. Who are the other competitors in the space?" And we've been emphatic that, "Hey, there were no other outside vendors in consideration." And this goes to the why. The why is, because there's nobody else in the industry that has the direct-to-flash technology, the IP, really the flash know-how and smarts, that we bring to the table. Okay, so then why is that important? Well, that IP, our direct-to-flash approach, the fact that we do not rely on commodity SSDs, it's that IP that allows us to bring out this much higher level of reliability, much greater degree of performance, much greater degree of efficiency through density, right?
To give you some data points, you know, when we talk about reliability, right, our DirectF lash modules, you know, they're generally, you know, our annual return rates are generally below 0.2%, right? Industry kind of averages standards for SSDs are north of 1%. Hard disks closer maybe to 2%. So when you think about that, okay, so that means we're five times more reliable than an SSD. Oh, but wait, we're, you know, our DirectF lash M odules, because they're simpler, we've removed a lot of the complexities that are inherent in SSDs, we can build much denser packaging. We can ship more storage per unit, you know, module than you can get in an SSD.
So to meet a customer's need, if I can deploy five times fewer of these modules and they're five times more reliable on a unit basis, well, my reliability is 25 times higher, right? And so you get these multiplicative effects so that when you step back and you look at it from, you know, system level, at the scale that these hyperscalers are considering, a 25X improvement in reliability is massive, right, when you're running a, you know, consumer-facing cloud service, right?
And I would imagine what you didn't touch on there, but the power consumption envelope. Like, as we met with Dell yesterday, and you know, one of the conversations was, "Hey, you know, a few years ago, a typical rack server might be 15 kilowatts of power consumption. They're gonna ship, you know, they've got a rack now that's 480 kilowatts in a single rack." So the power consumption envelope is a big deal. I mean, to unpack that a little.
Yeah, absolutely. So we generally, you know, require power that's about a fifth to sometimes a tenth the power of competitive solution all-flash solutions, and when you look at disk, it's sometimes even, you know, even a greater difference than that, and a lot of that, again, comes down to the direct-to-flash approach, right? At the end of the day, the core difference between our approach to delivering flash and everybody else is they rely on SSDs. SSDs do a ton of extra work, a ton of extra translation that introduces a ton of additional inefficiency. You can measure that inefficiency in cost, in space, in power, in reliability.
And for us, you know, never having had that inefficiency built into our software, well, every step along our roadmap, as we increase our densities, as we move to newer generation NAND, the benefits we create out of that IP, the benefits in terms of reliability, performance, and efficiency, just get magnified each step of the way.
Again, on the DFMs today, we're shipping 75, 150s. The roadmap from here is.
We're expecting the 300s next year. You know, we see a path to going beyond that. But you know, look, the 300s, even the 150s represent a 3X-4X increase over where we were just 18 months ago. So you know, we're in an incredible growth period in terms of improvement of the technology. And again, bringing this back to the hyperscaler, right? I think that's another reason why we were chosen as a design partner for this hyperscaler's designs, is you know, these are highly technically sophisticated organizations. They wanna work with bleeding edge innovators such as ourselves.
Yep. And then, you know, I'm gonna move on after this with the hyperscaler discussion, but, but, you know, clearly last night, this is a different sales motion, different type of monetization, for this opportunity than traditionally selling to an enterprise. So it's a licensing model, based on capacity under management. I'm 100% certain you're not gonna give me the numbers, but just help us understand maybe the mechanics of, of the monetization piece of this.
Yeah, absolutely. So, and we discussed this on the call last night. So, you know, first things first, the hardware, while the hyperscalers will be deploying hardware as part of their total solution, hardware will not be part of our sale to the hyperscalers. You know, we'd expect the hyperscalers to source the hardware through their own supply chains, etc. What is part of our arrangement or construct with the hyperscalers really is, as you alluded to, Aaron, two pieces. One is, think of it as a technology license. Would expect that to be on a per-capacity-deployed basis. And then the second piece of that would be more of a support and maintenance agreement, and would expect that to be on a time basis. So think of it as not dissimilar to an annual maintenance and support type of agreement.
So the hurdle from here is, I mean, the deal's done, right? It's just getting through the processes at that customer to start volume deployments, correct?
Well, so what I'd say is the deal, the design win that was announced is really being selected and, you know, being part of the plan of record in terms of how they're gonna go build their total, you know, their end-to-end solution, right, and design for that data center generation. That said, right, you know, there's work ahead, right? The work ahead really is for the hyperscalers to go through those phases to move to production, go into, you know, completing the rest of their integrations, implementations, you know, do much more detailed field testing, dotting the i's, crossing the t's, you know, building up at smaller scale testing all the way to the full ramp. We certainly expect to and will work very closely with the hyperscaler along that process.
You know, but in terms of, you know, in terms of big hurdles to cross, you know, we don't see any specific ones other than business as usual.
Yep. That's perfect. So I'm gonna shift gears. You being, you know, the CTO and thinking about, you know, there's the perpetual file flywheel of, like, the what's next. As you think about Pure, you know, company's done a fantastic job of building out really truly a portfolio company. Now you layer on this hyperscale opportunity, but really broadening, you know, the platform strategy, if you will, of the story. You mentioned Arista in an earlier comment. How do you think about the cadence of, like, what's next on a roadmap? Is there areas where you're like, "Hey, this is it feature functionality from here?" How do you think about that from a CTO perspective?
Wow, no, no rest. We just announced the design win last night, and you're.
I'm always thinking about what's next.
No, it's a great question. So, I would call it a couple of things. You know, so I think we've, you know, we've matured our sets of capabilities and really put the last puzzle pieces in our portfolio in terms of discrete products and areas of workloads and market and the market that we wanna go cover. I think the couple areas I'd call out are number one is really explaining our expanding our platform strategy and building that platform effect. And Pure Fusion has a big part to play in that. You know, when I step back from it, you know, where we were five, seven years ago was a point product company, right? We had individual point products. You've got this workload. I've got the best thing for you, right?
Over the last, you know, the next couple of years was about filling in different point products in that map. I think we've got a pretty good coverage of those key workloads from super high performance all the way down to our AI to archive.
Yep.
And we can do it with one consistent software architecture, one consistent hardware technology. I think the next leg forward really is about how do we tie all this together in more of a platform effect? How can we make it so that, you know, when a customer selects Pure, they're not selecting Pure for just one workload. They're selecting Pure for, you know, their entire estate or large portions of their estate. Pure Fusion has a big part to play in that, by virtualizing how customers manage and have to think about their data estates, by moving them from thinking about and configuring and purchasing storage array by array, workload by workload, use case by use case to much more of a horizontal enterprise-wide purchase, right?
If you as a customer don't have to think about, "Hey, I need this much storage at this performance level for this database, which is different than this other workload," but rather, "Hey, I need a little bit more of Pure at the high performance level, and I just drop it in there. It's a pool of resources. It automatically goes and gets deployed." That's kind of the next leg forward for us. A lot of technology that has to go into enabling that. We've been working on that, you know, rigorously, vigorously, I should say, over the last couple of years. And the recent release of Pure Fusion 2.0 is a big step forward. The second area I'd highlight is continuing to build on our subscription services, right? And it's more than just subscription.
I should say, continuing to build on our full as-a-service portfolio. We've made great strides in kind of the first innings of that in terms of Evergreen One, but I think we've still got room to run there.
Yep. And on the Fusion strategy, you said it earlier, like a virtualization layer. So it's like the management layer that allows you to manage, depending on your SLA, your workload requirement, how do you kind of click and select certain capabilities performance-wise or whatever it might be. Version two, I think at your end-user event mid-year, you kind of outlined a roadmap of where that strategy's going. So version two, what's the next things there?
Yeah. So, you really see the big difference, and you know, I think you summarized the capabilities quite well, and we've been working with early customers over the last year with this set of capabilities, but our initial challenge has been that it was net new, you know, applied to net new deployments. What version two brings is a non-disruptive upgrade for existing customers in the installed base to take an upgrade and all of a sudden have all the capabilities of Pure Fusion, so it's really about making that technology much more broadly available to our installed base and across our entire product set, and so that's really the big, you know, the first big step forward.
You know, I think the other thing I'd call out, so there's really, ultimately, at the end of the day, it's about bringing the cloud operating model to customers on their own premises, right? If we think about, you know, some of the lessons we've learned in working closely with our now hyperscaler customers, it's about how they run and operate their, you know, storage services beyond just the hardware, right? Hyperscalers, you know, when you as a customer go to a hyperscaler and you, you know, sign up for an IaaS service, that doesn't create an IT ticket that causes somebody in the hyperscaler to go and take a physical action to deliver that thing, right? That's all automated. It's all done by software. It's all, you know, managed automatically and provisioned. That's not how the enterprise data center works today. That's what we're solving with Pure Fusion.
Yep. And is that, remind me again, Pure Fusion, when you go in version two and you provide that capability as a non-disruptive upgrade, is that a monetization? Is that actually or is it like a seamless upgrade? It's a free optionality too?
Yeah. No, it.
So that's one thing that Pure's done really.
Yeah. It's a great question. So it's a seamless upgrade. You know, we want to drive adoption to this, right? Because, again, we've done a great job. I would argue as a company of making individual arrays, making enterprise technology that has historically and still our competitive set makes horribly complex. We've done a great job at making that very simple. We're now at a phase where, hey, how do we make somebody who has 100 arrays not have to do something simple 100 times, right?
Mm-hmm.
And so a lot of the initial phase of Pure Fusion is really about driving adoption, making it more widely available because, you know, it will make it easier and incrementally better for customers who have large installed bases of Pure technology to add to that, to add to that pool, if you will.
Yeah. So it's a wallet share gain opportunity rather than a monetization. Talk a little bit about in the time we have left, you know, the AI opportunity in storage, you know, is definitely discussed a lot. I think it's, for myself and I think a lot of investors, hard to really tangibly see the evidence that it's a catalyst on this traditional storage market at this point. I think we can all see, you know, we see somebody like Weka and VAST and some of these guys that are in some of these large-scale, you know, AI, you know, cloud deployments. But how do you how does Pure think about AI, you know, as an incremental growth driver? What have you seen today, or what are we looking for, you know, over the next 12 months?
Yeah. It's a great question. And we really see AI, the impact of AI on data storage falling into a couple, you know, I had three high-level categories, right? Certainly, you've got a relatively smaller set of, you know, directly connected to GPUs, AI training environments that need high-performance storage. You know, secondly, you've got a set of, I would say more enterprise-driven AI deployments, folks that are doing inferencing, folks that, you know, are trying to deploy RAG-type architectures. And then, you know, the third set is really, I would say AI-influenced storage modernization, right? Enterprises that are maybe sitting on, you know, massive amounts of very siloed, oftentimes locked away on cold, slow disk data that they wanna make available to AI, but, you know, if you can't connect the data, how are you gonna go do that?
And so we see the effects on data storage really falling into these three buckets. And you know, we continue to pursue each one of them, right? We've got some great leading customers in you know the AI training space. Meta, as we've spoken about over the years, has been a great partner in their Research SuperCluster utilizing our technology. We were able to expand our partnership with CoreWeave this last month in terms of enabling their environment, their storage environment in the GPU clouds, and a whole host of other customers that you know in the AI training space. I would say the enterprise inferencing space is, I would say, earlier innings, to be clear. You know, in my discussions with the enterprise, you know, I see the whole gamut, right, of customers that are.
Sorry about that. I don't know what that was odd.
Nope.
You know, I see customers spanning the entire gamut of early adopters, folks that are leaning in and heavily experimenting, you know, to others that are really just trying to figure out where to get started. I think the enterprise inference opportunity is gonna still take a year or two to play out. I don't think the technology is consolidated quite enough yet. I think folks still have a number of things to sort out in terms of security and compliance and also just finding the right use case. So I think that will take some time to play out. I would say in some ways, the third category, AI driving customers to modernize, I would say that's one of the larger categories that we see distinctly appear.
So you mentioned and one of the questions I get every once in a while is, you know, the Meta AI RSC, you know, footprint. You know, there was, obviously, you know, news about, you know, a competitor getting some opportunity within that deployment inside of Meta. But your engagement with Meta's, you know, on that footprint is still very much active. Is that a fair assessment?
Yeah. No, absolutely. It's our, you know, our engagement with Meta on multiple environments, you know, continues on an ongoing basis. I'm personally involved in many of them. We just don't speak of, you know, most of those environments. And to be clear, you know, I think the competitor you're speaking of, I think that was a different project. That was Meta's. A very large company. And, you know, we're able to speak of just the RSC environment.
Yep. That's very helpful, and then on the CoreWeave side, just real quickly, you know, that relationship was announced a month or so ago. I lose track of time. But, you know, it was the cloud, the large-scale GPU specialized cloud guy that was alluded to a couple quarters ago. That is an Evergreen One relationship. So walk me like, CoreWeave's customers are given the option to use Pure under Evergreen One as a backend, you know, back support.
Yeah, no, so your recollection is absolutely correct, so I think it was 4Q of last year.
Right.
We talked about an eight-figure GPU cloud deal. That was CoreWeave. We can now say that was CoreWeave. That was an Evergreen One deal. And just a few words on that. You know, I think Evergreen One was an ideal construct for CoreWeave, because, you know, they're effectively a specialized cloud provider, GPU cloud provider to their end customers. And so Evergreen One gave them a ton of flexibility and optionality in terms of, you know, rate of ramp, rate of actual storage consumption, you know, optionality in terms of mix of performance versus capacity required, and just ultimately gave them one less thing to have to go and take risk and have to go think about. And so that deployment's grown quite nicely.
And really, the expanded partnership we announced, yeah, it was about a month ago, is to take the learnings from that, the success from that, and, you know, make that more broadly available for other customers within the CoreWeave cloud.
So that's your.
So now in that context.
System sitting in CoreWeave data center?
That's correct.
That you guys own?
That's correct.
Right.
And we own as part of the Evergreen One relationship.
Okay.
but that Evergreen One relationship is with CoreWeave. That's not exposed to the end customer. The end customer is buying services from CoreWeave.
Okay, so it's actually the consumption is actually coming directly from CoreWeave. It's not they're signing up a customer, well, I guess it's they're all kind of tied together.
Yeah.
But it's okay. With all that said, you know, I'm gonna go back to where I started. When you think about the strategy of Pure, you think about, you know, this hyperscale deal, you know, when we think about the growth profile going forward, what excites you the most? Maybe let's take the hyperscale and set it to the side. What else excites you about the growth ahead for Pure?
Well, I think we're in a great position. We play in a huge market. If we just look at core storage today, we play in a huge market that we're now in, you know, great position to go and expand within, right? That position is driven by a, you know, very complete but also very tight set of products and technology, right? One hardware technology, one software technology, two basic architectures that can cover the entire span of price performance levels, AI to archive, block file and object. We're now in a position to walk into the enterprise and not just have a conversation about, "Hey, do you have this workload?
I've got the best, you know, thing for you," but to be able to go and say, "Hey, how would you like to go and modernize your entire data storage estate, get all of these benefits, reduce a ton of risk, and just make storage something you don't have to, you know, think too hard about anymore?" And so there's, you know, we're in a position to do that against a competitive backdrop where, you know, look, our competitors are not keeping pace. They're not innovating. They're not, you know, I love my competitive set. Let me put it that way.
Yeah.
And so I think there's a ton of opportunity there. At the same time, we're continuing to push the boundaries of what people think are possible. You know, I think we were speaking before this. I think there were a lot in the market, the financial market as well as the industry that had doubts about our prospects in the hyperscalers. Hopefully, we've been able to allay those doubts. You know, but we're continuing to push boundaries in terms of performance levels that we can deliver. We're continuing to push boundaries in terms of business models, right? This idea of taking a true SaaS experience and bringing it to customers in their own data center. It is early innings.
We talked a little bit about the just kind of quarter-to-quarter dynamics on that, but it's absolutely the way the future's gonna go, and then also the cloud operating model. I think we're in early innings of just transforming the way people, you know, manage their infrastructure. It's super, super antiquated today, and it doesn't have to be.
Rob, thank you so much for taking the time.
Thank you.
Appreciate it.
Great discussion.