NetApp’s CEO on Unifying Data for the AI Era

(upbeat music) - Hello, and welcome to the Tector Struppler Podcast, hosted by Bloomberg Intelligence. In this podcast series, we speak with sea level company executives and management teams about their views on disruption and how it's driving their decision-making strategy. Bloomberg Intelligence is Bloomberg's research arm. Bloomberg Intelligence covers roughly 2,000 companies globally across multiple asset classes, backed by Bloomberg and third party data, which is supported by nearly 500 research professionals. My name is Woojin Ho, and I'm a Bloomberg Intelligence, and today I'm pleased to have NetApp CEO, George Curian, onto the podcast. George, welcome to the show. - Thank you for having me, good morning to you. - Oh, great. Now, for those who don't know NetApp, it's a leading provider of enterprise storage and a bell-weather in enterprise technology. So I do lean in on this company quite a lot to see how the company is tracking or IT spending is tracking. Now, George joined the company in 2011, and was appointed CEO in 2015. A strategic thinker, he has gotten NetApp through multiple storage transitions, from strengthening its flagship operating system on tap to driving the shift towards all flash storage and leading the company to the cloud era. Now, as AI emerges to the next major technology shift for enterprise storage, I'm really excited to have George on the podcast to talk about the market dynamics, and more importantly, NetApp's AI storage strategy and opportunities. Now, George, hopefully that gives you, I did you justice with the intro. I'm curious, what's your elevator pitch? - Thank you for having me. NetApp helps our clients unleash the value of their data to accelerate competitive advantage. We do so by helping our clients build intelligent data infrastructure solutions that enable them to unify their data across all the places and applications that that data might exist to have the right controls, meaning security and guardrails, and to be able to transform it rapidly so that you can apply AI techniques to that data. - Okay, that's perfect. Now, I wonder before we dive into AI, right? I actually want to set the landscape for the audience, and you know, 2015 as you started CEO, so this is, you've seen 10 years and 10 years of technology is almost a lifetime, right? Now, the external storage market was about 33 billion in 2024 and it's expected to reach 40 billion by 2029 according to IDC. AI is driving unprecedented demand for fast-efficient access to massive data sets like you described, and where do you see the biggest storage transit solutions benefiting from this storage, and how is NetApp position to capture that opportunity? - Yeah, I think first of all, you know, there's been one enduring trend about data, which is particularly more relevant even today than it's ever been, which is that to really get insight from your data, you need to unify that data estate. You need to unify it across types, across applications, across time and across all the locations where your data is. We have been the pioneers in building solutions that enable customers to unify their data. The second big transition has been the need for scale and speed in the infrastructure that supports that data. And so we've built solutions that combine high performance all flash storage technologies with advanced capabilities for bigger and bigger scale and higher and higher performance. And so flash storage, for example, was less than 50% of our storage business a four years ago. It's now more than two thirds of our business and we are not only the largest, but we are the fastest growing flash storage provider. - Got it, okay. Now, when I talk to investors about your company, they ask me if you're a hardware company, I actually have to argue quite a bit that you're not a hardware company. You're actually a software company, right? - That's it. - And one of the key products that you managed when you first started off a net up back in 2011 was on tap and you actually spearheaded on tap as a CEO, as the foundation for what the company is today. Tell us a little bit about on tap and how it differentiates net up in this market. - Absolutely. I think your characterization of net up is dead on target. We, our value add is the software that we build that allows us to package technology and systems from third parties, but really build unique value through that combination. This is why our operating margins are higher than the gross margins of our competitors. And so the unique things that we build is the world's most widely deployed storage and data management operating system called on tap. It is not only deployed in all of the world's large enterprises, to host and store a large amount of corporate data, media and entertainment, data, medical images, images from the Mars Rover, you name it. As well as it is the only operating system that is deployed in every hyperscale data center in the globe. And so what it allows us with the richness of the capabilities of that operating system and the reach and scale of the distribution of that platform to give our clients unparalleled flexibility and capability to use their data well. - Okay, so let's talk about that. That's actually a very, very good point, right? Because if we think about the deployments today, so when you started in 2015, it was still a client server type of environment. You had to manage a company going from client server to a multi-cloud type of environment. How are you helping your clients using on tap in navigating hybrid and multi-cloud environments to make sure that there's the single pane of glass of managing the data in this whole disparate world of cloud and on-prem? - Yeah, I think first of all, data is by far the hardest problem to solve in a cloud. It is the thing that has gravity. It is also the asset that malicious actors are after and in regulatory and kind of legal disputes, it is the asset that needs to be protected in a sovereign environment, right? Compute is fungible, it's stateless, it's therefore a short period of time and disappears data as the enduring asset of a business. And what we saw soon after I got to NetApp was the idea that our clients wanted to use a variety of different landscapes to build their applications. They wanted to use cloud, they wanted to combine cloud with on-premises environments for a variety of reasons. And what we gave our clients is ultimate flexibility. So, you know, a client, for example, in semiconductor design, many clients, the largest semiconductor companies in the world have large data centers where they run their semiconductor processing, but they want to use the cloud for temporary peak usage. And we help pretty much all the larger semiconductor companies in the world to do that. The second idea might be that, you know, clients are want to combine certain applications in the cloud with certain applications in their data center. And we help bridge that where you can have, you know, copies of your data in the cloud and in your data center synchronized. So, for example, in financial services, many clients have their trading applications in their data center, but they want to, at the end of the day, grab huge amounts of compute in the cloud so that they can process and provide, you know, the results of the trade in a very short period of time. We allow a lot of different asset management firms in the world to do that. And then there are clients like software as a service providers that want to be on multiple clouds at the same time. And here, what we allow them to do is to have a single software stack that's certified on NetApp, and we help them deploy it, you know, you test once and you can deploy it on all the clouds because we are in all the clouds and we give you massive amounts of efficiency in the speed at which you can deploy new capability to clients. >> So, so let's talk about all the cloud relationships. I mean, clearly it sounds as if it's important to have, I guess, connections to all the clouds. I believe you have relationships with AWS, Azure, GCP. You know, my understanding in the past is that clients tend to use one cloud but has that become increasingly important to have all of those relationships, to have that relationship and does that help differentiate you from others as well? >> Yeah, very simply, you know, two reasons for serving a broad range of clouds. Not only do we work with the hyperscalers, we also work with a lot of sovereign cloud providers, for example, in parts of the world where they have concerns about putting data on American clouds. I think that with the hyperscalers, listen, most clients use a combination of cloud providers for their application portfolio. They might choose, you know, one cloud for their enterprise apps and another cloud for AI. They might use one cloud for software development and another cloud for, you know, operational, you know, capabilities, right? And so for us, it was important to work with all three of them. And then the second, and that allowed us to have a much stronger value proposition for our end client. And then the second was, you know, frankly, when we work with all three of them, it sucks the oxygen out of the room for any of our competitors to get a foothold in the cloud. >> Ah, there you go, okay. Now the other trend coming out of COVID that I started to see was when the enterprise IT, the CIO, or the CTO, so their cloud bill, they're like, oh my goodness, I didn't know that my cloud bills were so high. We were seeing this trend of cloud repatriation. One, is that something that you're seeing? And then two, how does that affect your strategy going forward in terms of cloud repatriation? I mean, granted, you're still gonna have your hooks to the cloud, but I'm assuming that, you know, that that's becoming a more important trend going forward as well. >> We've always told our clients that cloud has unique capabilities for certain aspects of your business, but like all technology transitions and architectures, it is not a solution for everything that a corporation might want. And so we've always helped our clients look at patterns of applications and data usage and said, hey, let's help you use the cloud for the right things and use your own engineered environments for the use cases where that is most beneficial. We do that through a combination of tools and software as well as professional services teams that work with clients on cloud data design workshops. The second is we help clients, you know, move data easily. If they wanna move it to the cloud, we can help them do that. If they wanna move it back from the cloud, we can help them do that. We also allow them to refactor their applications. You know, sometimes the applications that a client uses is not well suited for the cloud because it's monolithic. And any clients now are refactoring it to make it more modular using technologies like Kubernetes and we help them do that. And so those are some of the ways that we work with clients. Listen, our view has always been, there is no one solution that solves all the world's problems. - Yeah, but I hear you want that. And I think we just start talking about refactoring and new technologies. This is a good segue to talk about AI 'cause you've been talking about AI for the last, in earnest for the last 18 months, right? And we're still waiting for that inflection to come, but what goes through the NetApp AI strategy and how does it align with the emerging enterprise AI workloads, because we're still waiting for that to ramp up? - Yeah, I think first of all, we have worked in the AI market for a very long time. You know, I would say that there were, AI is the culmination of both mathematical tools as well as using large data sets to bring sense to business problems. We've worked with clients around large scale analytics, like clustering and regression, pretty much when I got to NetApp almost 15 years ago, we also have many, many clients using us for what you call predictive AI that helps you improve decision making, right? And those are mature markets. I think the place where there's a lot of excitement and innovation that's happening today is really around generative AI because you can use natural language tools so that you can broaden access to AI capabilities much more widely. You know, for an enterprise to use AI effectively, there's sort of three or four important things that they have to deal with. You know, 80% of the time that the AI and data science team in the enterprise spends is really around, you know, what you call data wrangling. Finding the right data, putting in place the right guardrails around proper use and access to that data, being able to transform that data so that large language models can use it and be able to keep the lifecycle of the model and the data synchronized. And so not only are we providing super high performance scale storage infrastructures that can feed GPUs data very, very quickly, but we are also, you know, delivering to clients the capabilities that allow them to, you know, find all the data that they have and search for it using natural language tools, build a semantic understanding of that data and not just what the data is, but how, what meaning exists, what's the relationship with different types of data, automate the transformation of that data and be able to keep track of implement guardrails, you know, keep track of the lineage of the data. And so we're really helping them solve the fundamental work and data workflow and data challenges that they have to make AI usable in the enterprise. >> So not only creating the links and trying to find the data, is there also a, is there a scale challenge as well, or is that something that you've already solved given your history with machine learning and AI workloads in the past? >> Yeah, I think first of all, many of the traditional applications and, you know, data paradigms work for what's called structured data. This is data that is in databases, which is organized using tables and columns and so on. Those are very small data sets, right? You can see a very large database being a few terabytes of data, maybe 10, maybe 50 terabytes. When you reach unstructured data, which is typically about 80 to 85% of the world's data is unstructured, of an enterprise data is unstructured, we are the dominant provider of solutions for that. You need to think about scale, you know, context indexing, you know, search in a very, very different way. And we have the tools given our experience to be able to do that. You've got to think about, you know, kind of a highly distributed system. You need to think about context and content as the data changes rather than in a batch model after the data is created and when it needs to be used. And so there's a lot of innovation that we've been working on over the last several years to enable that. It doesn't happen overnight. - Okay, so I was watching NVIDIA GTC when Jensen was on stage. And I was actually surprised. Well, he does call our companies all the time, right? And he didn't hurt him NetApp, right? And as part of it, he said that storage needs to be reinvented for AI, right? How many understand it? What does that mean? We have explained some of it, but just, I'm trying to better understand what does that mean. And it says, you know, I don't know if that meant for the hyperscale cloud guys, but does it also mean what does it also apply to the enterprise customers as well? - Yeah, I think there's probably two or three ideas there. I think the first is, you know, a lot of the storage technologies have been really focused on sort of managing devices and trying to bring scale to different types of devices. I think one important transition there is to think about how does, you know, the management capabilities move from the device to actually the data, right? And that's an important step. A lot of what I just talked about was related to that discussion. You need to have not just large pools of data, but you need to have an understanding of the metadata that is the data about the data and the relationships of that metadata, the semantics and the content of that. I think that's one important transition that, you know, people need to make. The second is, you know, if you look at the infrastructure of an AI factory or of a data center, you need to be able to, you know, aggregate resources in really flexible pools so that you can optimize the processing pipeline all the way from compute through network fabric to data access. And there's a lot of innovation around ultra low latency connections and ultra high scale composable distributed system architecture that we'll be showcasing at our customer conference upcoming. And then I would say the third idea there is, you know, as the models change from essentially, you know, kind of predictive models to more reasoning models and diffusion models, you know, the idea of how you manage the full stack across compute and storage where you're constantly accessing data in loops. And how do you make that efficient? Is another area of innovation? We, for example, have ways to feed GPU pipelines extremely efficiently. Those are sort of the core ideas. And then, of course, the big idea has always been, how do you unify all the data in a corporation so that you can bring real depth, right? And so there, what we are also working on is in addition to having the infrastructure to unify data across the hybrid landscape, we are working on intelligent data agents that can actually analyze the data so that the application and workflow agents talk to our data agents, rather than have to go and dig through all the data. - Okay, I actually do want to dig in a little bit about that. Sounds like there's a little bit more software that's involved. - That's right. - But you hinted at RAG, right? And in one of your prior comments. But Jensen also talked a lot about semantics search. - Right, that's right. We could talk a little bit about the difference between RAG and semantics search. And how should we think about the opportunities as inference is really starting to play a bigger role in the overall use case of adoption? Because it's been mainly training for now and now the discussion of inference is starting to ramp up. - Absolutely. To derive value and to be able to generate value for organizations, inferencing has to become capable, successful, and a deployed widely across a corporation, right? Because I would say training is just getting the tools ready. Infencing is where the real money and the productivity gains that AI promise really happens. So it's crucial that inferencing succeeds. With regard to the idea of going from RAG to semantics search, is especially like the idea of semantics is you have both structure and meaning, meaning context and content awareness to the data, right? In the database world, semantics were essentially implementing a data schema and allowing a queryable access model like using SQL or other types of query methods to that data. In the world of documents and unstructured data, for example, you need to be able to have a structure to that unstructured data that's typically through rich and deep metadata. And you need to be able to have an understanding of the relationships of that data, right? So that if you say blue cat, maybe the nearest other reference to it is green cat versus blue dog or something else, right? So that idea of not only content, but context and relationships. And so we are working on technologies that allow clients to have not only deep and rich metadata about their data across the entire hybrid landscape, but also to build content and context awareness to that data so that when you do a search, the search is much more relevant and credible than what you've seen to date. - Fair to say that you're leveraging AI to enable these newer technology technologies that you're trying to deliver. - Absolutely, our technologies are not only built to enable our clients to use AI, but they're also powered by AI. How do you manage the data? How do you identify the relationships with the data? That's automated in our systems using AI capabilities. - Okay, now driven that you are enterprise exposed and you made mention of it that for AI to really work, you need inference to work, to real inference adoption to ramp up. - I have a couple of questions related to this. One, have we started to see this inflection point of enterprise adoption as it relates to inference from your lens? - In specific industries and use cases where the data is well organized, we are already seeing adoption, right? So for example, in parts of healthcare, where the data sets are well structured, they've got well organized, you know, kind of metadata associated with the data, because of the need to comply with regulations, you are seeing good progress, right? You know, in other places, we see what is happening in the early days of cloud. It's almost a mirror, right? Where they've got an AI center of excellence, they're starting to pull data together, they're starting to run proof of concepts, and the way that you see that happening is you go from proof of concept to production over probably an 18 month period. So we've always said, hey, the, you know, the time for inference is probably 2026, second half of 2025, where you start to see the leading players go from proof of concept to scale production. - Yeah, and we're already hearing some Neo Cloud providers talking about mode, you know, 40 to 50% of their workloads are now inference base, and Jensen has talked about, 40% of the workloads are inference, and some of the commentary around open AI. So we're starting to see that pick up it seems on the cloud side, and it sounds as if you're seeing that as well. - That's right. - In terms of the opportunity, as it relates to inference, is the financial opportunity more on bigger pools of storage, whether because they want to save more, or they're creating more data, or is it managing the growing complexity as it relates to the data that's being generated from AI and trying to find the data to help train or actually get the responses for their AI models. - I think it's more on, there's opportunity in both, but it's more on, for me, I think the real value we deliver clients is on the ladder, right? Which is, we want to make it easy for them to extract real value from their data. Though I think there will be data growth, there will be, you know, more data generated because it's easier now to generate more data and all of that, but I think if we don't deliver on the promise of we make it easier for you to use your data, you know, that's an important strategic priority for us as an organization. - No, I'm going to have you at my next statement saying that this is not long-term guidance on your part, but when I framed the beginning with the IDC growth, it was always like a low single digit growth forecast going out to 2029. I almost get a sense that they may be a little bit conservative because there is an AI pump opportunity that they may not be fully considering. And I have, look, the IDC guys are my friends. So, you know, I almost get a sense there's a little bit of conservatism from the growth forecast of fully factoring an AI. - When a capability allows a business to grow or to drive productivity, there is a lot of spend on that set of capabilities. I think AI has that opportunity. I think that, you know, if we are able to catalyze both applications that drive, you know, top-line growth by serving clients better, as well as productivity, we should see the benefits of that in terms of customer spend. - Okay. Now, there's my last question. I actually think this is one of the more important questions as it relates to AI. You kept on mentioning one key phrase that you kept on mentioning is part of our conversation is guard rails, right? You know, we're seeing a lot of sovereign AI adoption. We're seeing a lot of on-prem AI, right? And then that's purposeful. But could you just talk about the security aspects of data and what you provide from a security aspect? To help differentiate from your peers? - Yeah, I think first of all, you know, if data weren't already one of the most important assets of a business alongside its people, AI makes that even more true, right? Because it is a proven fact that with better and higher quality data, you can derive more insight and more value out of the same AI model, right? Bigger model is not the only way to actually derive value, actually having better data is key to success. As a result, we see, you know, the threat from malicious actors grow in terms of, hey, I want to, you know, steal your data, I want to modify it in ways that compromise the value of that data. And so we have a lot of capabilities that allow you to detect in real time when somebody is trying to modify the data or delete it or exfiltrate that data. And we won the data protection product of the year in one of the most prestigious, you know, competitions this year. The second thing that we are working on is, you know, organizations have for many years implemented access controls and, you know, sort of ways that the right people have access to the right data. What many of them say is, hey, now I loaded up in a model and pretty much everybody in the world says, I want that data to be loaded up in a model. And it blows up the controls model that clients have had around governance and risk. And so we are enabling our clients to actually carry those controls forward as they think about utilization in languages, language models, right? So for example, if you have PCI, you know, data, we will either mask it or we will signal that you have PCI data when you try to load that data into NLM. If you have data that changes and you want to make sure that you have lineage of that data so that the model is, you know, has the right up-to-date data and it doesn't have skew, we have ways to automate and say, this model has access to this data and it has changed. And so there's lots of sophistication in the way that we make it easy for organizations to carry forward their guardrails into the world of AI. - All right, that's it for me, but I have one final question. You mentioned that the inside conference is coming soon. Any preview for a financial analyst like me to get excited about? - We have our customer conference called NetApp Insight in October and we will be showcasing a whole range of technologies all the way from, you know, sort of advanced, super high scale, storage infrastructures, new collaborations with the hyper scalers around their tools with NVIDIA around its AI data platforms that deliver real value to clients. And a lot of the capabilities that I talked about around, you know, how do you use your data? How do you have a semantic understanding of your data? How do you apply guardrails? All of those capabilities, we will be making available to clients. So I'm excited for the show. I wanna thank you for having me. - Yep, look, I look forward to the keynote and thank you for being on the show, George. Any final parting comments for the audience? - Listen, we have been privileged to be part of humankind's long, you know, search for meaning and for storytelling, sharing its traditions with succeeding generations using data. The first data storage and management solution was actually a bone carved with humans, you know, with stories almost 30,000 years ago in Central Africa. And today our company works with the leading organizations in the world to unleash their potential, their, you know, unique value to the world using modern data tools. And we're privileged to do that. So thank you for having me. - George, that's perfect. And thanks everyone for joining us. May and my colleagues have a great lineup of future disruptors like George and that app. So hit the subscribe button to keep up to date with the Tech Constructors podcast and not to miss an episode. With that we'll wrap.