Okay, excellent, so welcome, everyone. I for the people that are coming in over the internet, if you don't know me, my name is Mark Moerdler. I'm the Bernstein Senior Analyst, in charge of global software. Have been at Bernstein now since 2010, covering a very wide range of companies and capabilities. I'm really pleased today to have with me the team from Pure Storage. And maybe what I'll do is allow them to make a little quick introduction, and start maybe with a question. We have Charles Giancarlo, the CEO of Pure Storage, and Kevan Krysler, the CFO of Pure Storage. So you know, maybe I'll start this way. Oh, let me. I have one important one that I need to make as requested, and that is 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 to in Pure Storage's most recent SEC filings on Form 10-Q and 10-K and 8-K. So now that we've got the.
Thank you, Mark.
Mark stuff to the side.
Yeah.
And you didn't have to do it. I had to do it. So, Kevan, why don't we start this way? Can you give us an overview of the company and the company's vision for the future?
Actually, I'll probably take that, so in overview of the company, I'm gonna give a very quick historical background of the company and then really go to our vision going forward. Company started in 2009, so we're a little over 15 years old now. We started under the concept that flash storage, meaning NAND, the same thing that's in cell phones and all your personal devices at this point, would eventually make its way, which was a new thought at the time, into enterprise storage. At that time, all enterprise storage at every sort of price performance level was all done with hard disk, and we were the first to come out with enterprise storage using flash. Flash at that time was very, very expensive compared to hard disk, almost maybe 30 times, 40 times as expensive as hard disk.
So we had to do a lot of things to make it economical at a system level, when compared with hard disk. And we were quite successful in those early years, grew very, very rapidly in what are generally known as primary use cases: high-speed databases, high-speed trading, things such as that. Fast forward, 15 years, we have developed a broad product line, fitting every different type of storage from there are several different types of storage. There's block, file, and object. There's large scale, small scale. There's scale up, scale down. We've been able to cover all that space with one software operating system. Another fact about the storage environment is generally every different type of storage, even if it's covered by a single vendor, they have many different operating systems to cover that vast array, if you will, of storage.
We do it with one operating system. We stayed very, very focused on developing that environment. The other thing to keep in mind is other vendors have added flash storage to their hard disk storage offerings. They did that with SSDs. SSDs were specifically designed to mimic hard disks. And when you take a semiconductor and make it look like a mechanical device, you suboptimize that semiconductor. We're the only company that has software that manages raw flash. And we do it at the full system level rather than individual SSDs. And that gives us a price performance advantage in the use of flash. Okay. So that's generally where one additional thing that I'll mention that we built early on that distinguishes us is we've designed our software and our systems to be consistently and forever non-disruptive upgradable.
Typically in the storage market, actually typical in the entire IT systems environment, a product gets old, eventually it gets old enough where it goes out of service. The customer has to replace it. And when they replace it, first of all, it's a new purchase, you know, of the same thing effectively, just the modern version. So it's a new purchase for the vendor as well. But in addition to that, it's disruptive. The application comes down, they have to replace the product, and they bring up the application. It's why enterprise systems might be down on any particular weekend while they're going through this upgrade process. We developed a capability where we can upgrade our systems on an ongoing basis, non-disruptively forever. And that means our products never get old. We can continuously upgrade them without the customer having to take down their environment.
We call this Evergreen. And it's the basis of our as-a-service offering where we provide a storage service as just a set of SLAs. And we manage that entire environment. And it's a true SaaS offering, but we can provide it where the storage is actually on the customer's premises if the customer prefers that. So it's a very flexible both on-prem and off-prem storage service that is provided fully as a service where we manage and we own the infrastructure. So that's where we are today. We're now just over $3 billion as of our announcement yesterday, quite profitable, adding to the profitability every year. Our vision now extends in two major directions that I'll point out. One is, last quarter we were able to announce a major design win with a hyperscaler.
The software that I talked about that has our software managing NAND flash in bulk allows us to have better price performance than SSDs and to match the total cost of ownership of hard disk storage in the hyperscaler. This then allows us to start replacing the hundreds of exabytes that exist in hyperscaler storage with a flash solution that meets the TCO of cheaper disks and at the same time improves their overall price performance at roughly one-tenth the space, power, and cooling of hard disk storage. So this allows them to save about 20% of their current data center footprint power and to put it into other things such as AI. So that's one area of potential growth for us.
The second is that we have developed software now, and I've got a little bit of something to show. Believe we all used, if you remember what these are, it's an external hard drive. For those of us that are old enough, we used to use it attached to our laptop or our desktop, and it would provide extra storage, right? But if your neighbor had one, as well, if you ran out of space of storage and you wanted to borrow some of the capacity of your neighbor's, well, that wasn't really so easy or practical. You'd have to move it over, so you'd have to go to the store and buy a bigger one, and then you'd have to move all the data from one to the other. File sharing wasn't so easy.
You'd have to attach it to an email or something like that. So very impractical, but it did provide extra storage. Believe it or not, that is the way that IT enterprise storage operates today. A storage array is nothing different than an external hard drive on an application stack, in the IT environment. What we're about to do, what our new software release allows is for all of our arrays to operate like a cloud of storage. Just like all of us have moved to cloud storage, the benefit is not only unlimited storage that can scale, just by us paying them a little bit more, not only the fact that it never fails and that the service keeps getting better without disruption, but you can easily share files. That's what we're bringing to enterprise storage.
All of our arrays will now operate as a cloud of storage in the not just in the data center, but across the global enterprise. It will allow easy file sharing across applications in the enterprise, which does not not just file sharing, data sharing that does not occur today. Data is captive in a data silo tied to an application stack. The only way to get access to it is through the application, which adds cost to the application rather than allowing real-time data to be accessed directly, let's say by an AI agent or anything else. So it's a fundamental change, bringing the IT environment finally virtualizing the data environment in an IT environment, and allowing the enterprise to place their policies in software rather than being managed manually. So that's, you know.
Yep. Thank you. Thank you. And I apologize for misnaming.
No problem.
Don't worry. I plan to ask the financial questions also, but maybe, Charles, can I follow up on that? How do you think today and long-term about what the right mix of hardware and software for your business?
Yeah, so it's always an interesting question because 95% of our engineers are software, you know, but of course we sell it wrapped in tin, you know, which is the hardware of the product.
Right.
But you know, our margins are driven by software. The value that we get, the wins that we get are driven by the software elements of our product overall. You know, as we go into the hyperscale environment, we have announced this as well. That'll be more of a net model. In other words, the hardware over time will be effectively procured or purchased directly by the hyperscaler themselves or their agents, and the value that we'll be providing as software. If we look at our storage as a service, which we call Evergreen//One, you know, it's interesting. We talk about software, but what is a SaaS offering? A SaaS offering is different yet again.
Right.
than just software, right? So, to our enterprise customers, we are increasing the percentage of our value that we provide to them as storage as a service, what we call our Evergreen//One offering. And so that's the direction we're heading into. It's less about providing our capability of software, more about just taking away the entire responsibility of managing the products, or operating the products, or even having to think about managing storage to our customers. Instead, we'll manage the storage, they'll manage the data.
Okay. So lots of places we can go from there. Maybe I'll ask, I'll ask Kevan, how does this direction impact long-term financials?
Oh, in a couple different ways, right? So when we think about what Charles kind of laid out in terms of the vision of the company, you've got a long-term growth potential. I think the two significant areas of growth potential is obviously the hyperscale design win is just the start of what we're seeing on that front. Charles spent a lot of time, in this earnings in particular, talking about Pure Fusion version 2.0, and really the, and that's really laying the groundwork in terms of our stages of growth, in terms of long-term, sustainable growth, again, in double digits, which we think are really gonna be driven by these two tenets in addition to the services.
Our Evergreen subscription services that we've talked about, our Evergreen//One services, all being launched in on the underpinning of our software, our Purity software. And so I think on the top line, we've got a lot of growth potential available to us to take advantage of. And then when we think about gross margin profile, we saw some nice gross margin expansion in our subscription services. Again, that's really around our Evergreen subscription services, Evergreen//One services, which is our storage as a service. And then product gross margins. And our product gross margins have historically been quite high and quite strong, really again, driven by our Purity software. Now we have expanded very aggressively against disk-based solutions with our E Family. So we saw a little bit of a drop in product gross margins temporarily that we talked about this last earnings.
But again, with the gross margin profile ahead of us with the hyperscale design and a licensing model that, Charlie talked about, we see some nice expansion, opportunity on that front. Now we will invest this year, in the hyperscale design win, especially, on the R&D side in terms of, really qualifying new NAND suppliers, building out our supply chain to, develop more DirectFlash modules at scale, at similar to what you would need to, provide for a hyperscale, type environment. So we will keep our operating margins flat, maintain them at 17%, which we've talked about. But after this year, we believe that we'll get back to an aggressive expansion of our operating margins, similar to what we've done, historically.
Yeah. So where does Pure, Pure Storage, where do they play in the cloud, in a cloud-first world? If we, you know, AI is gonna shift things theoretically more fair, more quickly to the cloud, how do you play in that space? Where does the biggest value come?
Yeah. We have, I'd say three ways in which we play in the cloud environment. One is what we just talked about, which is there'll be customers that will want to maintain their data in their own control for a variety of reasons.
Okay.
and what we talked about in terms of storage as a service, our Evergreen//One offering, they get the benefits of operating like a cloud, being able to view their data as a cloud, but also being able to operate it on-prem, you know, in their control. And so that, we see that increasingly attractive to enterprise customers. The second is our software, in addition to operating on our products, also operates today on AWS and Azure. And our container-based software, which is called Portworx, operates on all of the hyperscalers as well as bare metal. So this allows that enterprise data cloud that I spoke of to exist not just on top of a customer's own assets, but even our software that may operate in AWS and Azure.
So that all appears as one cloud, prior to a recent renaming of a company we all know. I used to refer to that as a meta cloud. I can't do that anymore. It confuses people, as you might imagine. But you know, so it's a cloud of clouds, if, if you will. That is a second way because our software that operates on AWS and Azure, as well as Portworx, operates exactly the same way from the customer perspective as our software on-prem. So they get to see that as one operating environment. Then third, as Kevan, you know, also mentioned, we hope to win either way. If we're able to achieve more design wins with more hyperscalers, ultimately that'll. Anything stored in the cloud will land on our storage and we'll benefit there as well.
Why will someone pick your storage on top of a hyperscaler versus using the base capabilities that they have?
Right. So there are two reasons. One is the base capabilities of storage at the hyperscaler lack a lot of enterprise features, lacks snapshotting, it lacks replication, lacks, you know, a wide variety of capabilities that customers use to manage data. So, and it's entirely different than what they're used to on-prem. So if they wanna move one of their enterprise applications into the cloud, they then have to somehow build themselves all of the data management capabilities that typically come in enterprise scale storage. So having it be consistent is a positive.
The second thing, and this is maybe the most, in a way surprising part of this is the combination of what we charge for our software in the cloud and the underlying assets that need to be used in the cloud to make it all work is less expensive than the base capability of cloud storage. And you might say, well, my gosh, how does that work?
Right.
Well, there are a lot of different reasons for it, but I'll break it down into two. Cloud storage does not have two capabilities that enterprises use all the time. One is called thin provisioning, and the other is called data reduction. Data reduction, I think we all understand. We take out some of the redundancy in data and we can compress it down to maybe a third of its original, three to one, four to one, something like that. But thin provisioning is even more important. Let's say that everybody in this room wants a gigabyte of cloud storage. The way the hyperscalers work is we all buy a terabyte of cloud storage, and we all pay for a terabyte of cloud storage, and they reserve a terabyte of cloud storage.
But you don't all use your terabyte of cloud storage at the same time. Thin provisioning simply says you're all entitled to a terabyte of cloud storage. It looks like you have it, but the fact of the matter is it all averages out. We don't actually have to provision a terabyte of cloud storage. So we do that. We only charge customers for what they actually use. So it's a huge difference. It allows a dramatic reduction in the actual amount of cloud storage that customers actually need to pay for.
Very interesting. Very interesting. Let's go back to the concept of data as a service. So that is on-premise and cloud.
It is on-premise and in cloud or a mix. Yeah.
And so, from the client perspective, it looks like a single uniform environment.
Correct.
No matter where it may be stored.
Correct.
So they don't really need to think about their incremental storage in the same way.
Correct.
I don't need extra drives in California.
Correct.
You know, I put it where I need it, and then I can use the extra wherever I need it.
You know, it's even more than that because if you think about an array tied to a single application stack, I might have a second one over here, and that one has extra capacity and this one doesn't, but I have to buy more here.
Right.
And so with our system, whether it's in the same data center or multiple data centers, data is placed by policy rather than having to worry about each individual system.
Beautiful. By the way, for those in the audience, I do have the tablet. So if you wanna ask a question, you can put it up by Pigeonhole or later on, maybe I'll even ask around and we'll move a microphone around. But just if you want to let, you can put it here and I will keep an eye on that one. You liken the value of Fusion at Pure to the value of vCenter at VMware.
Yeah.
I worked with VMware's tech years and years ago before I came to Bernstein, so it'd be interesting to see where and why you would say it's the equivalent.
Right. So if we think about the original value or actually the current value of VMware on a server, on a single server, it was that in the prior to VMware, you would run a single application on a single server so that you had scalability on it. But you also couldn't run different operating environments on the same computer. That was just not possible, and a lot of times this computer might be 20% only using 20% of its capacity, but you couldn't put anything else on it, 'cause it had to have this, it was designed with a single operating environment. With VMware, you could use the full server to full capacity 'cause you could put multiple different workloads on it, right? That's all well and good.
But then in terms of managing an entire data center where you really wanted to manage lots of different workloads that might need more servers over time or more capacity, what vCenter allowed is for the dynamic placement of workloads on computers. It allowed the backup of work environments from a server so that you could easily go back to a former version. It allowed, if you will, application management on a large scale, right? Almost on a global scale. That's what vCenter did. If you think about what we're doing with Fusion, it's the same thing for data replacement. We allow for the efficient placement of data across a large scale data environment, and we allow for policy-driven management of that data on a global basis. Let me just explain what that might mean.
Please.
There are companies that have lots of policies about their data. Some of it is actually compliance with regulation, right? In terms of who can see the data, in terms of where geographically that data might be able to sit. But also there's cybersecurity concerns. You know, how many times is that data replicated or copied? Who's copying it? Where's it going? All of that today is managed manually. And in terms of managing or keeping a record of where the copies are, it's a common cybersecurity question I ask in board meetings everywhere because generally there are no records kept of what's been copied and where it's going. What Fusion allows is for any copying, any replication, any snapshotting, any placement to be done by software, by policy that's set in software.
And once that software is set, you'd have to go outside of that software to make any change. And even that will be kept a record if they're doing it through Fusion. So we keep a catalog. The standard management of the data is done by policy that's been set by the enterprise, and therefore it can be standardized on a global basis. It's not done manually by local teams anymore. It's done by software.
You really look at it as the movement of all the smarts and all the data out of the storage itself up into a software layer.
Into a control layer. That's right.
Right.
A control layer above the storage teams.
Right. And so therefore, in theory, you have you don't have to worry about the availability of a specific drive A from a certain manufacturer, certain configuration, whatever goes away. Today you have the problems of how do you deal with that because you may be mixing and matching and everything else. You don't have to worry about those same types of things.
You don't have to worry about. We're bringing it to the point where you don't have to worry about an individual array. You know, you'll have a team that's worried about the total capacity or their total performance.
Right.
But they won't be managing the data anymore. The data will be managed in software.
Right. And you don't have to think about the same way about the hardware itself because you're not touching it.
That's right.
Just like we went through in the early days of the PC where we had to deal with all the stuff.
Correct. We're abstracting the hardware away from the, that makes sense.
Yeah. Okay. You have now, you offer Pure Cloud Block Store for Azure and AWS.
Right.
Okay. How exactly does that work? Do you procure that through the vendor? You procure it from you? Who manages it? What's the process? What's the differentiation?
Up until a recent new entry that we've put in place, you generally procured it through us.
Right.
and operated on top of one of those clouds. It is available as a credit against your AWS or Azure, against the marketplace, against whatever your commitment is.
Right.
And you can get it from the marketplace, but generally most customers procure it through us. We're currently in public beta of a new Azure service called AVS, which is Azure VMware Service, where Cloud Block Store will be one of the checkoff items and it will be fully Fusion enabled. And what this means is now a customer's entire VMware environment that's on Pure on-prem will be completely transparent to an AVS environment in Microsoft, both the compute side, which is VMware, as well as the data side, which will be on our Cloud Block Store. Makes it much easier to make a transition from VMware on-prem.
Right.
Onto the Microsoft VMware license.
Right. So you use it as a stepping stone and then you keep using it 'cause it gives you all the value and advantage of it.
Right.
In an Azure environment because the large enterprise is looking for that level of service.
Correct. And it's the simplest for customers that are looking to get off their VMware license.
Right.
It is without a doubt the simplest, most straightforward way to do it because it is VMware. It's just now in the cloud rather than on the Azure cloud rather than on-prem.
Right, and so they can then build new stuff in the cloud side or on the, not on the, VMware side as required.
Correct.
Okay, so on subscription, okay, how the software is obviously software that comes with the hardware itself, but the real model that you're moving to is it's a true ratably recognized subscription model that you're gonna be charging more and more on a go-forward basis, especially it sounds like in with.
Evergreen//One for years now has been entirely SLA based, service level agreement based.
Okay.
You know, customer, it's our choice as to hardware that we use.
Right.
We're just signing up to SLAs regarding performance, capacity, reliability, power, space, and cooling because one of the things we do, if you think about a service, any type of SaaS service, generally the SaaS vendor owns the data center or some of them, actually, of course, are now in the hyperscalers, so they may not even own the underlying asset, but what they do provide are SLAs for their service, right?
Right.
When we sell Evergreen//One, we're only providing SLAs to the customer. Where to put the storage, how to manage the storage, exactly what storage it is, is completely at our discretion, and the customer gets to use storage, but again, they don't manage it at all, and because of that, when a SaaS vendor either has their own facility or they use Amazon or AWS, obviously one way or another, directly or indirectly, they're paying for power, space, and cooling.
Right.
When we place our equipment in the customer's data center, we will pay them for power, space, and cooling 'cause they're just hosting the service on our behalf, even if it's for their own use.
If we take a step back to and think about the key elements that go into our subscription revenue, right? So you've got Evergreen//One, which is a complete as a service. We own the infrastructure. We provide the service all SLA based. You've got Cloud Block Store and AVS that Charlie talked about. Again, that's all a software offering as a service, and that's going through our subscription services as well. You also mentioned we attach Evergreen to a traditional sale. So we have an Evergreen//Forever, Evergreen//Foundation, and that's a pure subscription, but all the subscription revenue is coming through as well associated with our install base and new traditional sales where we have the Evergreen attached. In addition to that, we've got Portworx, which is another subscription for us as well that's coming through there.
So those are really what makes up the majority of our subscription services. We do have a hybrid model of subscription as well, which we call Evergreen//Flex, where customer would purchase the hardware at really white label costs, and then have a subscription for the underlying service as well. So as you can see, we have a lot of flexibility in terms of what we drive through our business models. And then that's carried to the hyperscale design win, which is really a licensing model, by design. Right. So why don't we turn to that and we'll come back. You announced a hyperscale design win last quarter with one of the top four. Can you talk to the importance of this win and the likelihood of additional hyperscale design wins moving forward? And frankly, why did you win?
Yeah, well, the importance of this win is that the first one's always the hardest.
Right.
And so, you know, it's taken many years for us to get to this point with that hyperscaler. It's also the hyperscalers, especially in terms of their infrastructure. They talk with each other quite a lot, you know, on advanced designs and where they're going. We certainly imagined it would and it in fact has accelerated. The win has accelerated our conversations with the other hyperscalers. Now there's nothing to announce yet. You know, certainly, each one will go through their own diligence, but the conversations around that diligence have accelerated since we won the first one. It's important because, you know, it's been our view since our founding. That's now 15 years ago that flash would eventually replace all disk, and that we firmly believe that.
The hyperscalers buy on the order of six or seven hundred exabytes of storage every year. They're 70% of hard disk storage purchases in the world. It's a big market.
Right.
It's, there's a lot of money there. So, you know, it's important from that respect. It's important for the hyperscalers for somewhat different reasons. One is the performance of flash is much greater than that of hard disk. So it promises a performance increase, but perhaps even more interesting is that the average data center uses about 25% of their power and even a greater amount of the space in the data center for storage. We're coming in at about one tenth the space, power, and cooling. So if you were to take all of the power used by data centers and you can now take 20% of that power and use it for other things, it's a huge power source.
The other thing is that when a data center runs out of power, when it's using the power, the cost for creating more power is extraordinarily high, because, you know, it generally in the world today, there's not a lot of excess power in region. So you might have to build new generators, which is very expensive. Even if you didn't have to build new generators, you have to build new distribution, which is very expensive. And then every data center has backup and they have to have cooling. So it's a lot cheaper to just recover power that you already have, and it's much faster to recover power you already have than it is to go out and get new power for expansion of data centers or for AI.
Yeah. It's a big issue. We've written a lot about, done some work with our power teams and the rest at Bernstein on the whole issue of power. And as you move into AI, whether you're doing, everyone focuses on training, but long term, the big opportunity is inferencing. You know, you're gonna have these big data centers that you're gonna have to retrofit with a lot more, not all GPUs, but a portion with GPUs. Where's that power gonna come from?
Right.
There's not enough power capability in the world. Are there other solutions that can do this?
SSDs, even we're more efficient and we have better price performance than SSDs, but SSDs also reduce power, not as much as by we do. But then SSDs can't meet the price performance requirements to be able to go out and attack the disk layer, whereas we can. Even though it's not, the requirements of power for storage specifically for AI are not that high.
Right.
Because AI doesn't actually use a lot of storage.
Right.
It's the power and scale of traditional storage for traditional compute.
Right.
That offers the opportunity.
Right. But if you're going to an inferencing world at some point.
When you want power from somewhere.
You're grounding, but you're also grounding the query in an enterprise.
Correct.
You know, if SAP adds AI capabilities, and that data that they're gonna use to answer the question is coming from the traditional storage.
Exactly correct. Yeah.
It's all very much gonna be co-located in some way, shape, or form and freeing up the power.
That's right.
That comes. And you can do this at a price savings also.
At a, well, right now we're at a similar TCO, total cost of ownership, as hard disk. Now that total cost of ownership is realized over a period of time.
Right.
Yeah.
Right. But that was the first big gate we had to clear.
Yes.
Was the TCO gate.
Right.
So when you think about the design, when we announced—what was that, about a year before that?—we had cleared the gate with them in terms of a detailed TCO analysis, which, obviously, power was a significant component of that.
Right.
No, that makes sense. Power is a big portion of it. Price of power isn't going down.
No.
Whereas the price of everything hardware-wise is going down.
Right.
Very over time. Okay. So what's the next steps on a design, from a design win?
We got the design win. Now, so the way that hyperscalers build data centers is the data center is effectively their product, right? So they design every aspect of the data center, everything from the power, how they power it, to the next generation of the operating system design, to the next generation of their management design, down to every component that sits in the data center, which is now a new component, and even the cabling to tie it all together, right? So that's a multi-year product design that they go through. Now they have to know what components they're gonna use. So that's sort of at the, let's say, somewhere in the 25%-50% of the way through that, their design cycle, they have to finish the software design and do all the testing.
Then, you know, there's multiple layers of testing as they go through it. The design win, let's figure if it's a two-year design cycle, about halfway through the design cycle, that design cycle will go through its completion this year. The big, then you start deployment, you know, data center by data center, certainly all the new ones. The way they handle existing data centers is they wait to the, what they feel is the end of life of that, not the physical, but the technical part of the data center. Then they just scrape the floor and put in the new design. They completely scrape the floor. Everything goes, and.
So even if they have different, you know, useful life in theory, they're gonna take out everything that's in there and put it in a whole new environment 'cause it's better than coming back and retroing one piece of it.
They want standardization. They want every data center of a particular generation to be exactly the same. They don't wanna have to worry about different components and different software in different data centers.
In the hyperscaler opportunity that we are talking about, it's not gonna be a hardware sale. Likely they gonna buy the hardware?
They will effectively, they or their agents will buy the hardware, and will provide the software for our component of this.
Right.
and it'll be a licensing.
Very nice.
Structure.
That's a very nice opportunity.
Yeah.
Maybe Charlie, you wanna talk a little about the larger ecosystem, you know, with our NAND suppliers, the supply chain?
Yeah. It's complex 'cause we're managing our portion of it. So even though they'll buy the hardware, we're preparing, you know, both the NAND manufacturers for it. We're preparing the contract manufacturers that'll be providing the hardware. All the testing is gonna be done effectively by us, or rather we'll build the tests. The hardware manufacturers will do the testing, but we'll build the tests, we'll build the standards for it. So we actually have to prepare the entire supply chain, you know, for the storage, even if they're gonna be doing the buying.
Right. That's a very large opportunity. Does this mean what they'll end up, you know, the scale of one of these data centers from a storage perspective, is that larger than you have in your traditional client base?
Well, to give you a sense, you know, on a, probably our most recent year, you know, we will sell somewhere between four and six exabytes of storage.
Right.
You know, next year we would expect to sell 10 plus. You know, to be clear, if calendar 2026, we would expect to sell over 10 to this one hyperscaler and scaling up from there, you know, 'cause that's the beginning of the ramp, if you will.
Is 10 EB one data center, multiple data centers? I guess it depends.
I would say 10 EB, frankly, is going to be, you know, 10 EB will be, not the entire part of several data centers in our sale, but I would say generally, most of their data centers will have more than 10 EB.
But that is large, 10 EB is larger than almost any company in the world has.
Oh yeah.
At least in a data center, if not global.
Oh no, no company, there's no company in the world that has 10. I don't think there's any enterprise company that has 10 EB. I, I just don't believe so.
I guess you set the new bar for the maximum amount of storage.
Yes, exactly.
I see. Yes, but the NAND companies must be very interested in conversations with you.
Oh yes. Yeah. Those conversations have accelerated as well.
Right. You're suddenly much more interesting to them.
Yeah.
How does this all play in with the AI world everyone's trying to create? We are in that. Can you add advantage?
First of all, stay tuned for GTC next month. We've sent out that hint. We expect to have, you know, one in particular, but several announcements at GTC. You know, to some extent, these are two phenomena walking in parallel. You know, the hype or the business and the great fortunes of companies that have been made in and around GPUs and equipment that goes along with big GPU environments such as networking. I think the hype has been overplayed. And I was indicating this for a over a year now, that AI actually, at least as far as LLMs are concerned, generally doesn't require a lot of storage.
Now, it may still be, you know, $1 billion, $1.5 billion, but in a $50 billion market, it's not a lot, it's not, you know, it's still relatively modest, right? And so, now on the flip side, as you pointed out earlier, eventually inference and RAG is gonna touch upon all storage. We actually think that the opportunity to upgrade a traditional storage for use in AI, of which Fusion is a big part by networking it and making it available, that's what we believe our biggest opportunity is. That being said, I view AI as the F1 race of the tech world. And, you know, having, you know, winning cars in an F1 race and your benefit to, you know, selling street cars.
Right.
So, we wanted to have an F1. We wanted to have a car in the race. Stay tuned for GTC.
You already have, as I noticed on your site, that you've done work with NVIDIA and their systems. There's obviously some work you've done to make sure this'll fit in nicely.
Interestingly enough, we've been working with NVIDIA for five or six years. We've had what was called an AIRI, an AI ready infrastructure product with them. We were the first ones to sign up and have that available with them five or six years ago. Honestly, we in a sense, I would admit, missed the LLM, the GenAI, environment because we didn't and we haven't been in what's called the HPC or high performance computing environment.
Right.
The high performance and that has really been the base of the strength in the LLM environment, so you know, that's about to change.
Excellent. So maybe I'll, since Kevan's had a chance to sit here and relax during the conversation, maybe I'll throw some questions at you on the financial side. For a hardware company, if you're defined as a hardware company, and Paul has been arguing with me that you're not a hardware company, you're a software company, you know, 60% gross margins are unusual. Is that because of the strength of the software company and, you know, any color on how high the software gross margins are?
Yeah. Look, I think it really comes down to the two things. It's the value of Purity.
Right.
As Charles pointed out, the majority of our engineers are focused on development around our Purity software. The value realization of Purity is significant, and you see that coming through across our gross margins. The other is service offerings. We’ve talked about all of the service offerings that are attached either to a traditional sale or we’re providing as a service with Evergreen//One or as Pure Software with Cloud Block Store or AVS. All combined, yeah, I mean, I think we really are more reflective of a software company. Now, because we’re selling across the data storage platform, we’re gonna go compete aggressively against disk-based solutions. That’s where you’ll see a little bit more pressure on product gross margins. But look, we’re gonna be aggressive in that space. We’re gonna look to win.
But long term, when I look at it, I still think we've got, you know, some expansion opportunity on our gross margins, both on subscription services, as we continue to grow our as a service offerings. But then the licensing model that we've talked about, specific to the hyperscale design win, that should be providing expansionary product gross margins for us as well.
On the operating margin side, the biggest OpEx cost is R&D. How do you see that over the long term? Obviously, it's critical. Software company, you're competing with very high tech, especially if you're developing things that are gonna run on the hyperscalers. And the hyperscalers, many of them like to build all their own stuff. So obviously you need to invest. But how do you think of R&D and OpEx in general over a longer horizon?
I like where we're kind of sitting, frankly, on the R&D investment, because we are, I mean, the amount of innovation we're putting out has been tremendous. And it, it's far beyond just things that we've talked about, whether that's Fusion, whether that's the hyperscale design win. We've got, you know, announcement coming out here shortly as well that Charles alluded to. There are some great investment that we're doing on the technology front. Where I think we'll get some benefits significantly is if you think about the hyperscale design win, you don't have a lot of sales and marketing costs associated with that. So you get some significant upside with that, with your services, your Evergreen//One services and your subscription offerings as those are renewing again, your sales and marketing costs associated with that are very low.
And so that's why I think the bigger opportunity on the OpEx side is. Now we're still taking advantage of the global talent across the engineering front. We've got significant resource pools of engineers in Bangalore and Prague. So we'll get some advantages there. But in terms of our focus and commitment to R&D innovation, that's gonna be pretty consistent. I don't see that changing dramatically.
Yeah. No, I think that's right. I, and I'll say one other thing. That's, that's what allows us to have the highest margins in the industry. And margins actually are created by, at the end of the day, three things in my view, right? One is value that you provide to the customer well and above the rest of the, rest of the industry, right? And I think that, that is seen in our Evergreen promise that we provide customers where their products never get old, non-disruptive upgrades.
Right.
Single operating environment across all the different forms of storage that they might have. So, value that you provide to the customer. Two is anything that you have in your business model that allows you lower costs than the rest of the industry. Again, that's provided by our software. Our Purity software allows us, because we have the DirectFlash capability to get lower costs for the same price performance than our competitors.
Right.
Right? We're able to make better use of the flash than our competitors are. And the third thing that drives gross margins is market share. Now we're actually quite low in market share at, you know, compared on an industry basis. We're quite high in market share when you count flash only, that is flash only. But when you count hard disk, we're rather low. As we gain more market share and we're the only ones in the industry gaining market share, as we gain more market share, that should also inure to our margins as we go forward.
Perfect. Let me just check. If anyone wants to ask a question, raise a hand quickly. No? Okay. Yes.
Yeah.
One second. I'm just gonna bring the mic over so they can be heard on the telecast.
For the customers who are looking for an off ramp of their relationship with VMware.
Yep.
Is that VMware Azure offering a viable off ramp for them? And are there tools and alternatives for them when they get there? If that's the path they're trying to go?
Sure.
So the fastest off ramp from a VMware license for a customer is to go to, in my opinion, I'll just state my opinion, is AVS today, because the move can be a maybe not huge lift, but there's no real, there's not a lot of software the customer has to change or build into their application to make that move. They're going from VMware to VMware, maybe slightly different change in management, but it's pretty straightforward. The other alternatives require moving to a different virtualization environment, which, compared to VMware, are just much less mature. And so it usually will require a lot of investment, or loss of functionality for them to be able to do that. On a longer term basis, where we're also investing quite heavily, you know, our belief is that customers are gonna move to containers and Kubernetes.
I did mention in this week's earnings call a number of customers that are moving to what's called Kubernetes virtualization or KubeVirt for short, KubeVirt, which is virtualization. This is. It's open source, but it's also part of Red Hat in their OpenShift environment. And it allows them to move to a virtualization environment that's also managed by Kubernetes. So, you know, what a customer, you know, when they look at the future and say, well, I have a management system that is. Doesn't matter who it is, but if it's vCenter, you know, which is VMware or any other virtualization environment, and then simultaneously I'm moving to containers, which is under Kubernetes, I have two different environments that I'm managing. Why not just bring it all under Kubernetes?
As you know, we made an investment a number of years ago in Portworx. If customers are moving to Kubernetes virtualization, Portworx is almost, I'd almost state it's required. There's really nothing else out there to manage the storage for them, that is, easy to use and easy to manage. We're a simple choice in that environment. So we're also deeply embedded there.
So VMware keeps coming up in the conversation. How do you think about your relationship and why is VMware such a sweet spot for a client base? Is it 'cause it's the largest companies in the world?
No, it's because we spoke before, when we made the comparison to vCenter, that companies embrace virtualization so that they can make better use of their compute resources.
Right.
And so there's a very large fraction of enterprise applications that operate on on VMware. VMware won that, won that battle. The reason why it's so much in the news is prices went up by about a factor of four.
Right.
Now they're the, look, we actually work very well with VMware. I have nothing, you know, I have nothing against VMware, but it's created a lot of interest by customers for looking for alternatives. So we work very well with VMware, but we're increasingly working with other options for our customers as well.
Perfect. Last question. There's something you wanted to say to an investor who doesn't know the company. What would you wanna tell them about Pure Storage?
I think what I'd want to tell them is that we are the one company that treats data storage as high technology rather than a commodity, and that focus and that point of view has really driven our success, you know, since our founding, and I think we'll continue to drive our success as we go forward.
Perfect. Thank you very much for your time. I appreciate you coming out.
Thank you.
Thanks.
Pleasure, Mark.
Thank you.