We are excited about today's discussion and the future of Everpure. 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.everpuredata.com.
Hello, everyone. Welcome to the Everpure Financial Analyst Meeting, held for the very first time here at our headquarters in Santa Clara. We have a full agenda for you today. We will start off this morning with technology deep dives on data management and AI. This will be followed by a technology-focused Q&A session. We will then break for lunch and resume again at 12:00 P.M. with the main session featuring Charlie, Tarek, and Rob. For anyone needing information on things like restrooms or Wi-Fi, we do have people in the hallway who can help you with any questions. When we return from lunch, Charlie will lay the groundwork with an overview of our strategy and our expanding opportunity. Actually, our significantly expanding opportunity. Rob will cover our fundamental advantages and our right to win in four high-growth market areas. These include our core plus three additional high-growth market areas.
Tarek will conclude with our foundation for future growth and our long-term financial framework. Following that, we will conduct a final Q&A session with Charlie, Rob, Kaz, Tarek, Bill, and Prakash. After the final Q&A session, for those of you here in person, we will have an informal cocktail hour. Before we get started, I will remind you, as I do every year at this event, that the value of everything that we do at Everpure is driven by our software. This includes our core business, as well as our three additional high-growth market areas. For all of our high-growth market areas, once again, it is always about?
The software.
Thank you. Thank you. The software. Okay, now I'll turn it over to Charlie for his opening welcome.
Good morning, everyone, and welcome. For those of you who are here in the immediate audience, thank you so much for coming out. We know that it's a long trip for most of you and many of you, and we really do appreciate your physical presence here and also want to thank everyone who's online who will be going through our discussion today. We're very excited. The team here is quite excited to speak to you about everything that we're doing. A lot is changing. Obviously, a lot is changing in the outside world, a lot is changing in our industry, but a lot is changing at Everpure as well, and it's a great time for us to be able to express this to you in many ways. One is we believe that our business is accelerating.
We hope to be able to show that to you today in numbers. It's also expanding into a number of areas, and that's obviously what Paul alluded to, and as you can see from the agenda, different areas that we'll be providing you a lot of background in. One additional thing, for those of you in the audience that is here physically, we left you with a paper on the table. It is a bit of a long read. It's about a dozen pages. But we think it's a seminal view of how IT architectures are going to fundamentally change over the next decade. We believe that this, brought on by AI, is driving the need for organizations to start to organize around data rather than around applications. That's a very fundamental change in how IT architectures are built.
That fundamental change really is going to open up very new opportunities that we, as Everpure, are pursuing, and other vendors are going in that direction as well. But unless you understand where we are today and where we're going as an industry, how that's changing and moving towards a different architecture and a different relationship between applications and data, you won't fully be able to predict all the changes that take place. So I'd highly recommend that in your leisure time, you scan that document and get a good understanding of it. With that, I'm going to invite our first speaker on the stage, Prakash Darji.
All right. Thank you. All right. I am going to double-click on the first area Charlie mentioned, which is this idea of data centricity. We call this data primacy. It is an inversion of the application architecture. As we get into our business, we are really talking about how we can help customers maximize the value of data wherever it sits. Pure, non-Pure. We are getting into the focus of managing the value of the data and getting the most leverage out of it. Two, why we believe we can do so is we already have the trust of lots of mission-critical data across a customer's landscape, both structured and augmented with unstructured as well. The third thesis we are going to get into here is why a storage company has to do that.
Because as you think about data, why does context need to be both traveling with the data and be persistent in storage? If you think about the biggest data management challenge today, and I bring this back to, prior to here, I worked at SAP 13 years building applications, both financial and database. A large portion of the world today is customers buy applications, and IT departments integrate the data between those applications. If you think about the total spend, actually 80% to 90% of all of the spend goes into the integration work between all the assets you have. It just compounds the bigger you get. Largely, the data integration costs are largely the biggest cost associated with running an enterprise today.
If you think about agents and agents' access to data, where it sits, where context is trapped in applications or in lake houses, et cetera, that actually is going to blow open and be more exponential in terms of integration cost. These silos of information that sit in applications, warehouses, et cetera, need shared context to do anything with AI. Agents work better directly on data. That is a challenge we are seeing in the industry today.
Now, if people are running applications, you will see SAP strategy saying, "Hey, you want to approve sales orders, so bring all your Salesforce data into SAP." Salesforce is saying, "Hey, you should use Agentforce and bring all your SAP data into Salesforce." Databricks is saying, "Okay, bring all your data into Databricks." Everyone right now has a part of the context picture, and this is what leads to the fragmented context across the application. Our first approach is how do you create shared context with data wherever it sits? Leverage the data in place and understand how to build shared context. Two, instead of moving the data between systems, how do you allow reusing the data where it sits? What that does with the integration complexity is it inverts the way you develop apps. Historically, your data has been owned and encapsulated by the application.
This is my finance data sitting for my finance application. This is my sales data sitting in my sales application. We do see a world in data primacy where application workflows plug and play and reuse data across multiple apps. It is a different way of building and architecting. In the early days of SAP, when I was there, we were trying to build a big schema to handle what we would consider end-to-end ERP. Largely that became siloed and fragmented as people started buying different applications and not consolidating to a single referential data model. In the world of AI, we do believe that this idea of plug-and-play workflows will allow you to reuse data where it sits. When you hire a new employee, you don't need to copy all your financial cost center data, you can just update your financial costs.
The separation between HR and finance should not be as wide as it is today. Let's get into it. Data primacy. If you take this approach, it reduces data integration, reduces your data copy, increases the data coherence, meaning the meaning of your data across your landscape. Obviously, less is always better in a cyberattack surface. By doing so, you can also optimize your token spend, meaning the inputs you have that are relevant to models, instead of building a large language model factory, an SLM with the right data and the right context will get you better results at lower cost. We've seen that over and over again, and that's how agents are being built. They're being purpose-built with targeted data. Our beliefs as a vendor. First, we believe data will be distributed. There won't be one thing creating data on the planet.
You will have sensors, and you will have edge devices, and you'll have mission-critical applications and innovation coming from everywhere. We do believe that's to continue to be true. Two, we do expect over time application development to change. A lot of application development was typically written in source code. I would write my pricing in source code. That was part of how you would do intercompany consolidations. We believe that will be moved to more configuration-based as data. It will be persisted as data. We are already seeing people do this as markdown files in Claude. Kubernetes actually started this. In Docker and Kubernetes development, people actually started moving configuration into configuration files and compiling those configuration files. It was kind of the first view of what I would consider data primacy for software engineering.
It moved to that approach where you had configuration that you could update and then compile. Now, context will be stored with data. This is interesting because if you think about data, and storage specifically, has limits. It's like, "what physically fits in this data center? What physically fits in this location? What is the performance of this thing?" Over time, as data expands, you typically have to say, "Okay, I'm going to expand it," or, "I'm going to place it over there or move it over here." But every time you do so, your AI would have to retrain. If you really want to preserve what you've learned about that data, you actually have to persist that context with that data. Largely, when you have that context, it's not about the context of what's in that file.
It is not, "This is a PDF with an invoice number," but I need to know where that invoice is being used across other data sets if I am loading data into agents to do invoice approval. Shared context is needed, and what this means is a storage operating system then becomes a place that needs to store the data context and provide shared context. It is a persistent memory problem, right? Over time, a storage operating system is the only way you can actually persist memory context for data. That is why we have decided to enter the space, and we look at why storage and data need to come closer together. If you think about the problems, there are different problems we have historically dealt with. A storage problem is typically oriented with data residency, and when we talk about residency, it is not just a physical location.
Where does it sit? What performance tiered or performance class do I put it in? What protection group do I put it in? When I was an application developer, I always thought, "Okay, where am I going to place this data?" I just thought about placing it in that thing, and I would inherit the policies of that thing. Chadd is going to get more into that later, right? What cost tier is it placed on? You have heard us talk about tiers of cost. A data problem is different, and in the world of AI, data sovereignty is who owns your data, who is allowed to train on it, who can use it, and does it still differentiate you? If you really want to understand, you need to go look at the data and have policies around managing the data and data access patterns as well.
Because the more valuable AI becomes, the more control over the underlying data you need. Otherwise, and I bring this up, many years ago, Walmart had a policy saying, "You know what? We are not going to use AWS as a cloud infrastructure provider." Why? Because they are a competitor in retail way back when. Largely there was a fear that they would train on their data. If you see what is going on right now, if you do not control your own data and maintain the ownership of how it can be leveraged, all of it is being used to train other people's models and IP, and that is what data sovereignty is about. What is fascinating is the market. The market that traditional competitors are actually working against data sovereignty.
You see a lot of vendors, hyperscalers, app vendors like SAP, Salesforce, Oracle, Databricks, different vendors, all taking a walled garden strategy. Bring all your data to me and I will provide value to you in my platform. It is about a data collection. Our view is you need to use the data where it sits, and you need to allow the customers to own their own data. Largely we have entered this segment with a set of capabilities for data management, and there are other vendors that could take that strategy. Integration providers, and iPaaS vendors actually work with data wherever it sits, but they are not focused on this data management problem. They are actually chasing agent studios, agent orchestration versus the data contextualization we are talking about.
So it creates this opportunity right now where we see this inflection point of a problem that needs to be solved and not a lot of people solving it. I am going to bring up Ashish and he is going to talk about how we are solving this problem.
Thank you. Hit the ticker.
Good morning, everyone. Everpure Data Intelligence is an important expansion to our portfolio and our offering. Historically, we governed the data and stored the data for our customers using our unified data plane and the Intelligent Control Plane . Everpure Data Intelligence adds universal Data Intelligence across any kind of data. We discover data regardless of where it sits, whether it is in the cloud, on-premise, even on mainframes. We then also think about it from a structured and unstructured perspective because data is data and it needs to be discovered. We look at it whether it sits on Everpure data storage or on any other storage as well. This allows us to then classify this universally accessible data in such ways that we can determine what is the sensitivity behind that data and how you should manage it. That is when the magic starts.
We start to contextualize all this data based on business processes and the relationship between data, whether this is an employee, whether it is a customer, whether they are both of them. This allows us to provide this contextual intelligence that is required for AI tools to have better inferences and much more accurate decisions that they can provide to their various customer business use cases. All of this will be available at full scale on the Everpure data infrastructure and is already being used across multiple different organizations today. I have spoken to about 100 plus customers in the last three years, and there are three very important themes that are coming up. First, customers want to have context-rich data and intelligence about that data right at the source so they can make good decisions based on that data.
Second, we are seeing that large enterprises are bringing a consortium of decision-makers when it comes to purchasing data management tools. This includes the CIO, the CISO, and the Chief Data Officer as well. As a result of this, the budget is tripling in terms of their spend on Data Intelligence tools. Because the CIO is using the information specifically to understand how should the data be managed in the infrastructure. The CISO is taking and prioritizing that intelligence-based information to see what is the posture of the data, what is the sensitivity of the data to take action on the data.
The Chief Data Officer is being added to the mix because they can now provide actionable intelligence based on AI-ready data which has context, which has relevancy, and which allows those cost efficiencies to come together on AI tools because it is bringing the most context-rich relevant data to Everpure. This allows us to have not only one scan to enable multiple different use cases, but three different personas also get the benefit of it. This is also the reason that this market for Data Intelligence is going from $6 billion today to $21 billion by 2030. As I was saying, this is used by Fortune 100 companies today, by large enterprises across the world, and Everpure Data Intelligence is delivering great results for our customers.
At a major credit card company, they scanned 14,000 databases in two weeks to really understand what their data landscape looked like, because they were having 700 Data Subject Access Requests coming to them. It used to take them 21 person weeks. Again, I will repeat that. 21 person weeks to respond to each Data Subject Access Request, which is a GDPR requirement. With Everpure Data Intelligence, they can do it in less than three minutes. Massive cost savings for this credit card company. Similarly, another payment processor found 91 million records of credit card information stored in a completely unprotected database. That could be a huge security risk. We were able to identify that.
To add to it, we were able to tell them that instead of what they thought that they had 189 million personal identifiable information, we could reduce that down to 39 million, largely because they had a lot of duplication of the data that Prakash was talking about. When you do integration, you start to copy data and make it available outside of those applications, which creates duplication. This reduced their cybersecurity cost tremendously, and it was allowing them to also be more efficient as a business. Lastly, a global large bank was asked to comply to GDPR restrictions and policies, and in two months, they were able to scan all their data and get ahead of the audit requests by providing full GDPR capabilities because we were able to provide them very specific visibility about their data.
Not only visibility, but also the context that is required to make better business decisions. With that, I would like to invite Prakash back up. Thank you.
Thank you. For those who do not have the history, this came from our acquisition of 1touch.io. Ashish was the CEO of the company. He is now our General Manager and running this business for Everpure. Largely, the history started in more of a CISO security posture area, but that leverage he mentioned around users expands. Because now in the world of AI, if you had a catalog of where your information is and you wanted to say, "Hey, I want to build that" I think I used that example of invoice approval. What if you could just query with MCP, "Where is my invoice data and what is primary source?" Otherwise, AI might think the spreadsheet on someone's desktop is the same thing as your back office NetSuite system. It does not know.
You absolutely need to catalog, discover, and know, and this relevance now is more applicable in the world of AI. What is interesting is we have a lot of leverage that allows us to work on the data that matters most. Ever since I have been here, we have had a strong net promoter score. It keeps building every year. 64% of the Fortune 500, 84 Net Promoter Score. Our customers trust us. We have strong trust in serving mission-critical data. Mission-critical like your most mission-critical financial records, transaction, mission-critical databases, SAP, Oracle, et cetera, run on our FlashArray today. With our data plane, we have extended to support block file object, structured and unstructured, so we can bring together serving this data to our install base in a turnkey way. We have a massive leverage point on how we provide context directly to customers today.
Two, that leverage compounds because as people look at Data Intelligence and run these scans of their landscape, we can now see our share of wallet into accounts and show customers the value of moving data to Pure. That provides a product-led growth opportunity and ability for us to expand our footprint into accounts. Two, as we take a look at our storage system, by having the Data Intelligence embedded and having this context, and Chadd is going to get into this in a little bit more detail, we have the ability to go ahead and say, "You know what? I could go ahead and provide database management for the storage system." Because today in storage, if you wanted to say, "I have got this invoice approval agent," how do you offer high availability? Has anyone even talked about that? It is mission-critical. It is approving revenue coming into the company.
A storage system is where you set policies for RPO zero, recovery points, et cetera. If you wanted high availability, you would need to provide it using a storage system, which means you need to manage your policies based on the data itself, not the storage. We are unique in the approach we are taking here. Whether you are a CIO and you start with the journey of storage where we talk about running workloads, automating configuration, delivering performance and capacity SLAs, you end up at a place where you are optimizing for data residency. If you start as a Chief Data Officer, we are talking to people about, one, the first step is total visibility. Discover your data, understand it, and reuse it. For CISOs, this gives you a unique view for cyber resiliency as well.
You have the ability to not just look at the data, but to verify it and recover clean copies. Use cases we see, customers wanting to build secure data enclaves. How do I ensure all Personally Identifiable Information or PII resides in this zone and doesn't leave this zone? It's a capability where storage and data management need to come together. Ransomware use cases where not only do you detect, but how do you verify data integrity? You need to look into the formats of the data, which Data Intelligence provides. We talked about the agent policy use cases. In AI, how do you get better quality of outcome at minimal token cost? Well, one, only bring the relevant information into your model because your input to output token expansion comes down drastically and your results go up overall. We've seen that proven over and over.
The early problems you saw in AI where people actually just you hear the word AI work slop. There was a lot of slop going into AI models saying, "Just give it everything and ask questions." You'd see a lot of things and questionable quality. Focusing on the data in is important. With that, I'm going to bring Chadd over, and he's going to double-click into some of the core platform. Thank you.
Good job, brother. All right. Good morning. Excited to showcase some of the innovation that we're building into the core platform today. Prakash and Ashish showcased a lot around Data Intelligence. It gave the benefits of being able to discover, classify, and contextualize data and provide that context that's needed down into the infrastructure stack that we're building around this core platform to allow customers to take the innovation that we built in and just continue to compound that value that customers have. You'll see in this presentation, I'm going to go through three main areas. These areas are the areas that customers are most challenged with today, and we spend most of our time digging in with them to help them along this journey. The first step here is helping them being able to unify on one core platform.
The fragmentation of data has become a fundamental problem and shown its cracks with AI. The second big area, as Prakash talked about, is the ability to be able to control data, apply policy directly to the data, and allow it to follow along as it moves within the environment. Lastly, we're going to talk about the ability to operationalize AI, optimize it, activate the data, and make it so it reduces the overall token cost. We're going to dig into all of these different areas today. I'm excited to spend a lot of time with you on just why we're building these and the differentiation in comparison to what we're seeing in the market today.
When we look at one core platform, one of the biggest things we hear from customers is, "I've got way too much stuff to manage." Silos exist all over the place. Each one of those silos have their own UI, their own console, their own ad hoc configuration. It's a mess. Every single application gets deployed, or a new storage requirement creates a new one. In the world of AI, you actually need to be able to access all data holistically, not just be able to look at one specific single system. Our competitors, comically enough, will tell you, "Let's go build another silo for AI. Add more complexity, create new copies of your data, and have you manage even more challenges that are out there." Each of these separate systems are their own snowflake.
It's very hard for people to be able to manage their data estate without them being able to actually unify this together and apply governance directly to the data and take the context that we're taking from Data Intelligence and apply it down to the infrastructure stack. We decided the model needed a complete rethink. The rest of the market was thinking down here. This is storage management. Now, we made the best boxes on the planet that produce storage solutions. A lot of great software to make them simple, efficient, reliable. But most people are continuing to stay down here, and our goal here is to try to move people up the stack into building better efficiencies and workflows that allow them to be able to really get value from the system. This first stage is about providing autonomous operations.
It takes intent from an end user, what do they want from that data, and allows the infrastructure to actually manage it. It's great. Think about this model for a quick second, where you're able to just define, what do you want? I only want models to access this data. I want it to be highly resilient and not have to manage any of the configurations to do so. On top of this, we're building in data management componentry, which we talked about earlier in Data Intelligence. This informs our Intelligent Control Plane and unified data plane, which I'll go through a little bit later, and what it needs to actually do based upon the context of data. Because just executing intent doesn't always allow you to understand exactly what data and what kind of context is associated to it.
To fix this problem, we decided to build what we call an Enterprise Data Cloud . What an Enterprise Data Cloud is it starts with an Evergreen architecture. It's one of the things that customers love the most about Everpure, is the fact that they can get new generations of hardware with no disruption. It's a subscription to innovation where they consistently get more and more value. It compounds every single year, and they never move off the systems whatsoever. We then built a unified data plane. It allowed them to break through all of their silos and provide one common user experience throughout, but do so with multi-dimensional performance so that they could handle every possible workload from archive all the way out into AI. On top of this, in order to create this unified data plane, we created an Intelligent Control Plane .
This is where the brains of the operation exist from an infrastructure stack. It takes the context from Data Intelligence and now applies it down through the infrastructure core, taking the intent of users and applying it directly to the infrastructure stack. As you saw with Prakash and Ashish talking about Data Intelligence, this just becomes the context and meaning. All of this put together is our core platform we deliver to customers, and the value that they get is substantial. We've put so much R&D development efforts into this that they have changed the way that they think about their model. Now they're moving from storage management into actually managing their data directly. Let's dig in a little bit on what a unified data plane is. I get a lot of questions on what exactly is this.
The first is we delivered one common operating model. There's no different consoles or UI. It's one single experience that users have across their entire data state. Second is we built this on a core fundamental technology, our key-value store that provides metadata-optimized access to data, so they get a very common experience across each one of their use cases. On top of this, we built this on an Evergreen architecture. A unified data plane can't take downtime. It can't take disruptions. You can't continue to be moving it. It's got to be solid as the foundation that this is built on. Inside the data plane itself, we've got a whole bunch of cool technologies. FlashArray does really well for transactional low latency environments such as databases and virtualization. FlashBlade is great for scale-out analytics and AI, where high concurrency and high throughput is needed.
For those that want to scale well beyond what an enterprise does into the neocloud scale, we have FlashBlade//E that we've built for tens of terabytes of performance. Then for our hybrid customers, we can go across cloud and the enterprise to build a solution where that data plane connects all of them. The benefits here is that we can apply block file as well as object capabilities across enterprise applications and also AI in the exact same platform, removing the need to create new silos and new complexities for customers. Our Evergreen architecture is the thing that everyone has loved the most about Pure from the very beginning. You never replace the gear. It swaps out modular in nature, stateless, so it never takes downtime.
Over the last 15 years, we've delivered nine different generations of hardware for our customers, and every time they get performance gains, capacity density gains, all without taking disruption. I use the analogy of imagine if you wanted to remodel your house or even move to a new house. Imagine while you're out to dinner with your wife or significant other, that your house when you came back was brand new. You didn't move anything. Moving sucks, right? You break some stuff, you lose some stuff. In a world where the house just gets better instantaneously is the experience we deliver to customers. On a unified data plane, this becomes incredibly important. Our software innovation continues to compound value for customers because of the fact that we deliver about 200 net new features to customers on an annual basis.
They are getting this subscription to innovation consistently, which adds massive value. Our DirectFlash technology was built originally in the enterprise, but now we are delivering this in hyperscale environments, and we are getting testing now at scale that no other vendor has ever seen before. The value of this is resiliency. We get 10X better resiliency than what you will see in off-the-shelf SSDs. Overall densities at 300 TB drives will be 10X more than what our friends in the industry around disk for rack. Then also reliability and operating costs become key capabilities of continuing to scale this. The reason why is because we built something so different that the rest of the market continued to just use the off-the-shelf devices. It was software-driven, globally managed, and we made it so it was highly efficient and resilient from the get-go.
All right, so now that we have unified the data plane, let us talk a little bit about what we are doing in the Intelligent Control Plane to actually apply this control mechanism around data. The first thing is Fusion actually fuses these systems together and makes this unified data plane, but it also becomes the control mechanism for managing data. What we do for customers, and what is great is we have telemetry data that comes to us every 30 seconds from all of our systems to our own cloud, and we can help them be able to define what intent should be based upon past experience. Once intent is understood, and even better when you actually plumb in some of the Data Intelligence capabilities, you get the ability for the platform to actually be able to execute autonomously.
The way this works, as Prakash talked about, is it is applied directly to the data, so that the system can understand the intent that the end user has based upon our contextual knowledge, and it can monitor the system to ensure it is delivering on the commitment they gave back to the business. This ability here not only increases value to customers, but just think about this for a second. You do not have to control anything and manually manage it. You apply intent that goes across every single one of your applications, and it autonomously runs without anyone managing it. The benefits of this are huge in an enterprise who is dealing with, in many cases, hundreds, if not thousands, of storage systems at once.
We take this context, and in many cases, it may be storage context, such as what Prakash talked about in data protection or resiliency or residency, but also, you will see here shortly, Data Intelligence context. We apply this to a grouping of data, a semantic data group, we call it. This allows you to have data sets that are organized now by meaning, not just because it happens to be sitting on a storage system. This meaning thing allows you to be able to define policies such as, this data set must remain in Germany, or only approved models that I have defined can actually use this data set. This intent-based environment is so ridiculously powerful to customers, but what is more important is the ability to autonomously execute these capabilities within our Intelligent Control Plane . We use this concept called supercontext.
It is a mix between storage context, which we know from the storage infrastructure space, paired with data context that comes with data's meanings and relationships and the like. The power that these two bring together allow people to be able to have better governance, more trusted AI, and more accurate autonomous decisions. It allows you to control the data at a data set level, but now actually understand what it means to the business, so that when you apply intent, it is not just make it more resilient. It has an awareness to the business intentions associated to that data as it moves within the environment. What has been great is Everpure Fusion, which is the capabilities that underpin the unified data plane and applies much of the intelligence that we built into the control plane so that users can take intent and apply it autonomously, has been well received by our customers.
We already had 2,300 customers that have adopted the technology, and that has increased almost 2x just within the last six months. Customers are loving the fact that they can take this technology and remove storage management altogether and autonomously have the infrastructure manage itself while giving them the opportunity to truly understand their data above within Data Intelligence. Options IT/Options Technology , as an example, went and saw massive savings, hundreds of hours saved, in fact, per month, per users, because of the fact that they were not actually managing storage anymore. They freed those engineers to do more strategic aspects in the business. Things like looking at what data they have out there, how they start making it more efficient for AI, and getting it available to get more and more value from the data. Evergreen//one takes a very similar concept.
It is our storage as a service offering that takes true SLAs. It is one of the most differentiated offering in the market today because of the fact that we commit to these SLAs to end users. This intent follows all the way through to consumption. Consumers now can say, "I declare my outcome. I want a certain performance range. I want a certain resiliency. I want a certain efficiency factor." And it will deliver that to end users as an SLA that the system commits to holistically. That then, as the user scales, it allows them to be able to pay for what they consume as they continue to grow. If you pair this together with Evergreen as an architectural approach, as we talked about earlier, Evergreen allows you to remove the need to worry about the technology. Think about this for a sec.
If you get upgrades to the newest generation hardware, performance gains, density gains every three years as part of this subscription, you do not have to worry about what technology is going to evolve to over time. You can take the guesswork completely out of it. Evergreen//one , though, then takes the guesswork out of the actual consumption itself, and so you do not have to predict the demand anymore. You can just have it actually scale alongside that environment, reducing the complexity and, of course, risk that comes with procuring systems as a whole. All right, so now that we have gotten the unified data plane, the benefits of that, how we are applying intent and autonomy, let us talk about what we are doing to optimize AI workflows to build in better efficiency and truly activate the data for our customers.
The first thing when you think about it, the nice part about a unified data plane is you are running enterprise applications on that exact same infrastructure. We talked about this earlier. We do not think customers should create another silo. This should run on the exact same thing because likely it will run on the same data that is already running the enterprise itself. What is even more key here is that multidimensional performance within the unified data plane allows them to be able to run AI workloads even more efficiently because it is both on the same infrastructure stack, it has the same policies that can be applied, and it does not require additional copies or silos to be built. Second is we started to figure out ways for us to activate data.
The first thing that most enterprises look at is, how do I take the data that I currently have today and make it so that it can be used for retrieval augmented generation, so that when I create prompts, it can actually give me responses based upon the data that is true to my enterprise today? What Data Stream does is it allows us to activate the data. It vectorizes the data, discovers where those data sources are out there, provides the vectors as part of the retrieval process so that it can actually access direct enterprise information and be able to apply governance directly to those vectors as well. When it comes to efficiency, we realize that the tokenomics and optimizing for tokens are going to be a big focus as time goes on.
The first thing we thought about is how do we make sure that people do not continue to generate off the same exact series of prompts? What we did is we created what is called key-value acceleration inside the product, which goes and keeps a history of prior generations, which allows for faster inference, 20x faster inference, and even more so in some testing that we have done. It allows you to reduce the amount of GPU utilization per run and inference, which allows you to utilize the GPUs more often, and it optimized for overall token cost, which becomes a huge benefit. We are actually leveraging much of this internally as we are building our own AI stack to allow us to be able to increase efficiencies in the way that prompts come in. Certain prompts do not need frontier models that are sitting out there.
Some can land directly on open-weight models sitting locally on our own systems, on our own data. All right, so we talked about quite a bit here. We talked about three main areas of focus for us that are helping to drive significant value for our customers. The first is around unifying the data plane, which gives us the ability to continue to grow the platform, continue to add new workloads, and be able to take AI and run it directly on the same data sets, which just drive more and more compounding value to our customers. The second big area is talking about scaling through software. We are taking our software capabilities within Fusion and our Intelligent Control Plane and allowing us to actually help govern data and control it.
In this new world of AI, this becomes paramount because you want to make sure that policy is right next to the data so that agents are aware of the intent and understand whether they should or should not access it. The last part is around optimizing the data itself, where we try to improve AI economics. We enable users to get faster inference, activate the data that they have today, reduce the amount of duplication that's sitting out there. That's a quick update on just the core platform. I'd love to invite Rob Lee up next to go through Scale AI. Thank you.
Awesome. Well, good morning, everyone. Great to see a lot of familiar faces. I'm going to pick up where Chadd left off and bring a wrap to the morning's technology sessions, by diving a bit further into what we see happening in the AI space. Chadd talked a lot about where we see AI being deployed in the enterprise, how we're meeting that need. We'll talk more about that in the afternoon. I want to switch gears here a little bit and take this up in scale by 10x, by 100x. Talk about how we're serving the needs of the Neoclouds, how we're doing this with FlashBlade//EXA , which is our flagship product in this space. If we step back from this and we take a look at what we see happening in the AI space, a couple things, right? Certainly, it's growing.
We're seeing also now a mix of new workload types coming to bear, and this makes sense. Years ago, it was all about training, about building frontier models, who can build a better model with more parameters. As those models develop, as the harnesses develop, we're now seeing a much broader mix of use cases being deployed. Yes, the training, but also then putting those models into production in inference, in agentic deployments, so on and so forth. What does this mean? This means now we have a continued explosion of AI deployments. You see this in terms of GPU deployments. You see this in terms of the high-performance storage attach that goes with that. You're also seeing a much broader spectrum, if you will, a much broader dynamic range of different workload types coming to bear. It's not just training anymore.
It's not just one workload type. It's all of these things coming together. There's so much of it that the enterprises can't house this internally themselves. They can't deploy all the GPUs themselves. They're not going to. It's more than the hyperscalers can deploy. What you see is you see the Neoclouds filling this gap. What you see then is the Neocloud serving as a point of concentration, a point of consolidation for all of these workloads all coming together into one. A lot of coverage around the Neocloud space in the market. A lot of the focus in this coverage has been around GPU deployment. Who can get the GPUs? Where's the supply? How much is it going to cost? Can I get the power? Can I build the power plants and the buildings and the data centers? That's super important.
But actually, for the Neoclouds, what is more important is driving GPU utilization. GPU deployment is about spending money, and GPU utilization is about making money. This is about how they charge for GPU hours, tokens per, whatever the metric is. Here is where storage has emerged as a significant bottleneck. The traditional approaches to storage have emerged as a bottleneck to driving GPU utilization. That could be because of performance limitations. It could be because of the inability for traditional storage approaches to be able to adapt to the mix of workloads now between training and inference and agentic deployment. It could be because of the lack of ability to adapt to the multiple different types of data that is being worked with. It is not just LLMs and text anymore. You are mixing this with images and video. Stability and reliability.
That is the easiest way to drop your utilization is if your infrastructure is not available, is not reliable, then the whole thing stops working. As we have engaged in this space, as we have worked with Neoclouds, this has emerged as one of the biggest challenges that they have to face once they deploy the GPUs is, hey, how do I make these things actually utilized to where I can go and bill for them? Interesting data point. There was a third-party study done that showed that AI clusters across the board run generally below about 85% GPU utilization because of storage bottlenecks or reliability or availability issues. So why is storage so hard here? It comes down to a couple things. Number one, there is a lot of data.
Whether you are talking about training and massive data sets, whether you are talking about inference, which produces and consumes a lot of real-time data, whether you are talking about agentic deployments, which certainly produce and consume lots of data, but then also need to be audited and governed. There is a lot of data moving very quickly, and it all has to be located close to the GPUs. Chadd just talked about how the last thing you want to do in the enterprise, much less if you are a cloud, is build another silo just to serve this data need. That is a big challenge. When we look at the type of data I talked about, the mix of workloads that are now coming to the forefront. When we look at the type of data that is being worked with, it is also changing.
It is no longer just text prompts, "Hey, help me plan my vacation." It is correlating that with audio, with video, with images, multiple modes of data all being worked with at once. This is where a lot of the traditional approaches that have grown up in the traditional HPC or High-Performance Computing era really fall down. If you think about traditional HPC, this was a set of technologies developed for particle accelerator labs and nuclear simulations. Things that you plan for two years to do an experiment. Experiment runs for a day and then you are done. Simulations. They happen in human time. When you think about how a Neocloud runs, you have got training, you have got inference, you have got agentic workloads all running for multiple users all simultaneously. This stretches traditional high-performance computing approaches and products well beyond their limits. Reliability. Sometimes people forget that Neoclouds are clouds. Clouds have SLAs.
They have obligations to their customers. When the infrastructure doesn't work, those SLAs have penalties, much less not driving utilization. There is significant financial penalties. These are operations that run extremely lean. They don't have large storage teams. They need storage that's just going to work, even in the face of incredible demands. I talked a lot about how the variety of different workloads is driving different types of data access demands. There are some workloads that just need absolutely mind-blowing performance. We see this in training, large-scale training environments with checkpoints. As you're training these large models, this could take days and weeks. You need to take checkpoints along the way to make sure that if there's a blip, you don't have to repeat that whole process. That ends up driving a ton of bandwidth requirements.
You've got a wide variety of performance needs all colliding into one. It's all happening at the same time. It all has to be super reliable, easy to operate, and by the way, it needs to be secure and governed and compliant. There's a lot going on here. What do we do? Let me start with what we're not going to do, which is build a brand-new product from a clean sheet of paper to go solve this. As it turns out, we're working from a solid product and technology base that allows us to go meet these needs by repurposing and repackaging a lot of the core technology we've developed over the last 15 years. We start with FlashBlade.
We take a look and we say, "How do we scale that up to meet these needs 10X, 100X?" We look in three key areas. We look at how do we go meet the performance needs at the scale. We look at how do we extend the flexibility of this core platform to be able to scale not just at the enterprise level from terabytes to petabytes, but now into exabytes and tens of exabytes. We extend that predictable linear performance scaling to the needs of the Neoclouds, and then we lean on our track record of mission-critical reliability, and ability to deliver a predictable performance without bottlenecks across the board. One of the key aspects here of what we've done with Exo is we've taken the strong roots that we have in FlashBlade//S that we've proven out in the enterprise, and we've disaggregated that architecture.
We've taken that core product and we've said, "Hey, we've built a software base that gives us a very solid foundation to provide one part of the performance puzzle, the metadata piece." We're going to go decouple that software. We're going to go pair it with a more open hardware ecosystem, and now allow us to provide the utmost performance in both metadata and data. This is the secret, this is the key to meeting that wide dynamic range of Neocloud needs. We've tossed around the term metadata performance and data performance a lot. Perhaps it's useful to draw a quick analogy. If I went and got a book and handed it to one of you and asked you to read a page, it might take you a minute, two minutes to read the page.
If I hand it to a speed reader, somebody who can process data really quickly, might take them 20 seconds. Okay, well, that is a lot faster. That is great. Now, if instead I pointed both of you at a shelf of books and said, "Go pick the third book off the top shelf, go to page 17, read two sentences, and then go pick a book off the fourth shelf, turn to page 237, read a couple sentences," so on and so on and so forth. You can see that most of the time you would be spending is actually looking up the words to read, not reading the words.
When you look at the performance required for AI workloads, you need that balance, that mix of being able to process the data really quickly, read the words off the page, but more so to be able to process that metadata quickly, be able to find what data is relevant to then go process it. We are uniquely positioned to bring both types of performance and to be able to scale both ends of that performance independently and predictably. The other thing that sets us apart is we are able to do this without putting a bunch of onerous requirements on the user. If you look at other approaches that try to do something similar in the industry, they come with fairly onerous requirements.
Like, okay, well, we can maybe give you better performance, but you just have to install this little piece of software on the GPU server. Okay, it does not sound too bad. If you are running one, 10, 15 GPU servers, that is not too bad. What if you are running 100,000? It sounds pretty terrible. We are able to build on open standards, extend our core technology, which we have perfected in Purity since day one, to provide this unique mix and balance of metadata and data performance, exactly what is needed to meet these workloads. We are able to deliver it on open standards without a ton of operational headache and complexity to the end users, and that is really what sets us apart. We will dive one click deeper into this. Since Paul always likes to remind me that our differentiation is in the software, I want to hammer this home.
How do we solve that problem, right? I highlighted how metadata performance is really important. Data performance, I think people understand. Why is this a hard problem? Well, it comes down to the fact that organizing the metadata, when you build storage software, organizing the catalog of what files sit where, and where on the media should we go to read this data? Essentially, in my analogy, how do we look up the pages to go read? This is a hard problem to begin with. It is a hard problem to scale as you scale capacity and keep performance. The organizational methods for doing this, frankly, in the industry are quite antiquated. This is a hard computer science problem. The good news is we have a building full of people that love solving hard computer science problems.
And what we've done is we've built the core of Purity since day one on what we call a key-value store, a key-value database. This is essentially the technique introduced largely by the cloud providers to scale without limit and with extreme predictability and with extreme parallelism. So we built the core of our Purity operating system, whether it's in FlashArray, whether it's in FlashBlade, whether it's in FlashBlade//EXA, based on this key-value store, plus a set of techniques, a set of approaches, I would say, that we've adapted from the distributed database community.
If you net all this together and you look at the software that underlies our products or platforms, whether it's FlashBlade, FlashArray, FlashBlade//EXA, what you'll find is a highly performant, highly scalable, without bottlenecks architecture that is able to scale very predictably and is free of bottlenecks that has served us well in the past. But as we look forward, and particularly in the Neoclouds, at the changing workload demands that we're seeing, positions us very well and positions our customers very well to be free of changes in the future. All right. So we start with a strong software base, and then we pair that with unique hardware to bring out the best attributes of that software.
In the enterprise with FlashBlade//S, we pair that Purity operating system with a fully vertically integrated hardware environment that accelerates that metadata engine, accelerates that data engine while keeping enterprise-level simplicity, integrated networking. Kind of the perfect package to meet the enterprise needs of size, performance, scale, manageability, and networking complexity. As we move to the Neoclouds, as we scale up 10x, 100x, we expand this platform from a hardware perspective. Same software. We decouple the software, but we now pair our FlashBlade hardware, which accelerates our metadata tier, with purpose-designed hardware to serve our data tier. Basically opening the pipes, allowing for much faster data transfer and two axes of scale, ability to scale performance at metadata layer, the ability to scale the performance of the data layer, and the ability to scale capacity almost without bounds. So how does this add up?
What does this end up delivering? Well, to start with, number one in performance. We top the performance charts across the board as we look at both the top-line bandwidth numbers, as we look at the industry benchmarks, which I'll dive into in a second. We're able to serve Neocloud needs at massive scale. Whether we're talking about thousands, tens of thousands, up to hundreds of thousands of GPUs. And we're able to do this on the same technology that, again, in the enterprise, we see starting at the tens, hundreds, and as we scale into the Neoclouds, up to three orders of magnitude. And at the end of the day, when you net this out, I started this conversation by talking about GPU utilization.
When you net this out with a higher performance, with a greater reliability, with the ease of operation, the fact that operators aren't having to go around configuring custom software and spending a ton of time managing the storage layer, we're delivering a significant reduction in TCO for not just our enterprise customers, but now our Neocloud customers as well. I mentioned number one in performance. MLPerf is one of the leading benchmarks in the space. These are recently published results in the newest round of MLPerf 3.0 benchmarks. Key things to call out here. This is the standardized testing configuration. Number one across the board in all configurations tested by a wide margin. The previous version of MLPerf, we were about 2x the nearest competitor, and what I can say is that in this most recent version, some of our competitors declined to submit.
That should tell you something there. All right. Let's take a look at a specific example from a Neocloud deployment. STN, a U.S.-based Neocloud, is one of our first customers for EXA. When they started their issue, where they started off to the left was they had obviously a large environment of GPU clusters. They were having to serve multiple AI services on top of them. They're essentially a GPU cloud. Their customers are AI service clouds that have their own service offerings, doing a mix of training, doing a mix of inference, lots of data. They couldn't keep up with the performance. They had large multi-petabyte file object stores, couldn't keep up with the performance, specifically around metadata. What did they do? They went and found a third-party product, separated the metadata, ran a separate cluster to go manage that metadata.
Now they've got one thing doing the metadata, another thing doing the data. Okay, still not fast enough. Had to put in another product, a caching layer, more hardware spend, more complexity to sit in the middle just to provide good enough performance. At the end of the day, they were left with not that great performance, a really complex stack, third-party dependencies, and way more engineering effort and time being spent on this than they wanted to do. They brought in FlashBlade//EXA . They actually tested our solution in our labs on a Thursday, placed the order on a Friday. We had them up and running on a Monday night. They were able to simplify the stack dramatically, take out all of these extraneous layers, improve their performance dramatically, 3x and beyond.
Going from being topped out at 30 GB to 40 GB a second, 100 GB a second per node with predictable scaling, and they've now grown that environment in multiple environments, multiple fold. At the end of the day, when you ask the CEO of STN, and he's spoken on our behalf before, what he said is significant acceleration in storage performance. But what matters the most is the acceleration of the end-to-end inference and training workloads. 10% to 20% faster end to end, that includes the GPU time, and most importantly, a big uptick in GPU utilization. All right. With that, I'm going to bring the morning's tech sessions to a wrap.
I think hopefully we've given you a sense across the board of all of the things that we're doing with the newer product areas to meet the needs of AI, whether it's driven by a new approach to managing and looking at classifying understanding data, how we're bringing that together on a unified platform, simplifying the management of that with autonomous policies, and then accelerating AI use cases in the enterprise, and then now taking this to the Neoclouds. We're going to transition to a Q&A. I'm going to invite Chadd and Prakash and Ashish back onto the stage. As they're setting up, I'll just remind everyone that we're going to have an afternoon session going deeper into the company strategy and the financial outlook. This morning's Q&A really should be technology and product focused.
Also would ask just for the folks online that as we pass the mics around the room, before the question, if you could just introduce yourself and your firm, that'd be helpful. Jason?
All right, let's do it.
All right, Jason Ader with William Blair. I got a couple questions. The first is just how do you think about the point solutions out there for like DSPM, data discovery and classification, governance? Are these guys competitors or partners? Secondly, on the Neocloud side, can you just talk about who you're competing in those deals with?
Yeah, I'll start. 1touch.io traditionally was working in the DSPM space as a DSPM tool. Yes, it competes with other DSPM providers. The major differentiators in that category still remain the broadest coverage, like covers everything from mainframe to SaaS tools, et cetera. A lot of the traditional DSPM tools were built primarily security-minded. Regex classifiers, this is a pattern match. But it could not tell historically whether. We've tested that when we were actually looking at 1touch.io, we've tested a lot of products as we were evaluating. One, two, three, four. Is it an order number or an invoice number? Most things would fail on that test, right? So the context in terms of the approach was more built around context and knowledge graph, where security was a use case in the kind of technology that we looked at.
We saw the applicability to broader than just the DSPM category. I will let Ashish elaborate.
Yeah. I will just add a little bit to it. We have got an architectural advantage when it comes to looking at data. It starts with the fact that we do not have to sit in the workflow to understand what the data is doing. We go to the data as a source and create those knowledge graphs and the relationships, which allows us to get to deployment a lot faster, allows us to get much higher accuracy. It is actually much more scalable because you are not waiting for the workflow to happen. You are actually going right to the data and delivering the intelligence right there. So in addition to more coverage, much better context, we also have an architectural advantage to deploy at scale and deliver accuracy.
Can you replace all those point solutions?
Well, the whole idea behind the universal Data Intelligence layer is to provide this intelligence as a Lego plate on which you can put the various different Lego blocks. DSPM focuses on security, but the same intelligence allows you to do things on governance, compliance, and then also the contextualization enables you to do AI-ready data as well.
Yeah. One example you would see is most of those solutions is like, here's an audit report for DSPM. Yes, this product can produce that, but it also provides an MCP for a query, what data should be able to go into this agent model. So the products that were traditionally security-minded and DSPM-focused were built for very narrow use cases, and this is more broad in terms of all entity types, not just security use cases.
Jason, to take the second part of your question on the Neoclouds, what I can say is we typically don't, in the Neoclouds, compete with our typical enterprise competitors. We see some folks that have grown up, I would say, out of the traditional HPC space based on technologies like Lustre and so on and so forth. We see some more niche players that have focused, more recently, deeply on the Neoclouds. What I'd say that sets us apart from these players is neither of them are able to address the wide set of performance needs that I articulated between data and metadata performance, needing both of those at the same time. As well, increasingly, I would say the focus, as we have conversations with the Neocloud customers, is on the gaps in the competition as it pertains to reliability, operational simplicity, and availability.
That is becoming front and center. Much less security, by the way. But those aspects are becoming front and center as these Neoclouds are, I would say, graduating into truly mission-critical workloads that have compliance, that have governance, have other requirements around them.
Hey, guys. Thank you. Erik Woodring, Morgan Stanley. Rob, I appreciate the color that you shared on FlashBlade//EXA there, and maybe building off of that last question, I would love to know what do you think has to change for Everpure to become, let's call it the default storage platform for the next 10 to 20 major Neocloud build-outs? What changes for you to become the de facto rather than trying to get into that market, so to speak?
Yeah, Eric, thanks for the question. I think one thing is, to be perfectly clear, we were not first to enter this market. We had focused on enterprise AI. We, I would say, made a very intentional and concerted effort to adapt our technology for where we thought the needs of this market would head. We are not first entrants. In many cases, it is about going and engaging in—and we are engaged with pretty much all the top Neoclouds. As you might imagine, these are longer sales cycles, very technology-focused. I think it is a matter of time seasoning and realization that these environments are super high-performance demanding across a multitude of dimensions, and they have all of the mission criticality requirements, and we are the only vendor that can meet those needs.
The other thing I would say that accrues to our advantage, and what we are seeing really good signs of already, is people, again, forget that Neoclouds are clouds. Clouds have other storage needs beyond high-performance GPUs. They have things like IT departments. They have things like virtualization needs for the user sessions. They have file storage needs, object storage needs. They even have backup needs. Our ability as a holistic vendor to go and meet the needs of these Neoclouds, whatever door we can get into, allows us to penetrate and accelerate the conversations around FlashBlade//EXA and the high-performance storage in ways that nobody else can.
Thank you.
Thank you. It's Mehdi Hosseini, Susquehanna International. It's a question for the team. When I look at your slides and think about application, it's mostly a GPU-based accelerator, open source, which makes sense given your product portfolio. As you look into the future, what happens when we actually have more of a custom ASIC solution, which seems to me would require more a proprietary operating system. The question is, am I right to assume that right now your data management solution is mostly applicable to open source GPU base? If that's the case, how do you extend that to more of a custom ASIC proprietary operating system?
Yeah, maybe I'll start with that one and others can jump in, and thanks for the question, Mehdi. I think what's important to realize is everything we've talked to you about this morning really is driven by software, whether it's Data Intelligence, whether it's what we're doing with Everpure Data Stream and the organization of data for AI, and even EXA, what powers EXA, is software driven. We've taken a very intentional approach to design that software not to be specifically locked into any one execution platform, whether it's specific CPUs, GPUs, or accelerators. We obviously keep abreast of new hardware developments in the market. But as it pertains to some of the areas you mentioned, whether it's Data Intelligence, everpure Data Stream, we are designing those layers and those modules, the Lego blocks, as Ashish mentioned, in a very modular way to be able to run on any hardware platform.
They'll run great on CPUs, they'll run great on GPUs. If some new ASIC comes along that makes it run even better, great, we'll run on that. Everpure Data Stream, the same way. We developed that in partnership with NVIDIA on their AI data platform. We're working with AMD. We're working with other GPU vendors to run that same technology. It's not in our benefit to lock our software to any one hardware platform. We're going to make that work on the best hardware platforms that exist today and what we see coming in the future.
Just a quick. Just to follow up.
If I just think about your unified data lake, unified data plane, it sits either below or on top of the operating system, let it be open source or more proprietary. I understand that you do not want to be specific, or you want to be independent of silicon, but what about the operating system? How is a unified data plane would work with operating system?
I can jump in on that one quick. If you think about what the unified data plane does effectively is provides open protocols for storage. We do block file as well as object in one common mechanism. So no matter what the hardware is on the other end of it, they are receiving that type of protocol stack. And even within block, we have various ones of a fiber channel and NVMe over Fabrics and over TCP, as well as other ones that are out there. So I would say we follow pretty much the networking connection as well as the things that operating systems will consume storage around and provide that as one unified view. And I think one of the benefits here is that we have products that actually cover multiple protocols.
FlashBlade, as an example, does file and object, to where FlashArray does block, file, and object together. The control plane starts to become the management componentry to decide where it actually lands based upon the SLAs and intent that the end user defines. And so it rolls through the operating environment no matter what the hardware system looks like.
Cool.
Steve from Morgan Stanley. Paul keeps telling us that you are a software company, and there are a lot of other software companies out there, and many of them also are very interested in owning the data governance stack and cetera. It is sort of the right to contest the market. But there is also, what is the technology dividing line where you go up the stack, there are people trying to come down, Snowflake, Databricks, and Okta , all the security guys. So who does authentication? Where do these policies come from? And I guess I am trying to. We can either construct a framing where it is a battle royal, where you guys are slugging it out there or you guys can find some kind of a clean place where you do this, I do that, and everybody is happy. How does this all evolve? I agree it is going to evolve.
The key question is what's your right to contest? Where do you define a line? What is your relationship to those guys?
I will start and then you can weigh in. Largely, I think, and I talked a little bit about it earlier. A lot of those vendors come from a different point of leverage. Their point of leverage. Oracle comes from the database point of leverage. Databricks comes from the analytics point of leverage. Salesforce and SAP even come from an application point of leverage. The data management ecosystem has been broad and grown up traditionally for a really long time. Most of them have largely not come from solving the problem needed for AI today. All of them are trying to build capabilities. Bring your data to me and I will do an ontology and knowledge graph on things in my ecosystem. Just one, philosophically, we are approaching the problem very differently. We are saying, "No, your data is not going to go into one of those ecosystems.
It will be all of them. It will be Databricks and Snowflake and SAP and Oracle, and your data will go and be in multiple places." That is this concept of data will be distributed. Ontology and knowledge graph needs to be in middleware above the layer. I will use an example of Microsoft has capability in their platform with Purview. They try to do it for Microsoft data. We have got customers that are doing that, but then also use our Data Intelligence technology because Microsoft cannot look outside. This idea of if you believe data is distributed and you need to work across multiple ecosystems, we provide a neutral way to work across data wherever it sits. We are not opinionated on it needs to be in Databricks or Snowflake or anybody.
All of the other vendors have an opinionated view of if you want any value, first give me all your data. Philosophically, we found that that resonates with customers, very differentiated from the market. In terms of are we slogging it out? Are we partnering? We are having a lot of conversations with some of those vendors saying they are interested in partnering with us for the things that they cannot see or do. There might be co-opetition in the direct, "Hey, everything needs to be here." As they start accepting that they do not have ownership of all of a customer's data, then they need this external view, and we provide that external view into a customer's end-to-end landscape. From a customer lens, we mainly stay focused on customer value. We have always done that from the beginning.
Customers need total visibility into their data wherever it sits. They have multiple vendors. We see that in almost 100% of our install base. We provide that so they can remain in control and get value out of the data where it sits. That is largely how we are approaching it, but feel free to weigh in.
Yeah, I think just to add onto that, two things. One is, I think the two-up slide that Prakash put up puts it very well, which is we are pursuing an approach that seeks to add value on top of those platforms. That value largely is in building the connectivity, the universal context that allows customers to more usefully connect data across those sources, whether it is SaaS vendors or the likes of Databricks and Snowflake. Another example, we have a partnership with Databricks today. That partnership is allowing customers to connect the Databricks processing engine to FlashBlade, to data stores stored in FlashBlade on premise. I think it is a good early example of what Prakash is talking about. That partnership is driven by the fact that customers have these data lakes. They, for some applications, benefit from Databricks' capabilities. For other applications, that source of data needs to be connected to other parts of their enterprise.
They need to control that thing, and they want the flexibility to be able to burst and say, "Hey, for this particular task or analysis, I want to use a Databricks engine. For this other set of needs, I want to use a different application environment." As we zoom back out and we think about our vision of data primacy and how this plays out in the future, I think this is a great example of first of many steps that are going to take place, which is to say the true store of value is in that centralized data thing. Then, the application or platform stacks over time evolve to who can do a better analysis job, who can do a better process encoding for financials stack or an ERP stack separate from the data.
Also be able to optimize the workflows to it. There's a lot of benefits for us to build good partnerships and efficiencies in the way those processes access our data. Very similar way that we did early days with VMware for virtualized environments. We made it very easy for customers to be able to embrace the technology. We're going to do very similar mechanisms there as well.
Excuse me. Thanks for doing this. Howard Ma with Guggenheim Securities. I want to ask a higher-level question, but I also want to tie it into the prior question. First, the higher-level one, which is when you talk to CIOs and CDOs and CISOs, I imagine when you offer this unified approach or discover data wherever it sits, provide the context and the governance, you get very little pushback to that concept. But I'm sure you get inertia because there's a lot of existing storage systems in place already. So question number one is, you're in 2,300 customers today with Fusion/EDC. Can you give us a sense of how those combos have evolved? Then the tie-in to the prior question is, as they consider Everpure with the existing tech stack, I kind of want to summarize it.
It sounds to me like, obviously, the database layer still needs to stay. Whether it's OLTP, analytical, apps, some parts of security like IGA, authentication, it seems like those are crucial complementary, but it sounds like you're saying data integration like MDM, DSPM, maybe even inference gateways, there are categories that if you sum it up, it's probably tens of billions of dollars of TAM, that that is what you're directly going after, and that's what these C-level people are considering replacing with Everpure.
Yeah. Maybe I'll start this long compound question.
Yeah.
So a couple of things. Absolutely, as we talk to CIOs, CDOs, and I would say decision-makers higher up in the stack, not only are we not getting pushback, we are actually getting significant pull. I have been on the road, as have the rest of the management team, since the acquisition, and I will say every single meeting that we have with a high-level customer prospect, the first topic and really the thirst and demand is for a lot of these capabilities. One thing I would correct, maybe, or modify in the question is, important to realize that as we deliver 1touch.io, as we deliver on that roadmap into Data Intelligence, these are a set of capabilities we can offer on top of anybody's storage.
Right.
The existence of other storage infrastructure in place does not get in our way. Yes, the benefits of having 2,300 customers on Fusion and growing, the benefits of having 15,000 customers on our core platform will accelerate that. But this set of buyers is very much needing and wanting and thirsting for better control, understanding of their data, where it sits, for all the reasons that Prakash and Ashish outlined. Then I think to the second part of your question in terms of, hey, as we look forward and we look at traditional markets, if you will, of data management, whether it is MDM, cataloging, so on and so forth, what I would say is that we, in many ways, just like we have done with storage infrastructure, we are approaching the market needs with a little bit of a different lens. We are not just seeking to reinvent and replace existing markets.
We see an opportunity, with technology from 1touch.io, with new technologies coming out of AI, we see an opportunity to fundamentally disrupt and do that thing better. If you think about a lot of why, and Prakash set this up well in the beginning of his discussion, if you think about a lot of why the enterprise architecture problem has been difficult in the past, it is because of the seams between all of those pieces, the MDM, the data catalog, and the integration and the ETLs, and a lot of that can go away. Our mission is to make most of that go away, and to go away in service of separating out the data layer, having that become primary, and then relegating the application stack to becoming a more stateless encoding of business process over time.
Yeah. Look, just to build on it, and I am going to go back and probably date myself, but I was around when at SAP we had something called R/2, which was more mainframe, less client and not client-server. Right? I was there when we were moving to client-server. I was there when web and internet technology kind of came about. We have had platform modifications and cycles of how we build applications over time, and I actually do think this is a long-term fundamental shift in terms of a way you build applications. In the same way we now assume client-server or web microservices, I do think a data primacy architecture will be a 10 to 20 year super cycle. I do think that is a new way of thinking. I do think AI has forced it.
I think it has created an inception point for us to think about it differently. As Rob said, we are not chasing things, and we know things do not change overnight. But today, when we build an application and we think about how customers are going to build applications. The idea that I am going to buy a ton of applications and integrate the data, versus AI needs markdown files and context and encodes source and config and logic and pricing rules and business logic as data. The business logic as data is a superior approach for building modern applications. This might be just a long-term frustration of mine saying, we never really got to ERP because the application world became fragmented.
I actually believe if the ability to build an application gets shrunk down by AI development tools, because software engineering is one of the things AI is getting good at, then the cost to build workflows comes down and the cost to maintain data coherence goes up. So by starting with data primacy and Data Intelligence right now, you can create a baseline of data coherence. I do think that is a way new startups today will build applications that way. New companies will. The long enterprise and everything that we have today will not change overnight. But the simpler the app, I would say there is a category of applications, and I will not say who these companies are, basically a web form on a table. I do not know that you need to buy those apps.
The category of applications that have deep embedded process logic, probably not the first thing that is going to change. So I do think that as this thing evolves, Data Intelligence is the first step. Looking at the data where it sits, understanding the relationships, and ensuring that both analytics apps and agents can work off that data with shared context is a first step, and that is why we are taking this step. It is not the last step. Data Intelligence is our first step in this journey.
Yep.
Thank you. Thank you for doing this. Asiya from Citi. Just a quick question. On the adoption that you talked about, I think you said 23 customers, 2X adoption in the last six months. Are these existing logos that you are already present at? Are you seeing this, and I am sorry if I missed this, or are these new customers that are approaching you given the problems that you guys have talked about?
Definitely.
The opportunity should be pretty huge if it's not already in the installed base.
Yes.
Thank you.
2,300 customers. 23 would be a little bit small. It's both, right? Existing customers obviously see significant value out of it because they're helping to build intelligence on top of the existing infrastructure. I'd say the benefits on the existing customer is they're starting to see this experience shift and looking at their other environment saying, "Why do I continue to do what I do with those other older legacy products?" You're seeing a push towards, I want to move to that new experience and a consolidation around this kind of unified data plane model. On the new customers, we've been talking to them about this experience from the get-go, so we start them off fresh in that manner. The game here is that you can have conversations that orient them in a very different direction.
You're starting to talk to them about business processes and requirements and what intent that they have as a, call it first principle to start with. On the existing customers, we actually have the benefit of telemetry, and we have lots of data that comes back from each of these systems. One of the first thing we're doing is helping them rationalize what do you do today? What's crazy is most people say, "Hey, we only have three policies for this," and we can show them they have 55 policies, and that's the problem of lack of consistency. The first benefit we get is actually being able to help them align business requirements and policies and then apply that to the infrastructure stack.
The nice part here is when they look at the other environment that they have, they realize, "Oh my God, I can just imagine how much of a mess that is. I would really like to get more consistency and governance built into the system." Hence as those things come up for refreshes, lots of it is landing back on this new experience. So both existing and-
One of the things I think I shared with you this morning when we were speaking just over coffee was this idea that Pure has consistently taken share and market over time. I think Charlie may talk about that a little later today. As we take a look, some of the main drivers of that, our SLA approach that Chadd talked about is something that is bringing more new net, new logos, given the unique differentiator we have to the competitive market space. The Data Intelligence approach, also, we see signs of opening net new logos for our kind of growth business. So we've always done well in taking share as a baseline, but those areas of the business have a higher net new logo concentration of a front door because it broadens our value proposition into the market.
Hi, this is MP on behalf of Joseph Cardoso from JP Morgan. Just a quick question. You talked about different architectures across enterprise solutions as well as the new cloud. Any differentiation or what's really different in terms of offerings or overall solution for hyperscaler customers? Thank you.
Yeah, I'll take that one. We'll dive a bit deeper into this in the afternoon session. But yes, I would say if we look at the different levels of scale and customer segments between our typical commercial enterprise customers and their needs, balance of performance scale, simplicity, size, et cetera, what we're seeing in the Neoclouds, and then now the hyperscalers. Same technology that underpins all of those solutions in terms of the software base, in terms of the IP. But in terms of the configuration and the offerings, we've specialized those over time really for those three segments. I'll dive deeper into the hyperscale solutions in the afternoon session. But really the takeaway point is it's all based on the same core underlying technology and really packaging and integrating that into each of those environments, which have very different needs, right?
If you look at the enterprise, they typically want a one-size-fits-all, put it in the corner, plug it in, and then the thing just works type of solution. As you go 100X in scale to the Neoclouds, they are now delivering a cloud service. They have SREs, they have operations teams. They are operating at a much larger scale, have different set of requirements in terms of performance, efficiency, how they go and manage systems. Then if you upsize that again by 100 or 1,000X, now you are talking about hyperscalers. As you might imagine, each of those orders of magnitude carries with it a very different environment and very different operating requirements.
Great. Thank you.
Thanks, guys. Matt Calitri from Needham. When you think about some of the earlier adopters of Fusion and even EDC, before you introduced this unified data plane, is Data Intelligence something that they were building out themselves, or is it just that new of an issue? Going back to why is Everpure at the right layer to approach this from, when you talked about it is a fundamentally different approach from a Snowflake or a Databricks. How does that change now that they have dove into interoperability and Iceberg tables and things like that?
Yeah. Okay, I will start. So two parts to that question, right? The first part of was Data Intelligence typically needed in the past in building EDCs? Typically from a traditional storage lens, I would say no. Prior to AI, storage teams managed north-south applications. This is my Oracle, this is my SAP. It was sufficient, right? Our Enterprise Data Cloud was built around intent and policy management, and the 2,300 customers that Chadd talked about were all pre the Data Intelligence era. Why Data Intelligence became important is when previously the predominant paradigm was application centricity. Everything was buy an app, deploy an app, right? That was most of what was going on. When agents start working on data directly, the paradigm shifted. So now you need to look at and manage and understand that data. You need to contextualize that data, right?
From a storage management lens, that means, like I said earlier, if you have an agent and you need high availability for it, so it can be resilient, that is a storage problem. It is uniquely a storage problem, right? In terms of how you can offer that capability. Doing the data management of that thing requires bringing that data context closer to storage. That is the leverage point. That is the uniqueness we see in terms of why A plus B is uniquely differentiated. Coming back to your. So why Data Intelligence and why now AI? It was just a change in inflection point, right? It happened faster probably than I think most of us
Yeah
would have expected. The second part of that discussion is that Data Intelligence advantage around why at this layer, why storage, et cetera. There is a more near-term and a longer-term answer. In the more near term, you just want to ensure that, hey, anything that happens physically doesn't break context upstream, right? That is a near-term storage problem where storage tiering, storage mobility, storage things retain context and memory. Otherwise, your AI is spending tokens retraining and learning on that data, right? Longer term, that leads to more persistent memory in a storage operating system.
If you do this at higher layers in the stack, you then have different resiliency challenges, different availability challenges, and you end up back in that walled garden strategy I was talking about earlier, saying, "Okay, well, that just means putting all applications. Like every application provider has to build that uniquely, and you fragment the management of that thing.
Yeah.
Maybe just to zoom out and put all of that in context, because we have thrown a lot at you guys this morning. Maybe the way I would think about it in more simplistic terms is the unified data plane, that has been our mission since day one. Build the best storage infrastructure. Do it one way, do it across all of the different storage needs that an enterprise has, and make that a unified experience and very seamless across the board.
When we think about the intelligent control plane and management, that is about saying, "Hey, once we have done that first thing, how do we let customers not manage 100 discrete arrays, but manage that as one pool of resources, like a cloud of storage, if you will?" That then puts us in a unique position to say, "Okay, well, what more can we do now that we are helping customers with the best storage infrastructure, managing it well, tying it to policies?" That now puts us in a position to go add value at the data meaning, Data Intelligence layer, both on our equipment and arrays as well as other environments. With that, we are a couple minutes over on time. I think this brings the morning session to a close. Wanted to thank all of you for the great questions.
I think we are going to adjourn next door for lunch and be back at 12:00. Is that right? 12:00. All right.
Thank you.
Thank you.
All right. Thank you.