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

Welcome to the Pure Storage Product & Technology-Focused Meeting for Financial Analysts. Let me remind you that we will be making forward-looking statements today that are subject to assumptions, risks and uncertainties. Actual results could differ materially from those anticipated due to a number of factors, including those referenced in the detailed disclaimer at the beginning of our presentation slide deck and in our public filings with the SEC, which we encourage you to review. The presentation slides discussed today will be available on our Investor Relations website at investor.purestorage.com.

Hello, hello, and welcome to our product and technology-focused Financial Analyst Meeting, which is being held in conjunction with Accelerate. We're so happy to see so many people here in person as well as those attending virtually. We really appreciate your interest in Pure Storage and your support, frankly, many of you that are here. So on to the agenda for today. The plan is to start with 1 hour of presentation material, followed by about an hour of Q&A, for a total of 2 hours. Charlie will start us off with Pure's platform strategy and the Enterprise Data Cloud, key enablers for us moving forward. Rob will then talk about Pure's fundamental advantages, which, frankly, are still not super-well understood on the street. Then Prakash will talk about our customer journey to Pure's Cloud Operating Model, which is also an area that's not super-clearly understood, and we hope to increase that understanding here today. And then Bill will wrap us up with Pure's Hyperscaler business. I hear there's just a little bit of interest in that. So we thought we'd cover that. So then as Charlie comes up here to start us off, I'm going to remind you all, as I've been known to do, that at Pure Storage, it's all about one thing, the software, the software. You're going to hear that all day today. Charlie?

Thanks, Paul. And thank you all. Good morning. Thank you all for coming. I appreciate you spending time with us this morning. So yes, I'm going to be speaking to you about our overall strategy and how it all fits together. And of course, we'll be speaking about this. This is substantially the same presentation that I'll be giving this afternoon to our New York customers. There's a lot of similarities to the presentation I gave just 3 months ago at our Las Vegas Accelerate conference. But I'm going to start off by just saying that we continue to outperform the market. Our expectation is that we will continue to do so. We had a very strong quarter this past quarter, as you well know. But something that I think probably is -- well, the milestone that I saw in this last quarter was something that if you dig down into the financial metrics of us and our competitors is probably something that might be a bit surprising. We are now spending as much on R&D as any one of our competitors regardless of size, if not more. We're spending more than most of them, and we're now at the same level as many of them. And so -- and of course, since we spend it just on one software environment to cover all of the different areas of storage, we can be very efficient and have a lot of leverage in that investment unlike some of our competitors. So we think this is a major milestone as we continue to scale in this area. There's really no opportunity in the data storage market that we can't go after at this point. If we look at the metrics of the company, of course, 13% year-over-year growth, strongly profitable this last quarter. In addition to that, now, as you know, we're in the 40s and probably will be through the year in terms of the percentage revenue in our subscriptions. And as I mentioned, not only are we continuing to be the company that leads in terms of the percentage of our revenue we spend on R&D, but now also in terms of the total dollars. So the question might be, well, what is spending on R&D give you? Well, one of the things that gets you consistently and now 11 years in a row, top of the charts in terms of the Gartner Magic Quadrant. The other thing is very, very high customer satisfaction. As you know, we've always led in terms of Net Promoter Score, and I don't want to stand in front of the number there. We've always led, and we continue to lead. It's consistently above 80. And it's not just based on the customer service that we provide, but frankly, it's on the quality of the products and the quality of the customer experience with our company and with our products. So now I'm going to switch gears a little bit and start talking about what is it really about our products and the direction that we're going in that we think is going to impress our customers the most. And I might start from a surprising point in time because I want to address the elephant in the room. So what is the elephant in the room? Well, it's AI, okay? And this may -- so bear with me. The thing about AI that we think is really going to be driving customers as we go forward is the fact that it changes the relationship between software and data. And before AI, you may remember a famous individual once said that software was eating the world. right? And software has really been the driver of the fortunes of tech companies and the driver, frankly, of productivity in business overall. But what's happening now is that data is starting to drive investment and certainly starting to drive the performance advantages of some companies over another. So maybe data is starting to eat the world and maybe even data, and this has been discussed by many of you, is eating software, putting software as a dominant element somewhat at risk. So it's really changing the nature of software and data. And of course, data is where we play. So how do we help in this environment? Well, let's step back for a second. Let's talk about enterprise data. What do I mean by enterprise data? I don't mean sort of the files that you all share with your colleagues in your business. I'm talking about the data that business runs on. So the databases, the unstructured data that corporations collect to be able to get a better understanding of their sales force, of their customers and so forth. So let's talk about how they manage that. They manage that, of course, in data centers. And how are these data centers constructed? Well, let's take a typical application, which might be a database application. The database application starts with the software and the database and the compute that the database is going to be running on. Well, then what they do, what an IT organization will do, is add storage, add networking, and it becomes a stack. And sometimes they call it a full stack. And the full stack is made up, as I said, of compute, switches and storage. And then there's another application. It could be e-mail, it could be just a customer database, it could be an understanding -- it could be their marketing database, and they set that up. Or it could be their virtualization environment. And each one of these is set up. Now a more sophisticated customer will have a standard for their compute. They may have virtualization. It's all virtualization standard so they can move their application around between different computers. And of course, they're usually standardized on one network environment. But it turns out that they choose their storage bespoke to the application, high performance, low cost block or file. And each one of these can be quite different, as you can see in the picture. So what does that really mean? What is that -- what is the effect of having this kind of architecture, vertical stacks and bespoke storage? Well, what it means is, first of all, the storage is provisioned individually and manually. It doesn't operate the same across all of these different stacks. What it means is that they can't share capacity from one array to another. Each array is unique to its environment. If it runs out of capacity, even if there's spare capacity somewhere else, it doesn't matter. They have to add capacity to that one array. It means that new services have to be configured manually on every array. It means that any kind of policy that the company sets for their data as a whole has to be done individually array by array. So your data governance standards are also implemented manually. Now let's think about it, not from the storage angle. Let's think about it from the data angle, how customers manage their data environment. So again, within the same architecture, data is now captive to the application stack. You have to go through the application to get access to the data. The data is not available otherwise. It also means that there are poor records. People can easily copy and paste data in different areas, but the records of that, again, if they're kept at all, it's manually kept. So think of it from a governance standpoint, right? It creates a mess, hard to follow. Raw data is inaccessible to applications such as AI, has to be copied to an AI-specific data storage array somewhere. And then finally, if you think about data governance, because data governance is done array by array and manually, it's going to be inconsistent because this is spread out all over a large corporation, different users, different administrators. So it's very inconsistent data management. Now let's contrast it. That's not the way that clouds operate. It's not the way that clouds are built. How are clouds built? Very different philosophy overall. They're built largely horizontally. They want everything to be the same across their cloud, not only in one data center, but across data centers. So it generally starts with compute, one compute layer. The compute layer is treated as a general purpose compute layer. That's all tied together, of course, with the network. Now the next thing to think about is what they do with data storage. The data storage is the same horizontally across the entire organization -- across the entire cloud. The only difference is they may have different price performance layer. They have a high-performance tier, they have a low-performance tier, usually a tier or 2 in the middle. And they'll even have, in many cases, an archive tier, which might be tape. And so what they're able to do here is any application can access any portion of the storage. Of course, having -- you need to have the right authorization to be able to do so, but there's no physical limitation to what storage can be used for what purpose, right? So it's a completely shared environment that is software defined. It's a management system that creates the environment that determines how to -- or what storage can get access, much more flexible architecture. So this is what we feel is allowing an enterprise storage architecture now to be changed. An enterprise storage architecture as it exists today is vertical, right, vertical stacks. A cloud storage architecture is horizontal with everything being the same across the horizontal limit. An enterprise storage architecture is largely manual. It is not orchestrated by software. Cloud storage architecture is fully automated and standardized. Enterprise creates data silos, cloud storage architecture creates accessible data pools. And then finally, enterprise storage architecture is very physical, with physical connections. Cloud storage architecture is virtual. It's all based on software, what gets connected to what. So we believe this sets up for the industry a new vision for data, how data will be handled inside the enterprise. And in particular, as you might be able to guess now, it's a cloud -- we want to bring a cloud operating model to enterprise storage. So how do we do that? Well, we've been working on this for many years. As many of you know, it starts with Purity. Purity has this -- it's a common operating environment for any type of storage with our, very famous now, Evergreen architecture, which means it never gets old. That now -- Purity we've designed -- we started with block, but now provides for block, file and object. In addition to that, it supports all types of data, including everything from the highest performance in AI, all the way down to low cost, now replacing all disk environments. And that we continue with our progress about hitting ever lower price levels for low-performance storage. We also provide all kinds of services, and that's all embedded within the Purity environment, and we continue to add more and more services to that, all managed by -- sorry, let me rephrase this. This is what we mean by a unified data plane. That is one software environment that covers every level of price performance as well as all different types of data. Now it is all managed by one management system, which we call Pure1, and it is able to be acquired either as product or as a service with our Evergreen//One management system. Now the big reveal, if you will, that we announced in Las Vegas, and we are going to continue to focus on is this capability that we call Fusion, which now creates a unified data plane. So we're able to support all kinds of enterprise apps, all kinds of modern apps. So there's really no application environment that our system -- that Purity doesn't support today. But with this addition of what we call an intelligent control plane, instead of our arrays operating as individual arrays, they operate as a cloud of data inside our customers' environment. So all of the arrays communicate with one another -- are able to communicate with one another and are able to operate as a cloud rather than as individual arrays. And this fundamentally changes the way that customers can access their data, use their data and manage their systems. And when we say that the arrays are networked operate as a cloud, we're talking about a global data cloud, not just within an individual data center. So -- and this is what we mean by an enterprise data cloud, which is the architecture that our system now, with Fusion, allows our customers to build inside their enterprise environment. So just to sum this up, we're taking arrays from being manually provisioned to being auto provisioned. We're going from dedicated capacity per application stack to automatic load balancing across an entire global network. We're going from governance being manual to global governance of the way data is managed through software control, true software definition of the way data is managed. We're going from policy setting to -- from being manual to being automatic, which means protection is not applied array by array, but by policy. And then finally, from a captive data -- from data being captive to an application stack to data being globally accessible for things such as analytics, such as AI. Or another way to think about it is the Enterprise Data Cloud allows organizations to make their production -- turn their production data into their data lake. Today, data lakes have to be just a separate pool of storage. Why is that? They already have the data. It's already sitting in a production environment. It's already real time. Why not have that be their data lake. And so it replaces siloed data, if you will, with a data cloud. So you remember, I started this with AI. The opportunity for companies -- one of their biggest challenges is their data is spread out everywhere and siloed. And AI is having companies relook at the way their data architectures exist, and we think this is what sets us up for really a fundamental transformation of architecture inside the enterprise. Now I want to speak about one more thing in this area, and that is -- I want to speak a little bit about our launches. And I'm not going to go -- this is -- you're going to see these are the launches that we're putting out just this year. This is also what our investment in R&D is bringing to the market now. And I think you'll see that we have an extensive set of announcements that we made and products that we're bringing to market that cover everything from unifying even more the data plane so we can cover more use cases, more tiers, if you will, of storage, but also the intelligent control plane that allows customers to more easily manage this -- not only manage our products, but frankly, to manage their data sets overall. So I'm not going to go into this, but I do want to touch upon one thing. Again, it's another area that I know there's a lot of interest in, and that is FlashBlade//EXA. So -- and the reason why I want to touch upon it is because -- and Rob will go into much more detail, but it shows how extensible and how flexible our overall architecture inside the product truly is. So EXA is based on Purity, and it's based on our FlashBlade physical architecture. FlashBlade is our scale-out architecture. And up until now, we've used FlashBlade as both a storage target as well as the product that is able to handle the metadata that allows applications to get access to that storage. What we've done is we've leveraged the fact that FlashBlade is actually the world's fastest metadata engine. And now by attaching additional data nodes to FlashBlade, rather than using FlashBlade as the data store, we now just use it as the metadata engine. And not only can we add sort of unlimited amounts of additional data at high speed to it, but we can replicate the -- and we can expand the metadata engine, and we could just keep expanding it and expanding it. And what's really magic about this is -- and this is something that every other vendor would love to have is it scales linearly. Every time we add a new FlashBlade metadata EXA engine, it just continues to scale the overall performance, all the way up to right now, internal testing, 10 terabytes per second of read, 5 terabytes per second of write, roughly 5x faster than anything that's out there today. So we're able to cover the entire range of AI needs from hyperscalers, which is going to be based on direct flash, not on EXA. And that will be covered a little bit by Bill today. FlashBlade//EXA, which covers really the GPU clouds, Rob will be covering that. And then regular FlashBlade, which is -- perfectly, is a great product and covers really the entire needs of most enterprises in the AI environment. So we're very well covered. And all of these -- well, that is EXA, S, as is true of all of Pure's products, are covered by Evergreen and allow for the ability to be able to do upgrades in with -- allowing customers to save the investment that they've placed in us. So summing up, the Enterprise Data Cloud now is really a new architecture for our customers to think about how they virtualize their data environment. It does a number of things. It creates a virtual cloud of data, but it creates a very consistent environment for all of their different workloads. It allows for data to be governed by policy and software rather than manually through fingers on keyboards. We build in cyber resilience. And, again, that can be done by policy rather than fingers on keyboards. And it is completely software-defined. So this is a really fundamental architectural shift in enterprise. It's one of the -- it's both an opportunity for us as well as a challenge because changing architectures in organizations that have been doing the same thing, the same way for decades is not for the faint of heart, but we think we're making great progress here. So with that, I would like to turn the stage over to Rob Lee.

Thanks, Charlie. All right. Good morning, everyone. Great to see everybody here in person. As Paul mentioned, I'm going to spend my time today talking about and going through Pure's fundamental technology advantages. And what I'm not going to do is actually go through product by product or unpack the portfolio. Instead, what I want to focus the discussion on is really looking at the various pieces of differentiated technology, IP, intellectual property, we've created across the entire storage stack. We'll then transition into looking at, hey, how do we go take those core blocks of IP, leverage them in different ways, package them in different ways to go meet the needs of different markets and pursue different market opportunities. And then we'll do a double-click into one particular area, which Charlie mentioned, which is the performance and scalability of our metadata, which we see really driving the success of our AI platforms. If you're out there and you've been following us and you're perhaps still wondering, hey, what's that one critical technology advantage that Pure has that sets you apart from the others? They're in storage, they're an outlier. They do Flash, others do Flash. There's something different up here. if you're out there still wondering what that is, if I've done my job right at the end of this, a, you'll have the answer to that question. And b, you'll walk away realizing it's not one thing. There are 4 or 5 different things that really come together that we're able to now put together, combine in different ways to go after the various market opportunities that we're pursuing as a company. All right. So we're going to go deep. Before I do, I want to set some context and really just step back and look at, hey, what's in a storage system, a storage system, any storage system out there, one that we might build, one that our competitors might build, one that hyperscalers might build for themselves? And at the most basic level, any storage system has to solve a lot of the same needs. You need physical media. You need somewhere -- something to write your data down on, whether it's hard disk drives, whether it's flash. You're going to need some software to make that physical media work, whether that's OS drivers, whether it's firmware, whether it's purpose-built software. You're going to need some system hardware, right? You got to -- you need chassis, you need sheet metal, you need power supplies, controllers to put all this stuff in and to run your software on. Once you've assembled all that, you probably want some way to organize, store, retrieve your data. You want all the protocols, performance levels. You probably would like the system to be resilient to faults. You'd probably like some protection, snapshots, replication, that sort of thing. And at some level, you're going to need a way to manage, monitor and administer this thing, right? At a 60,000-foot level view, all storage systems have a very similar set of needs. Now the thing is as you dive down to deeper altitudes, as you come down to 30,000-foot level, as you come down to ground level, even though the needs are very similar across the board, hopefully, what you'll see through this discussion is the way that those needs are addressed couldn't be more different in our industry. All right. So let's walk through each of these layers and take a look at how the rest of the industry has tried to approach these and what Pure has done differently, right? And we'll start with the physical media and really the software approach that's been taken to control the physical media on which storage is placed. And this is perhaps the most readily apparent difference that we have relative to the competitive set. Everybody else out there is reliant on SSDs to consume flash. So if we take a step back, what is an SSD, right? An SSD is a technology coping mechanism to make the physics of flash work in a world that is built with software that's designed for hard disk drives. Flash behaves very differently. All the software in the world was built, prior to us, with hard disk drives in mind. So the industry came up with the SSD as a coping mechanism, as a translation mechanism to make that software work for flash. Now what does that mean? It means that the SSD has a tremendous job that it has to perform. It has to do a ton of work internally to translate flash into a world where it behaves and looks like a hard disk drive. It has to manage the flash at a physical level. It has to perform a lot of remapping, a lot of complex translation of instructions. It has to do background work like garbage collection. It's really complex software. And SSD is basically like a little computer, right? It's got an ASIC in it. It's got a processor. It's got DRAM. It's got complex software. They just call it firmware. It's basically a computer. The challenge with that is really twofold. One is, as with anything, if you add translation layers, if you add a man in the middle, so to speak, you're fundamentally adding inefficiency. You're adding extra work to be done. That extra work comes at a cost. That cost can be measured in components. It can be measured in performance. It can be measured in reliability. It can be measured in limitations. The second challenge with this architecture is that as flash advances, as the industry builds denser and denser flash, it becomes harder to work with. As you try to put more flash into the drive, you're complicating that task even further. The SSD architecture has reached its limits. So what has Pure done? Well, we had the benefit of not being saddled with retrofit architectures. We did not have software built for hard disk drives. So we started our software top to bottom, built specifically for flash. What this has allowed us to do is to build a media software layer -- layer of software to manage the physical flash that doesn't add all this translation work, right? We simply avoid this duplicative work by designing a software layer that allows us to expose the best properties of flash directly to the software that's running above. This also has another side effect, which Bill will talk to in a bit in his discussion, which is that as we look at different ways of packaging the software, we're now taking on the task of making all the different types of flash and NAND out there in the world behave and look the same. For everybody else that's relying on SSDs, well, the SSD manufacturers have put that work of qualification, put that work of the different behaviors of firmware on the consumer, whether that's the enterprise OEM vendor or whether it's the hyperscaler. All right. So let's move up the stack a little bit. Let's take a look at the system hardware. These are the chassis, the enclosures, the power supplies. And this is an area where our competitors, frankly, treat this as an afterthought, right? You can look at this, you can tell this by -- based on what they ship, right? It's largely third-party OEM hardware, it's repurposed servers. One of our competitors, at their top-of-line product, if you look at their manual, will say things like, when you install a storage array, you have to leave extra rack space above and below this thing to allow for sufficient airflow to cool the thing so it doesn't overheat and to allow us to go and service it. That's how much care and thought they have put into the system hardware. Obviously, that results in complexity, bloat, operating costs. We've taken a different approach, right? In our world, our system hardware is integral to what we do, right? It's integral to the Evergreen promise. Evergreen at its most basic level is a promise that once you're on Pure technology, you're on a path to non-obsolescence, right? You have the ability to non-disruptively evolve, upgrade scale to future needs, whether they're -- it's future technology, whether it's more capacity, more performance. Well, the only way we can go and deliver that is a hardware platform that has the capability to be forward and backwards compatible that's designed for decades-long lifetime. Our system hardware is core and integral to what we do. And also by owning that system and really investing heavily in it, it allows us to create significant advantages that come through in reliability, performance and simplicity. All right. Let's move up the stack one click. Take a look at the main layer of storage software, right? And this is where I would say some of the biggest differences really start to come out. If I look at the competitive set, this is where the retrofit architectures rear their ugly head yet again. Most of our competitors are building on retrofit software stacks designed in an era of hard drives. They were designed in an era where hard drives are basically the same speed they are today when flash has gotten 100 to 1,000x faster. They were designed in an era where networking was 100x slower. They were designed in an era where CPUs and processing was 10x slower. As you might imagine, right, the choices in how you build software, how you architect, how you design software in that world are very different than how you would design in a world of hardware resources that we have today. I'll give you a very simple example. If you're designing software to place storage on a hard disk drive, it's a physical moving platter. If you want to optimize for that, you want to lay out your data sequentially because it's a sequential mechanism, you want to lay out your data sequentially, so as the platter moves around, you very easily read it. Well, in a world of flash where it's highly parallel, it's semiconductor driven, if I want to go and optimize that, I want to place my data on multiple chips, and I want to go and access those entirely in parallel. These are very orthogonal decisions. This is one of many examples of orthogonal design decisions that are central to software stacks that are designed either for the hard disk era or for the flash era. As you might imagine, these are incredibly hard to unwind. And one of the other offshoots of the software that was designed in the era of hard drives is that because performance was so difficult to achieve, the competitive set generally had to build different software stacks for different purposes. Charlie mentioned this before in terms of how different storage arrays have historically been provisioned for different applications. Well, the competitive set generally has had to build separate products and software for each different need to meet that performance. Those of you who cover software will realize that it's incredibly hard once you have that fragmentation to bring it all back together. So what has Pure done? Well, again, we have not been saddled with. We did not grow up in an era of retrofit software. We were able to design for flash since day 1. We were able to build from the ground up for the speed and scale that flash was capable of. And we made those hard investments to really take advantage of not just the performance that was available to us back in 2009, but where we knew flash was going. More importantly, we've very, very intentionally invested in keeping a unified and single software base and single hardware technology. This now brings us back into -- and is a key enabler for Evergreen as well, right? If you think about the core Evergreen promise, which is non-obsolescence is the ability to non-disruptively grow, scale, evolve in different ways. Well, how do you do that in a world where you have 6 different software stacks to meet different performance levels? It's really, really difficult, right? One of the things that is critical to enabling Evergreen, besides the system hardware, is the flexibility and the dynamic range that our software allows us -- our single software base allows us to meet the entire set of needs from AI to archive. All right. Click forward. Thank you. All right. So let's take a look at management and monitoring. Again, this is an area where if I look at the competitors, generally thought of as an afterthought. It's a bolt-on. Frankly, most of our competitors' management tools are more focused on all the tuning knobs and configurations that are required to make those products perform. If you look at how we treat management and monitoring, it's central to not just how we deliver a customer experience, but also to how we build a full as-a-service model. Let me step back and unpack that. If you look at all of our arrays since day 1, right, the Phone Home data to Pure1, the intelligence that we can gather from the arrays, we manage -- if your home thermostats, if your AV system at home is able to be cloud managed and that service is able to be improved over the air, why can we not do that in the enterprise? That's been our focus. That has allowed us to have the NPS score, the customer satisfaction, the proactive support experience that we've been known for. That also allows us to better understand the arrays in time and transition that into a full as-a-service offering. You heard Charlie talk about the intelligent automation. Prakash is going to go into this quite a bit more. All of this stems from our investment in the management and monitoring part of the stack. All right. So we've unpacked. Let's move forward. Thank you. All right. So we've walked through each of the layers of the stack. What I want to transition us to now is looking at how do we put these pieces together, how do we leverage the investments we've made in this differentiated technology at the different levels to meet the needs of different markets. And the 4 markets that we want to focus on are: number one, the core enterprise; I want to take a look at what we're doing on the cloud and cloud-native workloads; the scale AI, which is kind of more the neoclouds, the GPU clouds; and then certainly take a look at what we're doing and how these pieces fit together for the hyperscalers. So the core enterprise workloads, and this is all the workloads you'd find in an enterprise data center. This has been the North Star. This is where all the pieces come together, and it makes sense, right? Our enterprise clients aren't coming to us with a partial solution. They want us to provide the whole kit and caboodle, right? This is where by tightly integrating all of the layers of the stack, the differentiation we create at each of these layers, we're able to create a very compelling value proposition across all these layers. And you see this coming out in our flagship products like FlashArray and FlashBlade. If we shift a little bit and talk -- and take a look at what we're doing for customers who perhaps are running on top of public clouds or running cloud-native workloads on top of third-party infrastructure, this is where, in that case, the public cloud -- the CSP might be providing the physical infrastructure. They're providing the physical media, they're providing the system hardware. But customers are seeing value in having the storage services, the same enterprise capabilities, the data protection, the reliability, the sameness of management across on-prem, hybrid and cloud. And you see this model coming out and these areas of IP coming together in products such as CloudBlockStore or Portworx. Shifting gears yet again. If we look at the needs of the scale AI community, the neoclouds, the GPU clouds, the largest foundation model builders, these are folks that have their own higher-level software management, monitoring. They have their own higher-level data flows. In a lot of cases, they're specifying their own physical hardware, ultimately kind of optimized for high performance. What they value, what they get from us is the storage software. The storage software that gives them that scalability that unlocks the performance of the hardware underneath us but also allows us to fit nicely into their software ecosystem, their management, monitoring systems. All right. And then last but not least, hyperscalers. I'm not going to go super deep here because Bill is going to come up in a few minutes and unpack this further. But this is where -- if we step back and we look at the hyperscalers, they've got their own management, monitoring. They think about storage reliability and storage services at data center -- across data center scale. They design their own system hardware, but they've got a challenge, right? They've got a need to move to flash. They're largely running on hard disk drives today. The SSD architecture, I've already explained, has significant limitations. We have the capability. We have the means in our DirectFlash software to dramatically improve their path. And the means are really driven by the software, right? There's nothing -- our DirectFlash Modules, there's nothing -- there's no secret sauce in there. It's -- the enabler there isn't how we solder chips to a board. The enabler there is what we have put into the software to make that work. And that's really key to the integration that's driving the Hyperscaler business, which, again, Bill will unpack further in a few minutes. All right. So if we step back, hopefully, this helps illustrate how -- starting with the core enterprise business where we focus on building a tight integration between all these layers, coming together to really meet the full needs of an enterprise across their entire set of applications has led us down the path of building core differentiated IP blocks. We can then go take those IP blocks and leverage them in different ways to meet the needs of other markets, whether that's folks that are running on cloud, whether it's cloud-native applications, whether it's the neoclouds and scale AI or whether it's the hyperscalers. I mentioned I wanted to do a double-click into one area, please. Thank you -- which is the scalability of the metadata. Charlie mentioned this upfront. I want to unpack this a little bit. We talk about metadata, we talk about data. What does that really mean? So in its most simplest terms, imagine you're trying to load a file. You're trying to load a file and transfer it somewhere. There's the work of looking at the file by name, figuring out what directory it lives in, figuring out, do you have the permission to open it, figuring out, okay, once you've identified that file, where are all the bits of the file located, on flash or on disk. And then there's the work of actually reading and transferring the file. There's the administrative work upfront, the metadata and then there's the actual data transfer work at the back. Well, it turns out both are important, as you might imagine. And it turns out that the metadata, the administrative work, scaling that, making that performance level super high is actually, in many ways, more challenging than scaling the data transfer performance. It's more challenging. And one of the reasons it's more challenging is you've got to figure out a way to organize that metadata, keep track of all those bits and pieces in a way that's highly optimized for parallel access. You've got to keep all that information in a way that makes it easy for you to scale and add performance into the system. And the second challenge you have is once you've organized your metadata in that way, you've got to optimize the coordination. You've got to allow that data to be accessed in parallel without the different processes stepping on each other, without creating wrong results, without creating corruptions, without creating inconsistencies. And so these are fundamental challenges. How have we solved these? Well, it turns out that we have focused in this area since day 1. I mentioned before, we were not saddled with hard disk drive-based software. We focused on the parallelism of flash since day 1. One of the key enablers that has allowed us to do that is how we organize metadata in the system, not just file metadata, but our internal metadata. We've organized it in what's called a key-value store. Frankly, we've borrowed concepts that come from the database community. We've borrowed concepts such as key-value stores, such as distributed transaction engines. If you open up our software and look under the hood, it actually looks a lot more like a high-frequency trading system or a distributed database than a traditional software stack. What this means is that the scalability of the metadata performance we can deliver, all of that administrative work to pair with the data transfer work is extremely high. We see these needs as super acute in technical computing, whether it's chip design, whether it's fluid dynamics or whether it's AI. And this has really been the driver of strength for the entire FlashBlade portfolio since really its inception, whether it's FlashBlade//S for the majority of the enterprise high-performance needs or, as Charlie mentioned, with FlashBlade//EXA, where we're keeping that strong kind of scalable high-performance metadata core and then pairing it with an open hardware architecture to meet the needs of some of the world's largest training and high-performance environments. And as Charlie mentioned upfront, this now completes our portfolio of serving the needs of AI at the entire range of scales from the smallest research groups to what the enterprise is doing, which is largely served with FlashBlade//S, your neoclouds, your tech titans, what we're calling the scale AI community with FlashBlade//EXA or what the hyperscalers are doing, which is really more of a horizontal high-performance design, which we're serving with DirectFlash, which again, Bill will go into greater detail on. So with that, I'm out of time. I'm going to turn the stage over to Prakash, who's going to walk us through how we're extending the management and monitoring plan to deliver on the Enterprise Data Cloud.

Thanks. So as we think about our customers, our goal is to help them move towards this cloud operating model. Charlie unpacked what an intelligent control plane is. But if you think about Salesforce or any SaaS solution on the market, realistically, how they're able to get economies of scale and SaaS margins is by standardizing operations. When you run everything, when you control everything as a SaaS-managed thing, your environment is very standardized. But if you go into a customer data center, it's not. Charlie mentioned it's very siloed. It's very fragmented. And for anyone who's been in a data center with a number of solutions and the number of operating systems that have historically existed, it's a mess. If you want to start moving people to an Enterprise Data Cloud and get SaaS economics, you need to standardize the components, create standardization at scale. And what that allows you to do is provide instantaneous value where you can spin things up at the need of what your requirements are. So our cloud operating model is driven to create the standardization. As I unpack this management and monitoring, historically, people thought about this box-to-box storage management. It's a capability where it's like I need this type of data, I need to provision it, block file systems volumes, et cetera. That paradigm to move to a SaaS operating model requires a paradigm shift. You can't think about these low-level objects and management anymore. You have to think about how do you move to policies and fleets. You need fewer components that are more standard with simplified policies where you can move beyond low-level management to higher-level management. Now this is fundamental in the shift we're driving in our customer and install base. We're the only vendor that's driving this innovation to this policy-centric view. Now that's a step in the direction. But when you move beyond these policy and management objects, you then get into the world of what AI needs, which is what are my workloads, what are my workflows. So managing data sets is the ability to take a look and assign these policies and these management principles to workloads. You want all of your SQL workloads to operate this way. You want all of your Tier 1 applications to operate that way. And all of these policies create standardization in a data center. It's the reason why a lot of hyperscalers who've tried to do go on-premise have failed because they expected standardization in a world where there was none. So you have to take people along this journey. Now once you get there, there's 2 approaches. You could do this. We can provide the capabilities with our management and monitoring for customer teams to deliver customer SLAs. Or secondarily, if you don't want to worry about it, why not trust us as the vendor to deliver vendor-managed SLAs? So when we talk about our Evergreen//One offering, it is a vendor-managed SLA. It's a vendor-managed service with standardized building blocks. You can get whatever storage you need, block file or object, any type. You can get it deployed wherever you need it in your data center. When we run our service, think about it this way, we're paying for the power and rack space we use. So we're just running our service and saying, "Oh, this is our service. It's a managed endpoint. We're running it in your data center. With our CloudBlockStore, you can get it in the cloud. We have partnerships with MSPs where you can get it in the Co-Lo, and it's available in 32 countries around the world." But the standardization, when we talk about that standardized operating model at scale, what do you want to standardize? Charlie mentioned performance levels. Without a single operating system, you couldn't get unified performance and actually deliver this at any reasonable margins. If you want uptime, 0 data loss, like we've had a lot of people think about data center consolidations right now. We all know about the power crunch. We have a site relocation SLA where we can help customers consolidate their data centers, their physical locations because our service can deploy new hardware and move the data seamlessly transparent to the customer as part of the site relocation service. With our Evergreen architecture, we have 0 planned downtime. We have guaranteed watts per terabyte with our energy efficiency guarantees and cyber recovery SLAs in case of ransomware attack, where we guarantee the time it takes to get up and running for customers based on their data sets. So we were unique in approaching this problem set, not from just a financial lens of cash or credit, how do you want to move to recurring revenue, but fundamentally changing the operating principles of how do you standardize data centers with this cloud operating model. Now it's built on the uniqueness that we talked about. You could not do this without Evergreen because a fundamental approach of a SaaS solution is for the same unit economics, your offering gets better over time. You expect -- if you're paying a subscription to Netflix, you expect more content. If you're paying a subscription to a storage system, you need to ensure it gets better over time. The Evergreen architecture allows us to have always upgradable hardware. And this intelligent control plane is fundamentally changing the game by linking together 2 disparate spaces, observability, where you collect data; and automation, where you can actually take action on the data you collect. So let me drill into this paradigm shift because I think this is very important for folks to understand. Our -- historically, we've collected about 70 petabytes of data in Pure1 on an annual basis, like so over the last year, we know everything from how people -- like how the system operates, is low-level NAND, Bill will get into some of that. But we also know how people use storage. Think about Tesla, there are advantages in miles driven on the road for self-driving. Our advantage over 15 years is we've collected more information on how people use flash than anyone on the planet. And if you take that information and you combine it with Fusion, where our arrays are now networked. Charlie mentioned the network effect of arrays, you can go ahead and actually take action on different things, saying, "Hey, your environment is unbalanced. Let's dynamically rebalance your environment." So you're completely creating standard building blocks and optimizing the landscape. So this paradigm shift allows customers to manage fleets, manage sites, these logical sites, have policy-driven workload management, whether it be protection, tiering, et cetera. And this -- the ability to do it continuously is like a CI/CD pipeline for developers. It's not onetime, it's not manual, it's continuous. So our hardware has been getting better over time. Now your operations are getting better over time. And that's the power of the intelligent control plane. Now this Evergreen//One offering we talked about, you get block, file, object, on-premises, cloud, Co-Lo and all these SLAs, are built on this concept of this intelligent control plane because we need to continuously think you're running a service, you have SREs running the service. You need to be able to go ahead and have the capabilities to say, "You know what? If you want the service to run, operate efficiently, how do I change things dynamically in the environment?" You need this approach to say, "Let's keep the site rebalanced. Let's ensure that we can offer these ransomware recovery SLAs through policy." So as Rob talked about the core bits of IP, our service offering to deliver vendor-managed SLAs is built on this management and monitoring paradigm with this differentiated intelligent control plane. Now let's think about this competitively. We invented this Evergreen architecture as a core fundamental building block early in our company's history. This wasn't a bolt-on. It wasn't an after effect. But our Evergreen//One offering now allows us at the end of a service life to continually just say, "Oh, we can keep this hardware upgradable and current." It gets better over time. We guarantee in our contract, you will never have a data migration once you move to Pure. That's part of this offering. Most of our competitors at the end of the service life cycle contract is like, "Okay, now you have to pay for migrations," or they'll have to subsidize that in their unit economics to migrate from one generation of technology to another. Our service definition isn't starting with a box and a lease. We define these standard building blocks and SLAs that are set up by site. Here's a service site. This is what I need at that site. I need this performance tier and this capacity tier. We have a simple contract. All of our SLAs are transparent. It isn't like, "Oh, go," all those terrible warranty claims that you have to go and read, exclusions, et cetera. Most of our competitors have exclusions for planned downtime because they don't have this Evergreen hardware. They have exclusions for data migrations. All of these exclusions exist very similar to -- I think I said this to you guys last time, it's like all the fine print on a drunk commercial. Transparent SLAs. An SLA has real teeth because you have real financial penalties, and customers can monitor and see it in your product. Our management and monitoring in the product shows you any SLA violations. They're triggered right from our product. And then finally, because we're running a service, we pay for the capabilities that we're doing. We're renting space in your data center, so we'll pay for it with rack space and cooling. So these are some of the unique advantages that make Evergreen//One unique. And while we see people on this journey to these SLAs, some customers prefer to do it themselves. So with our intelligent control plane, customers have large service delivery capabilities, and they're building customer-managed SLAs. But increasingly, we're seeing strong adoption where customers prefer, "You know what? I'm not going to be able to standardize. I've got too much variability in my environment. I'm going to trust you as a vendor to offer vendor-managed SLAs in my environment." And we've now crossed over 1,000 customers in Evergreen//One in just about 4 years. So this offering is now starting to see broad adoption across a wide variety of sectors. Now this Enterprise Data Cloud creates a network effect because once you're on this platform and once you have these policies and once you have the standardization, it's a one-way street. This unified data plane gives you a single operating environment that you can program with our intelligent control plane right into the automation and build in your business. When you can provision things, you get built into customers' workloads and workflows, which creates a platform effect and stickiness for Pure as a vendor. So this allows us to make sure the next purchase that a customer thinks about isn't a jump ball. It's not subject to just NAND economics and dollar per terabyte because your value is critical to a customer's workflow. And if you think about getting built into a customer landscape, how do you think about vendors who do that well? Salesforce did that with Force.com. It was a platform where you could build extensions right into the environment. So if you think about platform companies, platform companies are unique in their ability to get embedded directly into the core operating environment of a customer's core IP that's revenue generating or cost optimizing. You're driving integration directly into their businesses. This platform effect with this Enterprise Data Cloud allows us getting built into real customer workflows and become core to standardizing their operating environment. So with that, I'm going to summarize and talk about this paradigm shift. Our vendors keep -- our competitors keep talking about this world, storage management. And some of them are even like, "Oh, I'm going to package up database management software with this stuff." But they're fundamentally missing this paradigm shift around how do you drive standard operating procedures at scale because that's the fundamental thing we need to bring into the data center environment. Charlie talked about the horizontal nature of unlocking the value. If you don't have standardized components and you don't have standardized operating policies, you're not going to unlock the value of your data. Customers can deliver that themselves, or they can trust us to do it. And increasingly, we're seeing customers trust us to do it. So with that, I'll summarize, and we'll bring up Bill, and he's going to talk to you about something I'm sure you're all excited to hear about is the Hyperscaler business.

All right. Thank you, everybody. Thank you for your time, coming out today. I'm the last presentation. Hopefully, it will be quick. I know we're getting fatigued. I'm going to talk about our technical advantage, our supply chain advantage and our structural advantage of our hyperscale offering today. Okay. So first off, I wanted to just step back and talk about what the offering is because sometimes there's a lot of confusion about that. And I talked about it last year, but I just want to go over it one more time, okay? So it's not an enterprise storage array. We're not taking our product and putting it in the hyperscale and just asking them to adapt to our solution, which is something that others have done, and you can make some money doing that. But this is an offering that is much easier for them to consume. We're taking our DirectFlash Modules and the part of Purity that controls the DirectFlash Modules, bundling that and offering it in a way that's easy for hyperscalers to consume. And it's not a point product. It doesn't offer one service. It can be used across their entire production environment, right? So we take parts of the DFM, which is like a next-generation SSD, the parts that control data placement, management of the NAND, we elevate it to software and offer that as a solution. And we've scaled the solution so that it can be consumed in very high quantities, which is something that's necessary in a hyperscale environment. And this solution has been fully released to production. So it's available today. So I wanted to talk about the elephant. So Charlie talked about the elephant. And the AI market is affecting this market in a big way. So AI has -- there's a tremendous amount of CapEx spending going on. And it's created this problem in the industry. So GPUs are in allocation. Everybody is fighting to get a GPU. And then you heard about HBM memory. So imagine all these companies spending money to try to get HBM memory to attach to the GPUs, it's created a run on HBM memory. Then it's happened to the DRAM market. Well, this tightness now has hit the storage media market, okay? So there's a tremendous amount of pressure on the hard drive market and on the flash market. Now the hard drive industry hasn't done much over the past decade or 2 decades to increase capacity, right? So they're sold out. There are 52-week lead times, and they're under a tremendous amount of pressure. This has also come now to the flash market as the hyperscalers are looking for any solution they can get in order to get storage media and increase the size of their estate and address these AI needs. So this has created a tailwind for us. There's a tactical tailwind now where hyperscalers are open to new solutions much more than they were in the past because they need to increase this capacity any way they can. And that's a market condition that we're taking advantage of. But in addition to a short-term tactical tailwind, we have the long-term structural tailwind of NAND flash versus hard drives. So -- if you think about a hard drive, it's a complicated electromechanical assembly. It's a very mature technology that was first released in the 1950s, and now it's become very mature. The cost efficiencies in hard drives, what they're doing in the factories and how they're improving things is raising the capacity very slowly, and it's lowering the cost per bit very slowly. NAND flash. NAND flash is a semiconductor commodity item, okay? Process improvements, technology improvements in semiconductors, lower the cost very quickly. These innovations have led that over the long term, we see a 2x decrease in NAND prices compared to hard drives. And this is a tailwind that behind the basis of the entire offering. But we don't need NAND prices to become cheaper than hard drives in order for this offering to work because the offering has inherent value. There are many parts of the offering that are just much better than any solution that they can buy today. So versus hard drives, we're 10x the density. We have 5x less space, lower power, we have fewer failures, and then we last twice as long. And what does this lead to? This leads to a total cost of ownership that's much, much lower for the NAND flash. It also means that the ancillary equipment around the solution is lower with flash. So imagine if you buy a bunch of hard drives, how many controllers, network ports, racks, power supplies, cooling, you have to put around this archaic architecture, right? But against flash, it's much more efficient. So we have this other advantage. So if you look at this graph with all its interesting and precise numbering, the hard drives today are scaling to about 30 terabytes, right? Our module this year, we will announce a 300-terabyte module. It wasn't that long ago that the average module SSD or DFM that you could buy was smaller than a hard drive. And this year, we will be 10x larger for 1 unit. We expect that to continue when we release our 600-terabyte module next year. All right. So the hard drive market is a very weird place, okay? The global supply of hard drives, 74% of the entire global supply of hard drives are consumed by a small number of companies, think less than 6 companies by all 74% of the hard drives produced worldwide. okay? This is a market that's ripe for disruption. As we work with these hyperscalers, as they move and change their architectures, this is going to have a major effect and immediate effect on the hard drive industry. All right. Let's talk a little bit about where our solution goes. So in the typical hyperscale infrastructure, they tend to have these horizontal storage pools that Charlie talked about. And they address the data needs of cold, warm and hot tiers. So I think a hot tier is something that's getting accessed all the time. You need very high performance, maybe you need low latency. Warm tier tends to be a combination of SSDs and hard drives and the cold tier tends to be hard drives and sometimes it gets super cold and becomes tape, right? Our solution has a way to span all of these tiers. Because we've elevated the solution to software, we can tune our solution and meet the needs of any of these tiers, different sizes of DFMs, different software tuning, different way to integrate with the hyperscale. And we are the only company that has one solution that can move across all of these tiers. All right. So we're not just competing with hard drives. We're also competing with SSDs. And as Rob was talking about, we don't really make an SSD. DFM is not an SSD. We've taken the functions that are complicated, that change a lot, that control the NAND flash and moved it to software. So effectively, we've taken something that's found in hardware that's hard to change over time, it's complicated, and we've elevated it to software. And that's made it easier for us to do things like qualify different NAND vendors. But one of the things that I didn't expect when we started this journey was that hyperscalers really value the agility. If you think about what's going on in hyperscale, it's a huge software development environment, dominated by software engineers. They change their infrastructure, meaning their software and how they control storage all the time. And they really value our agility. So SSD companies don't have that system expertise. They don't talk in the same ways that hyperscalers do about the storage stack. Since in our enterprise solutions, we've done the entire storage stack. That's not what we're offering here, right? But we've done the entire storage stack. So our engineers think at the system level. They can talk about system ideas. So our agility, our ability to change the offering really quickly, offer different protocols, different performance levels, how we handle metadata, how we pass errors forward. It's really been a very strong plus for us over using an SSD. All right. So on the technical advantages, I talked about higher capacity, density, the flexibility of the software. our ability to work with different NAND vendors. Also on the structural advantage, I talked about this tactical win we have right now where there's a lot of froth in the market and people are very open to new solutions, and that's giving us a tactical advantage. But I haven't talked at all about our supply chain advantage. And when you're purchasing something at very, very high volume, the way the hyperscalers are, supply chain is super, super important. They're not interested in a solution that's hard to buy. They can't get quickly. They can't get a lot of it, okay? So we have signed definitive supply agreements with Kioxia, Micron and Hynix. And those came out in press releases over the past year. Now in the total NAND market, there's only 6 or 7 suppliers of NAND flash, and there's only 5 of high volume. We have 3 of them signed up. If you look at the total NAND market here for each one, how much NAND flash each one make, 15%, 13%, 20%. But all of these are in our solution. So with our solution, we offer to the hyperscalers access to almost half the NAND flash in the world with one solution. We're the only solution like this. If they use an off-the-shelf SSD, they get access to like 15%, 20% of the market. With our -- and this is an environment that's really under a lot of pressure. So we're the only solution in the world that offers access to almost 50% of the market, and that's highly valued by our customers. All right. Thank you. I'll bring Paul back up to wrap us up.

Thank you, Bill, my hero. So as they bring up the chairs for Q&A, Charlie is going to have just an initial comment about one particular topic.

Yes. Thanks, Paul. So yes, as everybody is coming up, I am going to make an announcement. Of course, we're online so that -- this is an open forum. As you might imagine, as we're working with our hyperscaler customer and hyperscaler prospects, they're very sensitive to their own intellectual property, their own confidential information. And as we've talked more and more to them, they would like us to be as sensitive as they are to having that information out in the wild, if you will. So from this point on, whether it's here in our earnings calls and so forth, we're no longer going to refer to any of our customers or prospects by name, and we're not going to be explicit about any specific numbers relative to those hyperscalers, that is shipments, et cetera. We're working very closely with them. They've made their interests and concerns known, and we're going to respect our customers' intellectual property. So I want to put that out there because I know there'll be a lot of questions here, and we've been a bit perhaps more explicit in the past, and we're going to be more cautious as we go forward with their information.

Okay. Let's jump into Q&A. Let's start with Howard.

Howard Ma with Guggenheim Securities. Thank you for a very informative and tightly packed presentation. Charlie, when I think about the evolution of your product portfolio to what is enterprise data cloud today, I can't help but think about Jensen at NVIDIA talking about a fundamental re-architecture of enterprise data centers. And I believe he said it's something like a 10- to 15-year opportunity. When I think about the advantages of EDC, and I wrote some of these down. So just to say it back to you, unified software management plan, removing data silos, different storage tiers for AI to archive, which I really like that phrase, infinite scalability, multi-cloud, governance, cyber resilience, these make for a highly compelling and I believe, truly differentiated approach. So when your sales teams and your technical teams go into your enterprise customers today, which I believe one of you called it the North Star, what is the level of resistance versus acceptance of this automated approach that you're delivering? Are we at a -- and sorry for the multipart, but are we at a tipping point, do you think, where the use cases compel this automated approach? And then finally, when you talk about the optimism in your fiscal year guide, when you raise your fiscal year guide, how much of that is due to EDC increasing pipelines?

Yes. Multiparts are okay here.

Yes. It's a great question, and things are moving very rapidly in the market. In terms of your question, what is the resistance, what's the openness? Things are moving very rapidly in this market. Data storage was -- has been traditionally a very slow-moving market. Resistance to any modification of the way that customers bought or even sellers sold data storage was very high. One of the reasons why I opened the way I did is that AI is causing -- regardless, there's been a lot of hype around AI in terms of how much you're going to sell into AI. What -- the benefit of AI for us right now is customers are rethinking their architectures, okay? It's not about how much we sell directly into AI. It's that how open are customers to rethinking their overall architecture in their environment. And so the discussion around enterprise data cloud has really opened up. Now it is not correct. It would not be correct to say, yes, now a majority of our customers are flocking to this whole idea. But what we are seeing is leading-edge customers really listening and it's starting to affect the way they are thinking about purchasing. Now again, this is at higher levels inside an IT organization, not at the lowest levels. I don't think it's really yet fundamentally changed individual storage admins, although I think it is going to in the near future. But certainly, at the higher levels of an organization -- we are speaking to higher levels in the organization, and we are finally talking to them about franchise opportunities rather than individual use case opportunities. So it's making a difference.

Aaron, please?

Actually, let me just finish.

Sorry, multipart.

And that's what's giving us some optimism about the -- it's not just about optimism. I mean, obviously, we have pipeline analysis in all of the individual financial analysis that we go through when we do forecast, but we are seeing this turn into bigger opportunities.

So let's go to Aaron and then Jason.

Aaron Rakers at Wells Fargo. I appreciate you doing this day. Paul, you should have never said that about multiparts. So I guess my first part of my multipart is, I can appreciate that we don't want to talk about the hyperscaler customers. That's no different than what we've heard from a lot of other vendors that participate in this market. But the comment that you've now released the solution, fully released the solution to production, can you just remind us again on the pace of deployment? I know last quarter, you talked about pilot deployments and importantly, how you see the breadth of these opportunities evolving beyond maybe just that one customer without naming any customers as we look into next year? And then I'll throw my second question out there. Just kind of building on that earlier question, this Fusion layer just Tarek, you're in the stage, how does that get priced? Is that -- is there a software element to what's evolving or deepening in the Pure story that we should all be thinking about here as we move forward?

Let me touch upon that second part first, and then I think between [ Kaz ] and Bill, et cetera, we'll touch upon the first part. So Fusion is not -- we are not monetizing Fusion directly. Fusion now is part of the Purity software. It is -- we call it out because it's a unique new addition to the Purity software, but it's part of Purity software. When we sell software as part of our -- that's embedded in the product, we don't charge extra for it. But I can think of a number of analogies. It creates a network effect. There had never been -- there has never been in storage a reason why the next array that a customer buys should be the same vendor unless it's economics or a TLA of some type. right? In fact, they often -- every use case, as Prakash mentioned, is another jump ball. It's feature versus price versus whatever. We think this creates a network effect, which is valuable for the customer because it makes their lives easier, it creates more consistency. It will create more efficiency for them. But for us, really, if we reduce the amount of effort that it takes to sell the next array, it will get us larger market share and the larger wallet share.

Yes. So I made a comment in my presentation that we're released to production. So that is a commentary on our status. So about the maturity of our offering, and we've gone through our testing, and we've released our solution to production, which means it can be mass produced. It's not a commentary on the status of where our customer is in their journey. So -- and I can't talk about their status. We continue to have engagements with others. Yes. I can't go into too much detail about what point we are with each one, but we are having conversations, and we continue to work with them.

And we're more convinced than ever that, again, this is the time the transition is starting. And over the next few years, there's no reason for any of the hyperscalers to not do the switch and you saw what it's going to do to the disk market. It's just going to wipe out disks because flash is a better solution that is advancing at a faster rate.

I think another way to think about this is you only have to believe that we are a third alternative to what are 2 alternatives today, hard disks and SSDs. And then the question will be what type of market share out of that do we get? Right now, it's all opportunity for us, right? I mean 2 quarters ago, it was literally zero. So it's all opportunity.

Okay. Let's go to Jason, then we'll come over here to Asiya and then okay, Erik, after that.

Jason Ader with William Blair. Good to see everybody. My question is on the hyperscalers. Can you just walk us through maybe like historically, what the split between HDDs and SSDs were for the hyperscalers? And then fast forward a few years, [ Kaz ] to your point, like how does that mix shift change between HDDs and SSDs? And then what piece of the SSD market do you guys think you can get? And like how -- where do you fit in into that mix because I imagine they're going to continue to use a mix of media.

Yes. So I can talk about that for a minute. All right. So obviously, in the distant past, they were all hard drives. In the -- recently, let's say, within the last 10 years, they moved almost all of that hot tier up at the top to SSDs to TLC-based SSDs. TLC is a version of NAND flash. It's the fastest. And those are -- that is sort of the first layer that many of their accesses go through. So most of that in the hyperscale has already moved, okay? Then you get to the warm and hard drive space. They're starting to integrate more and more technology, including ours, in the warm space and then growing down into the colder and colder spaces. So they are in the transition of doing it today. Does that answer your question?

Yes. I mean I'd probably guess single digits for SSDs as a percentage of exabytes?

Yes. Obviously, it depends on which hyperscaler it is. Some are further on their journey to others, but there are some hyperscalers that are still in single digit on the NAND flash.

You guys are the TLC SSDs, they don't go away.

Right. But our solution can be used in that layer as well.

Okay. But you are actually more in the warm layer, the QLC SSDs.

Warm is where there's a lot more exabytes and that's sort of the best opportunity for us right now. Yes, we want to take the hot layer, but the warm layer is much larger and the economics are just fantastic for us. And then you move down into the whole layer. And if you're a hyperscaler, you're going to start at probably looking at the warm layer because you say, I already have SSDs here. I get less of a boost by switching that. So you're probably going to start looking at the top of the warm layer and again, move down. And as we do a good job for them, move up. Charlie is right that we're never going to have 100% market share. None of these guys will ever bet on a single solution. They always will keep their optionality. Our job is to be the best supplier and get a huge portion of the market share, but we'll never be 100. We'll never be 90%, right? Because they just physically would not ever, ever do that. And so we want to capture every bit that they have, we want to play for, and we want to capture as much of that as possible.

And then Jason, just add one more thing to that. A part of your question was, hey, historically, how much of the mix of deployment in the hyperscalers has been hard disk drives versus flash? I think this is an area where the hyperscalers have almost paradoxically lagged the enterprise in mainstream adoption of flash. They have been able to, through R&D, through just extraordinary means, let's say, be able to squeeze the last bits of optimization out of hard disk drives. We're seeing that start to change. And when we -- when it changes, I believe it's going to happen in a very quick way, right? The hyperscalers get incredible economies of scale through standardization. They're not going to -- it works against their operating principles and economics to shift gradually over time, 5% here, 5% there over time. They design their data center architectures in generational horizontal designs for a reason once they design that, that now gives them a template to go and replicate and operate with uniformity. And so to Bill's point earlier, the hard disk drive market is an interesting place right now. We clearly have our view on how that plays out over the next couple of years. And I think it plays out in pretty discrete steps.

One other thing I'd add, you asked the question, it opens all the ads. In talking with some of the top engineers at the hyperscalers, they will point out to us how much difficulty they have in using all the capacity on the hard drives, right? As the hard drive grows in capacity, but it doesn't grow in performance, then they need the data to be colder. And so that's another advantage that if you remember Bill's chart, as our capacity goes up and up and up with the flash, they can use all those bits. But if the hard drive vendors came along tomorrow and said, "Hey, we've got a new drive that's 40% larger. The hyperscalers are like, well, I can't use those bits because I'm not getting more performance, and so they lose the economic benefit. It also means that they'll tell us things like we're overbuying the number of hard drives to get extra performance by 30%, 40%. As we start to look at flash, we could shrink that number. So that would further dampen -- dampen the growth of total bits. But given that there's like 1,000 exabytes to play for, we have a long ways to go before we're ever worried about that part of it. But you have to recognize like disk as a technology is coming to the end of its life. It is not improving the way it used to. And people have been twisting themselves and knots trying to use it better for a decade now. And that's why this transition is going to happen and why it's going to flip.

One quick last one, Paul. The TLC, the hot layer versus the QLC, is the TLC SSDs, does that have higher performance than what you guys are offering in the QLC?

Bill?

Well, so TLC offers lower latency access than QLC. But one of the things that we're seeing is you can address the very high-performance workloads with QLC. So our AI offerings, we talk about EXA and FlashBlade. FlashBlade is one of the highest performance storage arrays in the world. It's all QLC. So we are confident that we can address a large portion of the market, the QLC. But in some ways, it doesn't matter. We have expertise in TLC. We've been shipping it on our DFM since 2017. If that's what it takes to hit that, we'll do that.

Thanks, Jason. Good multiparter. Asiya?

Sure. If I can just dig into the neoclouds opportunity, doesn't really get a lot of attention. I mean, I heard a lot about your scale AI. Just help us understand, I mean, there's some competitors there who are obviously dominating a little bit in that space, at least in all the media, and I've heard their presentation. So how does Pure play in that space? And can you maybe identify that opportunity and how big that could be for you guys as you kind of scale this out?

You bet. It's an interesting new market. In the past and also in my earnings calls, I've tried to put it in perspective in terms of overall size. In a $50 billion storage market, not even counting hyperscalers, it's still just a few billion dollars, right? But it's an interesting new area. Now Gen AI evolved from the HPC space, right, and the high-performance computing space. The high-performance computing space has always been an interesting niche in a number of different businesses, everything from the compute area, you remember -- you perhaps remember Cray Computing, right? And that was -- became part of HP. But also in terms of storage, it was a niche space as well that a couple of private storage companies played in. One of the things about HPC was it was historically sort of like Hotel California, an organization would go in, could never extract themselves because the values in the HPC space, which were largely focused on scientists were very different than what enterprise values were. And you would get caught up in the benchmark competition that was taking place in that environment, which really didn't merge very well or suit itself to an enterprise market. So we chose early on to stay focused on the enterprise market. Gen AI takes off all the practitioners come out of the HPC market. We were behind. I'd be very direct. We were behind in that space. Now we have always, I think, exhibited great discipline in making sure that we did not build bespoke software for every new segment that came along. So we looked at what we had, what capabilities that we had in the company, and we were able to then rally around our metadata engine and using that as a way of entering the market. We introduced FlashBlade//EXA earlier this year. We believe -- well, FlashBlade//EXA is now benchmarked as the fastest AI storage engine in the world by several factors. But we still have work to do in order to really get to the head of the pack from features and so forth. So Rob, maybe you want to take it from here.

Yes, absolutely. Just to add a few comments to what Charlie said, just reinforcing the point, we've intentionally through the company's history, not started with a focus on HPC, but rather, as I mentioned in my presentation, the needs of the broader enterprise. What we have seen through now multiple new markets is the opportunity to leverage those investments, the IP we've built there in different ways to go address the needs of different markets. Said more simply, if you look at the HPC space, companies that start there have a lot of trouble moving into the broader enterprise. We've been able to go the other way and say, "Hey, let's start in the enterprise and now take some of those fundamental IP blocks and go address the needs of the neoclouds in this case. One of the things that really was behind -- kind of the driving principle of how we built FlashBlade//EXA from the IP based on FlashBlade is we step back and we looked at what are the core strengths of FlashBlade? It's the software. It's the scalability that it delivers in the metadata. For the enterprise, right, it's like any design decision, you design for a certain space of possibilities. We deliver really, really good performance, really, really good economics, really, really good efficiency that covers the needs of 99.5% of the needs out there. What we have now done is to say, hey, if we go and take the software that's in FlashBlade, that scalable metadata engine, pair it with an open hardware architecture, we now have a very compelling offer to now supercharge the performance there to go after that last 0.5 percentile. And so that's really the evolution that we've gone through in terms of addressing the needs of the neoclouds. And as Charlie said, it's -- we just went generally available with that solution in Q2 and in early discussions with a lot of the top names and really good initial signs of interest.