Everpure, Inc. (P) Earnings Call Transcript & Summary

Perfect. So why don't we get started here? I'm Aaron Rakers. I'm the IT hardware and semi analyst here at Wells. And I'm pleased to have with us this morning, Rob Lee, the Chief Technology Officer from Pure Storage. Before we start, Rob actually said that he would have the pleasure of reading the safe harbor statement. So go for it, Rob.

I'll try to do this in my best legalese voice. Statements made in these discussions, which are not statements of historical fact, are forward-looking statements based on 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 our Forms 10-K, 10-Q and 8-K. And with that...

All right, Rob. So that was a great job. So for those of you that aren't familiar with the Pure Storage story, those of you in the audience here and on the webcast, company reported pretty solid results last night. Continuing, in my opinion, to highlight what really is the fundamental building block of why we're constructive on Pure, which is the architectural differentiation from the platform perspective.

So why don't we start there? Rob, if you wouldn't mind, give us a quick overview of Pure's differentiation in the architecture approach and how that's competitively enabling you to win in a fairly competitive landscape in all-flash storage.

Yes. Absolutely, Aaron. So I think there's a couple of things I'd point to in our core technology decisions that we made really going back to the beginning of the firm that have expanded our differentiation over time and really are driving a lot of why we win today. I think the 2 main ones are really the software architecture, which we've designed specifically to work with semiconductor flash. And the second one is this idea of everything that goes into Evergreen, right? This idea that the software, the hardware that we build is really designed to continuously be upgraded, continuously be modernized, all of that creating disruption. And so why did we make those choices, right? When we started the firm and we looked at the possibilities that flash had to offer, it was pretty clear at a high level, top line, great. It's like a ton of performance, a ton of reliability, the idea that you could build a much denser and much more reliable systems potentially, great. Now the details get a little bit trickier, right? To work with a flash, it works very, very differently than magnetic disk. And to extract all the right benefits out of that, to break down the price differential, to create efficiencies and the cost structure required a fundamentally different approach to the software, a fundamentally different approach to how we treated the storage media. In the most simplest terms, if you think about a spinning disk drive, right? You've got a motor, the disk drive spins around. If you want to read or write data, you've got to locate the head in the right place in the disk and basically read and write the data as the disk flies by. Well, that drives you to make a certain set of decisions. For example, if you want to make that really performant, you're going to lay out your data sequentially, so you move the motor once and you kind of slurp the data as that disk flies by. The way that semiconductors work, very different, right? The idea that, hey, without spinning parts, I can go and access all parts of the flash chip simultaneously now opens up a whole different set of possibilities in terms of driving performance. Now to take advantage of that, you've got to make a lot of changes. You've got to make software changes in terms of how and where you store the data. You've got to make changes in terms of how you keep track of what you wrote in different places in the media. And we recognized all of these necessary changes as really an opportunity to come in and disrupt the traditional incumbent storage industry because we knew that our incumbent competitors and really, our larger competitors today, were built on and based on large legacy software stacks that had made a set of decisions that were optimized for hard disks that really fundamentally could not be easily retooled for flash. And so if you play that forward a decade, those sets of decisions have really enabled us to drive tremendous cost efficiencies, tremendous density efficiencies, data reduction efficiencies, especially as we've integrated that software as well into our hardware approach with DirectFlash. The second piece of this I would call out is we recognized upfront that semiconductor technology was capable of just being far more reliable, right? And when you look at the disk industry, the disk-based systems industry, people buy disk-based systems, they expect them to last 3, 4, 5 years. And then they expect the disks to fail and they expect to go and replace them, right? We knew that flash, that solid state, was capable of much longer service life times. And when you go into it with that mindset, well, now you've got to work with your customers to go and take them through modernizing the software and all of the surrounding hardware of those systems. And so we've built a lot of those capabilities into our core software, which have now become really the underpinnings of Evergreen and our Evergreen subscription set. So a long-winded answer to the first question, but if I look at the fundamental differentiating principles that we have, they really go back to those 2 core pieces of technology that trace their roots all the way back to the beginning of the company.

Yes. So Rob, I mean, I could ask you 40 questions underneath of what you just outlined, and I'll try and kind of keep this somewhat succinct. But really, what you're laying out for us is you've got purity at the operating system level. You've got DirectFlash as you mentioned. And really, when I look at that competitively, you're competing against architectures that retrofitted for flash vis-a-vis SSDs. So I'm not kind of making a statement but asking a question. Like your ability to drive an operating system that sees the entirety of flash as a singular pool enables that efficiency architecture. How would you characterize that efficiency? If you look at your solution versus -- not to name names, but a competitive solution that uses the SSD off-the-shelf architectural approach, how much more efficient do you guys typically see yourself?

Yes, it's a great question. So I think you can break efficiency down in a couple of ways, right? So number one, a lot of the technology you just described allows us to do just a far better job of data reduction, logical data reduction, de-duplication, compression, than legacy software architectures. And that's where we rely across the fleet drive, 4, 5-into-1 data reduction rates for our customers. We don't even give them an opportunity to turn it off because there's no performance impact. If you compare that to the competitive set, right, typically, customers are forced with a trade-off of, hey, do I want some data reduction? Am I willing to pay the performance penalty? And so you get significant efficiencies there. If we look at the core flash technology and efficiencies there, I would look at, for example, our DirectFlash technology compared with enterprise SSDs, right? Two things there. One is we're able to drive, kind of, call it, order 20%, 25% raw efficiency just in the use of flash. You take one of our 18.3-terabyte DirectFlash modules and you look at the flash chips that are inside of it and you look at a off-the-shelf 15-terabyte SSD, you'll see largely the same number of flash chips, the same actual flash chips. But we're able to drive incremental, call it, 20%, 25% of raw space out of that. Well, why is that? Again, it comes back to our flash technology, the fact that we see -- the software sees all of the flash in that drive as one big pool. We can be a lot more efficient with it.

So that's kind of the over-provisioning attributes of how better you are in wear leveling, air correction, all of those attributes of that architecture?

Exactly. And the other piece of it is because we are so much better in those attributes, we drive less wear to the drive, the drives are much more reliable, have much longer service lifetimes. And so again, as you start thinking about -- as our customers start thinking about total cost of ownership, that becomes a very significant driver as well.

So on top of those building blocks, I think one of the things that I've been impressed by with Pure over the last many years of covering the stock has been you've been at the forefront of kind of key technology trends, right? I think you were first to incorporate native NVMe in your architecture. You've definitely -- I think, if anybody didn't hear the transcript or see the transcript last night, are leaning forward on QLC. Is there other things that we think about the iterations of architectural changes looking forward that further catalyze flash, which still today is only 15% of total data storage capacity shipped? Further drives it farther as far as penetration opportunity?

Yes. No, absolutely. I mean I think we're really just getting started, frankly, with QLC. I mean it's growing. Our QLC products have been growing extremely well. We talked about last night on the call continued great performance out of FlashArray//C. FlashBlade//S is off to an amazing start. But I think we're just getting started, Aaron, to your point, in terms of the mass replacement of disk. There's a long way to go there. And I think that, that transition really is catalyzed by the same set of technologies I just spoke about. One of the reasons why we've been shipping FlashArray//C for almost 3 years now, just about 3 years now, and we're really the only enterprise vendor that's shipping all-QLC arrays. One of the reasons why we're kind of out in the field alone is that in order to make QLC cost-effective, reliable and enterprise ready, you really need this set of technologies, right? You really can't go and accomplish this around enterprise SSDs. And so this is a space where we think we've got a tremendous advantage, and we're going to be leaning in hard to.

And to be clear, last night, I mean, the comment that was made on the call, which I thought was definitely incremental, was not only has QLC -- and again, for those that are not familiar, that's a -- you get 30% plus more capacity per die on a piece of NAND flash. But not only has it approached or crossed through the TCO level of hybrid, but you're moving towards -- I think you made a statement that it would be equivalent, on a TCO basis, relative to a nearline capacity. Is that as we see it today, just the progression of the product portfolio? Or...

Yes. Yes. No, it's a great question, and let me expand on that. So if we look at on a TCO basis, which accounts for certainly, power, cooling, savings, operational, like human operational costs, break-fix as well as asset lifetimes, I would say today, we're in a zone where there are some customers that -- sorry, on a TCO basis, our QLC products, I think, largely have crossed over with nearline and hybrid. The more interesting crossover zone is on the acquisition cost, right, where you're just looking at day 1 acquisition cost of the array, QLC-based with Pure or the nearline disk alternative. I would say that we're in a zone now where there are some customers, depending on data reduction rates, depending on the configuration capacity and density, there are some customers where we have crossed over with hybrid and are in the zone with nearline today. I would say that there's very -- many more customers that are right on the brink of that. And there's going to be some that are further out. But on a TCO basis, I think we're pretty much there.

Yes, because it never has to be price per gig parity, right? It's when you start getting into the attributes of flash and you think about power consumption, footprint savings, there's all other pieces of that TCO equation.

Yes. And I think the other thing to recognize also is that as the media costs decline, right, and the cost of the acquisition cost declines overall, the other cost elements that go into TCO have an out-weighted effect, right? So said more simply, in the day when you were spending $1 on the storage and $0.05 to power it, the overall weighting of the power was noticeable, but it maybe wasn't going to swing the decision. If you get to the point where you're paying $0.10 for the storage and you're paying not $0.05 anymore but $0.10 for the power, now all of a sudden, your TCO equation starts shifting pretty dramatically, if that makes sense.

That's a great point. I'm going to shift gears here a little bit because one big part of the Pure Storage story has been it was commented that you've gone into the second phase of the deployment last night with Meta. Maybe give the audience and those in the web a quick review of what your role is in that Meta deployment for their AI research supercluster, where we're at in that deployment cycle. And ultimately, I just want to kind of simply ask, like why did Pure win?

Yes, absolutely. So a bit of a description of the Meta research supercluster environment. This is an engagement that goes back really over 5 years. We began working with the Facebook team at the time in one of their AI research environments, supplying them with FlashBlade arrays to do direct training to their NVIDIA GPUs. As that engagement became successful and that project on the AI research side grew within Facebook, they decided to really scale it up into production service. And we started working with them as they were spec-ing the infrastructure for that. And to the second part of your question, as they started looking at their needs for that production service, the couple attributes that became the most critical -- certainly, price and cost was always there. But performance, naturally. But space and power were very critical, right? They literally had data centers built out. They had power limitations. And fundamentally, they needed to get as much storage in there in the smallest amount of power envelope, so they could use their power budget for GPUs and processing, right? And so when it came time to looking at really all of the options at the table, I'd say, certainly, we won the deal. Certainly, it was helped by our prior engagement with the Facebook team. But really, we were the only option on the table that was going to meet the needs and the balance of needs between price, performance, physical footprint and power. And so as that engagement has grown and what does it look like, so we still supply the direct training piece with FlashBlade. The larger part of that environment and footprint really though is served by FlashArray//C. And we think of that as -- and it's really being used as more of a bulk data storage environment, right? So think about in an AI training environment, as they're working those algorithms, the algorithms have to go work on large pools of data, whether it's images, video, text, so on and so forth. We're supplying the bulk storage of that training data on FlashArray//C. When we first started talking about the shipment in Phase 1, I think it was 3Q of last year, Facebook subsequently came out and wrote a blog and kind of described their intent to that environment. At the time that blog came out, we were in Phase 1, about 185 petabytes -- 175, 180 petabytes. They indicated that they see with that service scaling, growing to an exabyte over time. And last night on the call, we described shipping the majority of the Phase 2 deployment to them in the quarter.

And those phases are kind of equal in deployment sizes?

Well, I can tell you that the Phase 2 shipment was kind of in line with what we set expectations at the beginning of the year, which was kind of what we shipped last year plus the company growth rate, so kind of in line with that from a volume and capacity basis.

That's perfect. And I'm not asking you to give guidance, obviously, and you won't say going forward. But Phase 1 and 2 are Phase 1 and 2 of what? Like how do I think about the progression to get to that full -- is it still 1 petabyte of deployment?

1 exabyte.

1 exabyte. Sorry.

Yes. So I think when we think about Phase 1, Phase 2, there's still a lot left to get to 1 exabyte. Hard to kind of say and forecast when that might be. But the engagement is going quite well, and our visibility and really expectations around the environment haven't changed.

So I think this whole deployment has kind of caught my attention, obviously, not just because of the size of Meta but just the thematic architectural attributes of flash and what that might play when we thought -- when we think about AI infrastructure build-outs and the need to feed these GPU engines and the capacity and the deployments involved in that. Would you characterize Pure as being in a position now to see opportunities at other like deployments beyond just Meta as we think about some of these build-outs?

Yes, absolutely. And I'd actually answer that in 2 ways, right? So if we look at the AI angle of it as well as the hyperscaler angle of it, on the AI angle, AI and analytics in general continues to be a very strong segment for us. I think it's a workload set that our technology is uniquely suited for, right? If you step back from the bits and bites of AI, fundamentally, it's a space where the more data you feed into the algorithms, the better results you get. Well, how do you feed a lot of data into the algorithms? You need a really, really fast storage, right? And that's really, at the end of the day, why we serve so many customers with our AI solutions. Now certainly, as the larger tech titans and hyperscalers deploy huge AI environments, those benefits multiply and get even more profound. On the general hyperscaler front though, right, because again, I'll remind you, even in the Meta environment, the larger part of that deployment with FlashArray//C really isn't specific to AI, right? It's bulk data storage, storing a lot of content, which happens to end up getting fed into AI engines. In the general hyperscaler environment, this is an area where we think we have a very large opportunity to go after replacing nearline disk, right? If you look at enterprises versus hyperscalers, I think hyperscalers are actually a little bit -- I would say a little bit behind enterprises in terms of mainstream deployment of SSDs and flash technology. And I think there's a lot of disk that we can go after there with, again, the core technology that drives the attributes that led to our success at Meta: price, performance and power and cooling savings.

Yes. So that's kind of a very interesting thought in my mind is that I'll just ask you the -- like as these cloud guys and you see the amount of capacity that they've deployed in nearline hard disk drives, a lot of their architectural approach historically has been deploy that, share that data or shard that data across x86 compute commoditized server platforms. Given the size and scale of what they are today and the attributes of flash, is their views changing on how they think and that's the opportunity for Pure to say, hey, Pure has had this operating system that's been optimized around flash. They don't want to go and rethink their stack for flash in corporations, so that opens the door for Pure? Does that make sense at all?

Yes. No, it does make sense. And I think that's part of the opportunity, right? So if we take a look at the larger hyperscaler environments, they have been on these special-purpose disk-based designs for a long time, really riding the wave of density in disks and continued cost improvements. They're now at a point where, look, they've got to find a way to bridge the gap over to flash, right? It's pretty clear that disk isn't going to continue to keep up. And once you kind of confront yourself with that and you realize, hey, to get the economics I need, to get the efficiencies I need, to get the density and reliability I need, I'm not going to be able to do it with SSDs, right? And so then what are my choices as a hyperscaler? Well, A, I could go and try to develop everything from clean cloth, right, and reinvent a lot of the IP that I've spent some time talking about here today. Or especially in an environment where I'm having to be a little bit more judicious about where I spend my R&D resources and decide, hey, is this really the highest value area for me as a hyperscaler to go invest in versus partnering with the market and technology leader, I think that creates a huge amount of opportunity as QLC and other semiconductor technologies really continue to improve.

Yes. That's fascinating. So I want to shift a little bit outside of just the cloud stuff and maybe broader a little. I think one of the things that Pure has done over the past many years is really emphasize and Charlie and team have focused on the idea of a portfolio motion, right? You started with the FlashArray, good success. You went into FlashBlade. All at the same time, keeping that core competency, which is Evergreen, wrapped around it. Can you talk a little bit about the success or some of the attributes of the portfolio motion that Pure has been able to do over the last couple of years?

Yes, absolutely. I mean it's been a huge driver of our continued growth and especially, I would call out the penetration of the enterprise, right? Five years ago, we could walk into enterprise accounts and say, hey, we've got the best product, best solution to go and accelerate your database environment. And they might say, great, yes, we absolutely recognize the benefits. But hey, can you help us out with these file environments? Can you help us out with these other workloads and applications? And without a broader portfolio to be able to go solve more of a customer's needs, it really becomes quite a bit of a hurdle, especially in the enterprise, to kind of bridging the gap from being a point provider to a really strategic vendor. And so as we broadened the portfolio, certainly with file an object with FlashBlade and the Tier 2 solutions, with QLC, that's absolutely -- and Portworx as well with the container space, that's absolutely been a tailwind for us as we've gone and really grown the enterprise business. And I think that's something where we're starting to see the fruits of that in both directions, not just using the FlashArray product to entry point and installed base to go and drive additional sales of FlashBlade and Portworx. We're also seeing a lot of cases where we're breaking the Fortune 50, Fortune 100 companies with Portworx initially, and then that's leading to downstream FlashArray and FlashBlade sales.

Is there any metrics you could share? And if not, I can appreciate it as well. But how much of your customers have bought more than one -- not one system, but outside of...

One type of product across the portfolio?

Yes.

We haven't broken that out in separate disclosure, but I can describe it as significant. Yes, significant.

And that's really changed over the course of the last...

Absolutely. Yes. It's something we track very closely internally. It's changed. I would say it's changed for two reasons. One is, certainly, the breadth of the portfolio and the maturity of products certainly has helped it. But also our sales teams, right? Our go-to-market teams really recognize that the power of the portfolio help them win in any of their engagements. And so they're now starting to grab on to all parts of the portfolio.

Yes. In the final couple of minutes I've got left, because we touched on it, you mentioned it a couple of times, but I kind of want to make sure it's well understood is this Evergreen attribute, right, which is I think one of the fascinating metrics that we've seen out of Pure's -- you're going to correct me if I'm wrong, I think 97% of the arrays that you sold 6-plus years ago are still in the active installed base. I think it was the metric, if I've got it right. But the idea of the recur-ability of the business wrapped around Evergreen, just help us appreciate what Evergreen is.

Yes, absolutely. And the one caveat to that metric is it was 97% of arrays, I believe, sold in 2014 at the time are both still in service but more importantly, have been modernized to look like new, like arrays that we've sold in the last year or two, right? And that's really the distinction. The principles of Evergreen really are that when we sell an array, our commitment to the customer is to be able to go and continually modernize that array, hardware and software, completely nondisruptively. And that nondisruptive piece of it is what makes it so sticky, right? A customer that's come from a competitor where every 3 or 4 years, they're forced to do a very disruptive, hard-to-manage migration, once they do a nondisruptive upgrade, nobody says, hey, that was a great experience and all, but I'm going to go back to the old way of doing things. And so every time that we Evergreen upgrade a customer, that's a competitive refresh and rebid opportunity that we've taken away from the competitors, right? That's in a state we haven't put up for bid. Conversely, right, every time a competitor estate comes up for refresh, we love that, right? That's a fresh new bid and RFP opportunity for us. So it's a huge competitive advantage in terms of customer acquisition, but it's also incredibly sticky.

And again, just to hit home on that idea is that, that started from day one. That was the architectural decision you made, clean sheet, OS with that upgradability, that's something that your competitors can't really replicate as far as ripping out the controllers and the SSDs and replacing nondisruptively in their installed base.

Yes. It's just one of those things that's really, really -- I don't want to say impossible, very, very hard to retrofit, right? It's like -- it's just like you can't take something that was inherently complex and retrofit simplicity on top of it. It's very, very difficult to retrofit some of these things on.

Rob, 1 minute left. I'm just going to ask you kind of the easy question, which is did -- when you think about in your role as a CTO, is there anything that we -- I didn't ask you about, architecturally, that we should be thinking about as we watch the Pure story continue to evolve?

No. I mean we'd a pretty broad-based discussion here. I mean I think the things -- there are a couple of major vectors I'd call out. Certainly, we spent some time talking about flash. I think we're just getting started with the nearline replacements, and a lot of that is driven with the core flash technology and the densities we're going to go and drive. I think the other piece is if you look at what we've done with the portfolio with FlashArray//C and FlashBlade//S, really converging a lot of the same core technology now with the QLC DFMs, that's going to give us quite a bit of just internal leverage and efficiency. But it's also going to simplify lives of customers, right? To be able to share equipment and so on and so forth. And then the third piece of this is really what this means for Evergreen, not just Evergreen//Forever but Evergreen//One, right? We're going to turn all this hardware and product-level flexibility into service-level flexibility as we continue to build out the Evergreen solutions.

Perfect, Rob. I appreciate you joining us today. Thank you.

Thanks for having me.

Thanks.