November 16, 2026

00:42:30

From The Field at NetApp Insight 2026 | Dhruv Dhumatkar, Mark Fioretto & Jurgen Hofkens on Data Sovereignty in APAC vs Europe

From The Field at NetApp Insight 2026 | Dhruv Dhumatkar, Mark Fioretto & Jurgen Hofkens on Data Sovereignty in APAC vs Europe
KBKAST
From The Field at NetApp Insight 2026 | Dhruv Dhumatkar, Mark Fioretto & Jurgen Hofkens on Data Sovereignty in APAC vs Europe

Nov 16 2026 | 00:42:30

/

Show Notes

Recorded at NetApp Insight 2026 in Las Vegas, this From The Field episode has KB talking with three NetApp leaders about how data sovereignty is playing out in APAC and Europe.

She asks CTO APAC Dhruv Dhumatkar what’s really holding AI back, and he points to poor data quality, not a shortage of GPUs. He also explains why NetApp builds cyber resilience into the storage layer.

Mark Fioretto, Area VP and MD for Australia and New Zealand, tells KB that sovereign AI and data sovereignty now matter to every organisation, not just regulated industries.

Jurgen Hofkens, CTO for EMEA and LATAM, tells her that if you can’t move your data, you aren’t really sovereign, and warns that AI pilot copies can become ungoverned islands of data.

View Full Transcript

Episode Transcript

KB [00:00:09]: What's up everyone? It's KB coming to you from MGM Grand here in Las Vegas for NetApp Insight. Everyone wants to move faster and with AI, and everyone wants control over their data. But those two ambitions can become complicated very quickly when your business operates across borders. Where does the data live? Who can access it? Which rules apply? And how much control do you actually retain once that information moves through your infrastructure? Sovereignty sounds like one conversation, but looking across Asia Pacific and Europe, we need to understand where the priorities line up, where they differ, and what that means for the organizations building these systems. So I'm bringing together my conversations with Dhruv Dhumatkar, CTO, Asia Pacific, and Mark Fioretto, Area Vice President and Managing Director for Australia and New Zealand, and Jurgen Hofkens, Chief Technology Officer for EMEA and LATAM, who all work at NetApp to explore sovereignty through those two regional lenses. KB [00:01:17]: We're getting into the tension between local control and global operations, and what businesses need to consider when the demand for AI moves faster than their ability to answer those questions. This is KB from the field at NetApp Insight 2026. Let's get into it. Joining me now are Dhruv Dhumatkar, CTO, APAC at NetApp, and Mark Fioretto, Area Vice President and Managing Director for Australia and New Zealand, also at NetApp. So today we're discussing what it will take to move from AI, from experimentation into production across one of the world's most diverse regions. So Dhruv and Mark, thanks for joining me and welcome. Mark Fioretto [00:02:00]: Thank you, Karissa. Great to be here. KB [00:02:02]: Okay, so Mark, I'm going to start with you. Everyone is talking about AI, but many organizations appear to remain stuck in experimentation. Given your role, what are customers across APAC and ANZ more specifically struggling to sort of move beyond? Mark Fioretto [00:02:23]: I think there's a lot of conversation at the moment around how do they move from that pilot phase to full scale production. What we're seeing is a lot of organizations now really leaning in on making sure that they're understanding the ROI they're getting from these pilots. How do they scale that, particularly around how do they access different parts of the data that they have scattered around the organization? And increasingly now we're seeing board levels leaning in, supporting those organizations by appointing chief data officers, for example, or chief AI officers, really trying to move, as we said, from that pilot to a full scale production, but with the governance and the security protocols around that as well. KB [00:03:05]: Okay, so Dhruv, I'm going to go to you, given your technical experience, and I know you've been with NetApp two decades, so let's drill on that a little bit more, given your lens. Dhruv Dhumatkar [00:03:16]: Well, picking up from where Mark left off, the biggest challenge is the data. So we're not short on GPUs, we're not short on space, we're not short on technology. It's the quality data. That's needed to feed these LLMs. Without the quality data, it's a case of garbage in and garbage out. So that's a big challenge that we see across enterprises, we see public sector entities, even specialist manufacturing entities. It's a very common problem that we see, not just in Australia, but across Asia Pacific. KB [00:03:50]: So staying with that garbage in, garbage out, do you think that people are becoming more cognizant of that now, nowadays? Dhruv Dhumatkar [00:03:56]: They are. They're starting to understand, that an effective ROI comes from a valid business case, as Mark spoke about, but then also having the right data to feed into the LLM, so you get quality answers once you start scaling into production. KB [00:04:14]: And so just to extend that a little bit more, would you also say that now people are cognizant what they're feeding into the machine, so to call it, that's then also that byproduct of the whole ROI conversation? Dhruv Dhumatkar [00:04:27]: Correct. Absolutely. KB [00:04:29]: So, I'm aware that this week at Insight, there's a lot of major announcements. So for an APAC leader, perhaps, Mark, I'm going to go to you. What has materially changed and why should people really care at the moment? Mark Fioretto [00:04:43]: Look, there are a number of great announcements that are happening during this week, but I think the important thing to focus on here is this is really an evolution of a technology and technologies that NetApp has had, you know, in the 30 year history of the organization. So the enhancements really are around, trying to help a lot of these organizations that are trying to scale their AI projects. Now, the truth is this data that sits in these organizations is really distributed and there's a lot of silos, both from a structural perspective, but also from an IT perspective. So a lot of these announcements that you're seeing this week is enabling customers to access that data in a very unified way, underpinned by our ONTAP operating system, which again, is present both on -prem and in the public hyperscalers. Mark Fioretto [00:05:28]: And so, what you see and what you will continue to hear this week is, you know, getting access to the data is important. Making sure that the data is secure is also important and making sure that you can do that in a timely and an efficient way as well. KB [00:05:42]: So yesterday at the press briefing, a couple of your executives mentioned about the fragmentation. So would you say that gap's closing now for companies in the APAC region? Mark Fioretto [00:05:51]: I don't think it's closing yet. What we've got to do is provide the tools and the ability for them to close that gap. It's not a, one and done thing. It's not like we can turn around tomorrow and close all of those gaps immediately from the perspective of how the data is structured and where it's structured. Accessing the data is the first part, getting access to that data, and then over time, being able to actually create a more unified way of bringing that data together. But today, the solve is how do you get access to that data in an efficient way without having to completely re -architect your environment as it stands. KB [00:06:24]: And just before I go over to you, Dhruv, Mark, what would you say, is that sort of the main, question at the moment from customers in the Asia Pacific region? Or concern? Mark Fioretto [00:06:34]: Access is one, cost is another. So as you scale AI, obviously there are cost requirements that are associated with that. So I would say definitely access, cost and security. So making sure that data doesn't fall into the hands of bad actors, that also internally from an organizational and governance perspective, guardrails are put in place. So I think that'd be the three priorities that I'm hearing more and more. KB [00:06:57]: And so Dhruv, would you say security? So what is that a main sort of conversation at the moment with customers that you're speaking to? Dhruv Dhumatkar [00:07:07]: I think so. Well, if you look at the APAC projected spend, APAC is projected to spend $68 billion by 2028 on security. And when we think about security, a lot of attention is given to identity, management, endpoint security, et cetera. But if you think about, you know, the high profile case studies of security breaches, data has been stolen. And so it's very much an afterthought. So to answer your question, it is a top of mind discussion amongst the boardroom, amongst C staff. And what we've done is we've deliberately engineered cyber resilience and cybersecurity features at no cost into the storage layer. And that is, pivotal and there's plenty of case studies across Asia Pacific where our capabilities have helped customers secure and protect their data across many industries. KB [00:08:09]: So you know what's interesting about that? It's sort of like from a media point of view where I'm sitting. I've had so many conversations over years about yes, the continuity side of it, continuous business, those sort of things. Then it sort of dropped off. But now with everything that's happening specifically around that critical infrastructure, cost of everything. It's almost like this conversation has made a bit of a resurgence then in the market. Have you seen that then as well? Dhruv Dhumatkar [00:08:34]: Yeah, absolutely. I have absolutely seen that. It continues to be a discussion at almost every customer meeting. It's something that we bring up proactively and it's something that customers respond to because they are thinking about not only how do they modernize their estate, but build a security posture as they modernize. So that's, yeah, absolutely. KB [00:08:59]: And so now I want to just switch gears and talk about sovereign AI and digital sovereignty are becoming major priorities across the region. So Mark, how is that sort of changing customer requirements? Mark Fioretto [00:09:14]: I think the important point to make there, is that it's not just about regulated industries. Sovereign AI is, and sovereign data sovereignty has become something that's now across the board for all. organizations. So what we're seeing is, for example, you know, we know that 25% of organizations either have moved or are planning to move their data to a sovereign cloud. So this is a major initiative now and a major concern for many organizations and something that certainly has the board leaning in on making sure that the data of that organization, the value and the privacy of that information is guarded well. So we are seeing that conversation now almost in every discussion we're having. Mark Fioretto [00:09:56]: some form of or some version of discussion around data sovereignty and sovereign workloads as well. Dhruv Dhumatkar [00:10:04]: What are your thoughts? Well, I'm just picking up from where Mark left off. Sorry, Mark. But I think that it's sovereignty in terms of the border control, in terms of the border control, but also in terms of understanding and being able to tag your data. So back to the discussion we had earlier about siloed data, what we're seeing is customers wanting to see what data they have, understand what value it's going to deliver, particularly in the context of AI. KB [00:10:33]: Again, going back to the sovereignty piece, spoken about a lot a couple of years ago, then nothing, again, spoken about again. Why do you, is it because this is the way the market's moving in the region or what are your thoughts there? Dhruv Dhumatkar [00:10:45]: Well, look, if you, if we go back to the earlier discussion about data being critical to AI, if we, if we look at the use cases of AI, there are, there's, it's about, increasing productivity of a workforce. It's about building competitive advantage or just operating more efficiently. You know, those are the type of use cases that we see across different industries. Now, the ability to, build a highly efficient AI, or adopt an AI initiative is dependent on your data. And that data is essentially the lifeblood of that organization. It's the sum of all knowledge, competitive product operations, and so companies want to protect their data and make sure it potentially doesn't fall into the hands of bad actors that are overseas or even competitive hands. So that's where it comes from, my experience. KB [00:11:39]: Now, APAC as a region is incredibly diverse. Are you both seeing different AI priorities across the markets? Because when I speak to people in North America, that's what's happening down in Asia Pacific, it's very different. KB [00:11:55]: So, I want to hear perhaps from yourself, Mark, and then I'll go to you, Dhruv, about what's going on there. Mark Fioretto [00:12:02]: I think the use cases for me that I'm hearing a lot about obviously starts with productivity gains within an organization. So making sure that, you know, from a structural perspective and from an ability to be able to serve internal customers as well as external customers, we're making it, customers are making use of AI as a tool to help them with productivity. Certainly there from there, you start to also think about, how do they, create more customer centricity? So how do they use that data that's been collected over the years through different interactions or through different, means that they've accumulated that data? How do they really tailor their offerings for customers to make sure that they've got customer centricity and driving value for that customer? Mark Fioretto [00:12:44]: And then certainly from a perspective of speed to market, I would say that would be the third thing where it's a competitive marketplace, regardless of which industry you're in. So using AI to drive competitive advantage, drive speed to market, whatever the offering may be that you're taking to market, would also be another one that tends to come up in discussions as well. Dhruv Dhumatkar [00:13:06]: how about I lend an APAC lens? I think when you look at countries such as Japan, China, Japan, Korea, India, you know, English is not the first language. And if you look at the frontier models, there's a huge opportunity that these countries see for building their own LLMs in the native language. Then when you look at countries such as Japan and India, these are large populations and the governments are moving to digitize many services. So they automate health services and the like. So there's a variety of use cases that we see that some are similar, but some are very different and very specific to the country in which they're being adopted. Dhruv Dhumatkar [00:13:53]: One other trend that we're seeing is the adoption of physical AI, which is essentially not large language models, but you know, using manufacturing robots. And we're seeing a huge increase, particularly in Korea and Japan, where there's large manufacturing industries for AI to automate a lot of the manufacturing process and improve the quality control. KB [00:14:17]: So do you see then the LLMs being developed in their native language? Do you see that as a good thing or is that positive? Dhruv Dhumatkar [00:14:27]: Well, they certainly see it as positive because if you look at the frontier models now, there is a translation wrapper around those. And what the governments of these countries feel is that they need to build their own language, large language models from the ground up using the literature of that particular country. And I think it's a good thing. They certainly think it's a good thing. And certainly by building models in the native language, AI is becoming accessible to all, not necessarily people who understand English. KB [00:15:04]: So AI factories and neoclouds are gaining a lot of attention across APAC. What do these models reveal about where then the market's heading, Mark? Mark Fioretto [00:15:16]: As we talk about scale and we scale out AI, organizations now are faced with that challenge of how do they actually provision for these larger models to be able to then bring you know, again that scale across the organization? So they're leaning into organizations like neoclouds to actually get access to performant compute or GPUs and also scaling from a capacity perspective as well. The other thing that we're starting to see now is consumptive models. As you may well know in Australia, we are fans of the consumptive models. Whichever type of technology that you're actually consuming And so we're also starting to see them not wanting to manage the whole stack. Mark Fioretto [00:15:56]: And so leaning on the neocloud type providers to provide part of those offerings or part of the stack and to manage it for them as well. KB [00:16:06]: So just with that, why do you think they don't want to manage it? The complexity of it? Mark Fioretto [00:16:11]: What, like, Is there a reason for that? Well, I think as we go through this stage of scaling, a lot of organizations are trying to decide how much of that they want to actually own, how much of that they want to try or to continue to, you know, pilot and scale. And so as that model evolves, they'll decide where that water level is of what they want to invest in, what they want to use on a permanent basis and what they may want to use on a semi -permanent basis as well. Dhruv Dhumatkar [00:16:37]: That is that is absolutely what we see across Asia Pacific. So the GPU as a service provider, the neoclouds, the ability just to rent GPUs for testing and experimentation, there's no shortage of ideas of where AI can add value to an organization or a government. And so you wanna test that in an area where it's a predictive cost, the infrastructure is stood up for you and all you bring is your software development and your data. And if that proves the ROI is proven in an experimentation, then you can bring it back on premises and scale it. And we're seeing neocloud providers come up in Australia, in Singapore, India, many of these major countries, we see neocloud providers coming up and finding no shortage of customers willing to experiment on their infrastructure. KB [00:17:34]: Operative phrase that you use there is predictive cost. So would you say traditionally people have been a little bit taken aback by some of the costs associated to it? Dhruv Dhumatkar [00:17:48]: I think AI, you know, the GPU costs are quite high. And of course we're in a phase of the of the of the market where costs of acquisition is top of mind for customers. And back to what Mark was talking about around consumption models, customers look at these neocloud providers in the same way they look at the traditional hyperscaler. I can rent capacity, I can experiment. And then if it's not proving feasible or it's not returning an ROI, then I don't need to use that infrastructure. It's the avoidance of acquisition of infrastructure for a potentially unproven initiative that is top of mind for them. And that's the value that the neocloud provider provides to all organizations, big and small across APAC. KB [00:18:38]: And would you say because of that reasoning, it allows latitude for businesses to test trial, they don't like it, can turn it off, can switch, et cetera. Dhruv Dhumatkar [00:18:49]: Exactly. And you know, like I said before, I think once you have an idea of where AI can potentially increase productivity of your employees, simplify operations, gain competitive advantage, it's that testing phase that gives businesses a level of comfort that if the idea doesn't work out, there's no long -term investment that's gone to waste. KB [00:19:17]: Okay, so gentlemen, closing comments, final thoughts. So Dhruv, I'm gonna go with you first. Dhruv Dhumatkar [00:19:21]: Look, it's been a very exciting set of announcements at the keynote, starting with the data platform and NetApp Novus. I'm looking forward to working with our customers across Australia and across Asia Pacific to bring this technology and add value. Mark Fioretto [00:19:35]: I think we're at a really exciting phase of the industry. We're going to unlock so much potential in different ways. The opportunity is gonna be governed by how fast organizations can actually scale some of these models. I think everyone is taking the right approach, which is to do it in a you know, in a considered way to really understand what the business impact is, the customer impact from a positive perspective as well. And then obviously with the consideration, the commercial considerations around that. But I think we're gonna continue to see an acceleration of the deployment of different technologies. And for us, we feel in a very privileged position to have so many customers trust us today with our technology and with our offerings that we've had over the last 30 years. Mark Fioretto [00:20:14]: And so we're looking forward to continue that over the next 30 years. indeed. Mark Fioretto [00:20:19]: Gentlemen, thank you so much for your time. Thanks very much. KB [00:20:24]: Joining me now is Jurgen Hofkens, Chief Technology Officer for EMEA and LATAM at NetApp. Today, we're discussing how regulation, sovereignty and escalating cyber risk are shaping data innovation across Europe. So Jurgen, thanks for joining me and welcome. Jurgen Hofkens [00:20:41]: Thank you for having me. KB [00:20:42]: Okay, so you spend a great deal of time speaking with customers across EMEA and Latin America. What is dominating their technology agenda right now? Jurgen Hofkens [00:20:54]: It's a couple of things, I guess. You know, one of course is AI. You know, AI is driving their innovation agenda right now. And then specifically, I think, how they get from, of course, pilots to production and how they get the data to go do that. But I think Europe is slightly, you know, different in a sense that sovereignty really becomes important. And it's starting to drive the agenda for a lot of people, not just the ones that are in public sector or, you know, regulated environments, but it's basically architecture, technical choice, you know, vendors, you know, sovereignty is really becoming important. And then we've had quite some issues around cyber resilience. Jurgen Hofkens [00:21:35]: So also understanding how they can build an enterprise that is resilient against cyber attacks is becoming really, important to our customers. KB [00:21:43]: And when you say driving the agenda, do you mean that's what's constantly coming up in conversations? Is that what's dominating perhaps in terms of the laundry list? Jurgen Hofkens [00:21:53]: What does that look like? That's what we talk about. That's what's on the top of mind for our customers. It's what's driving decisions in terms of, you know, technologies that they deploy and it's driving their IT strategies. KB [00:22:09]: And would you also say, Jurgen, that the sovereignty piece is specific to Latin America as well? That's more like European, given, you know... Jurgen Hofkens [00:22:17]: It's getting everywhere. Of course, you know, the context, the global context has, you know, made people much more aware of, you know, the fact that they need to have a sovereign strategy. But it's really important in Europe. But it's, you know, you see it in the Middle East as well. And you see it in LATAM as well. So I wouldn't say it's just Europe, but it's really, important. KB [00:22:46]: Because even a few years ago, people talked about sovereignty and then it went away and now people are talking about it again. But would you say a lot of that has been powered by the AI conversation that now it's like, okay, we're going to focus back on the sovereignty piece again? Jurgen Hofkens [00:23:04]: It's partially that, of course. It's partially the political context. You know, it is what it is. You know, I think specifically in Europe, but, you know, we want to be less reliant on, you know, you know, cloud providers that are very, specific in, you know, legislations that may not be very, you know, very friendly to Europe. So there is part of that. But AI creates an additional dimension. It's creating an impetus for investment, you know, making sure that we have the technology in Europe, we have the technology in the Middle East and Africa that allows our enterprises, our public sector to be competitive and not just relying on, for example, models. Jurgen Hofkens [00:23:45]: Right. You know, that are produced outside of Europe. So it's both, I guess. It's not just AI, but AI is absolutely changing the game and for sure changing the speed at which this is being looked at. KB [00:23:58]: So you see AI projects that succeed and those that stall, for example. So where do these projects typically break down? Jurgen Hofkens [00:24:10]: You know, the most important thing is when you start scaling, this is not anymore about, doing a little prototype. You need to do this at enterprise scale. You need to make sure that the data that's being used by the agents or by the AI, you know, the AI project or the model is, you know, company wide, you know, with the right resilience, with the right governance around that. And so once you get out of the, let's just do a little pilot and you need to start, you know, building something that has strategic impact to the company. You need to get that data piece right. You need to understand which data you have in the first place. Jurgen Hofkens [00:24:53]: And once you have the data, you need to be able to use it in a confident way with the right security, with the right governance around it. And that is stalling a lot of the projects. It's less about the model and it's less about, you know, the actual software and the agent. It's less about, you know, the agenda, you know, environment you build up. But it's, you know, what is the data underneath? What I see a lot of, customers do is build a pilot and then just copy some data, you know, just for that pilot. It's easy, you know. And then suddenly these copies become, you know, get alive on their own. Jurgen Hofkens [00:25:28]: And you get islands of data sitting around that are not governed anymore. And this is where it starts failing because now suddenly those people don't have the right information anymore. The AI is being used, doesn't have the right information, doesn't get updated. You know, it's, you know, it's actually also, you know, from a security perspective, you know, that copy creates an additional risk surface. And now suddenly that just doesn't work anymore. And now we need to get the governance in place and that delays everything, right? So you really need to think, start thinking about that before you do your pilot. And that's where I see a lot of, like, maybe not failing, but for sure delaying projects to really get into production. KB [00:26:09]: And would you say that there is still that lack of confidence perhaps around, like you said, the security piece, the governance piece, would you say that's what's holding people back? Or do you think that confidence will come in time? Because we're still figuring things out as an industry. Jurgen Hofkens [00:26:24]: I think we're still figuring things out as an industry. And it's, I would claim it's a bit more maybe than confidence. It's the realization that now suddenly these agents who can look up data on a, you know, on a, like, much faster than humans, now suddenly, have access to data that should be governed, right? And so that realization, you know, makes a company think twice, you know, specifically if you're in a regulated environment. So it's just a little bit less about, I guess, just the confidence on its own, but really understanding, okay, so this agent can now get access to this kind of data. How do we make sure that, you know, it doesn't see too much data? Jurgen Hofkens [00:27:10]: And when it gets the data, how do we make sure that it doesn't abuse the data? You know, in a way that we don't want the agent to abuse the data. And so what's the technology around that? How do we make sure that we secure that, we actually monitor it, we audit it? It's just something that a lot of companies have to learn, to your point, create confidence around. But it's the technology piece that just needs to get there, and they need to get it right. KB [00:27:36]: Would you say that learning curve is relatively steep at the moment? Jurgen Hofkens [00:27:40]: It is steep. We're, of course, from an IT perspective, We're helping them to make it less steep. But yes, it's steep. It's new. It's new in, you know, a technically ambiguous situation right now. A lot of that technology is, you know, evolving and evolving rapidly. So you have that steep curve of the technology on its own, and then you have the learning curve inside of the organizations. So that combination just makes it, yeah, steep. KB [00:28:06]: So would you say the lack of confidence because it is steep learning curve, and then that learning curve is, would you say that people are concerned around the velocity of that learning curve or the ambiguity around that learning curve? Jurgen Hofkens [00:28:20]: The velocity is, I think, less important. I do see a lot of companies giving a little bit more freedom to, you know, to people to, you know, get that velocity right. I think it's more about do we have all of the information? Do we have all of the data? Do we have the data we need to make our project successful? Do we then represent it in a way that the AI can use? do we make it develop in a cost -effective way, you know, without creating too many copies? We just discussed about that. Do we, and then, you know, do we make sure that the governance around which agents can access what kind of data and what they can do with it and the enforcement of that governance is in place? Jurgen Hofkens [00:29:03]: And that's just, yeah, just needs to get, you know, get done right. And there's very little basis. I think from an enterprise perspective, there is some basis on data governance, of course, but now everything speeds up. These agents are just much quicker than human beings. And so that speed up makes people think twice and make sure that they have the right things in place. Does that make sense? KB [00:29:26]: Definitely. So I want to move on and talk about NetApp says AI agents can now read, move, act on data faster than governance was designed to follow. KB [00:29:39]: So I'm from a security background. So. Perhaps then for security leaders, what does that then actually break? Jurgen Hofkens [00:29:48]: There's a couple of things. So that you need to look at. So on the read side, if you want, you know, we just discussed that, right? On the read side is what kind of data do they have access to and what are they doing with the data and how can I audit that, right? But also on the write side, you know, if you look at an agent in reality, an agent is going to get some information, is going to share the information in a prompt, with an LLM and is going to get some information and write that information as a context back to its file system, right? Jurgen Hofkens [00:30:19]: Now that suddenly that information that is written back, if you write it to a simple file system becomes ungoverned, but it's a combination of data that used to be governed that the agent got access to, a mix of what the LLM has done with that data when it provided an answer back and, you know, the basically the reasoning that the agent is doing around it. Now suddenly you get that mix of things on that file system. Now if you don't do anything around that, you don't monitor that file system, you don't understand what the agent is doing with that data, suddenly a lot of the governed data that the agent had access to might actually become completely ungoverned. Jurgen Hofkens [00:30:57]: Also, the file system in that surface, if you want, that the agent is using to write their context back becomes an ideal platform to understand what agents are doing. Like if suddenly an agent starts reading too much in a way that it didn't used to read or starts writing too much, or is starting to share a lot of information after reading some of the confidential information in the file system, you know, you start to get a good understanding of that agent going rogue. So suddenly that file system becomes a really interesting surface to do cybersecurity around, or at least the monitoring around it. And so that this, you know, a place where, I would suggest, you know, some of the CISOs to really look at is like, where is that information being written? Jurgen Hofkens [00:31:45]: And are we making sure that we're using that surface to really enforce some of our monitoring around what agents are doing and actually understanding whether or not agents are going rogue, if you want. KB [00:32:00]: So, I'm then curious to understand, people talk to me in my interviews about like, theater governance, theater, for example, so do you think that, is it, the industry is still trying to work out like where everything sits? Cause it's kind of like people are saying that they're governing things, but then in actuality that they're not. Jurgen Hofkens [00:32:21]: I think they're not, just to be very clear. I think, which may or may not be a good thing, right? like sometimes like you have to decide how much governance you want and the governance will, you know, if you want the scale of governance will also determine the scale of speed of your innovation in a certain point of time. But I think for the moment speed is becoming, is a little bit more important than governance in some industries and for some of our customers. That's why they start creating additional copies. That's why they don't really, understand how to use their existing, you know, data systems to enforce the governance. Jurgen Hofkens [00:33:02]: It's changing though. Because of course these agents become, you know, they're not in a pilot anymore. It's not a, you know, let's play with it. becomes a really important part of the existing structure and strategy of a company. Now the data they attach becomes more important. Becomes more, there's more need for regulation. Also the impact they have on the actual company strategy and the exposure becomes more important. So it's changing. It's changing rapidly. but I guess, you know, it's still, you know, I guess, you know, people, you know, prioritize speed over governance for the moment, but it's changing. KB [00:33:44]: So I'm going to focus on Europe for a moment. And Now, it appears that Europe is taking a very different path from the United States, even Asia Pacific on the sovereignty piece. Spoke to some of your other colleagues in Asia Pacific. There's a bit of that, but what does genuine sovereignty look like once an organization begins running AI workloads? Jurgen Hofkens [00:34:08]: So I think, so if you would go and ask people in Europe about sovereignty, they would give you a, for now, a fairly static definition of sovereignty. You know, they would say it needs to be, the data needs to be in a certain location. It needs to be in a place where a, you know, certain legislation, drives whether or not people have access to the data. It needs to be, you know, the data or, you know, more than the data, all of the software can only be accessed by certain nationals, but it's very static in a sense. It's just like, you know, that's the definition. Jurgen Hofkens [00:34:49]: AI changes that significantly. And the reason why is because if you look at what you need for a successful AI project, what you need a lot. You need the data. We just discussed about it in the right location. You need a specific model to be able to do, you know, a specific large language model most of the time that model needs to run on accelerators that need to be available in that legislation. And then you have to have the right regulation around that. And if you start adding those four, it becomes really, hard to actually find a place or infrastructure to actually do your project, right? Jurgen Hofkens [00:35:34]: And then on top of that, this might be true today, but with the speed of change tomorrow, that picture might completely change. And so when I'm talking about sovereignty to our customers, I say the static governance and the static sovereignty is important, but they have to think about how they can dynamically change where they use the data to a place that follows the governance or the sovereignty in a certain moment of time. You cannot just use a static definition. So for example, say, hey, let's go and do all of our AI in this location because this is where we have the legislation, et cetera, et cetera, and then just stay there because it might well be that in two or three months from now, the model you need you know, changes completely, the accelerator you need changes completely, and the actual data needs to go to a complete different location. Jurgen Hofkens [00:36:27]: And so if you can't move the data at a certain moment, you're not sovereign anymore. You're linked to that specific location. So I think dynamic, static sovereignty, really, important, and this is how today regulation is looking at it. But I think AI creates an additional part of complexity because it changes so rapidly. You also need to be sure that you are independent from where the actual AI is happening to be sovereign, and you have to be able to move your data. The data gravity cannot be so heavy that you're actually becoming dependent rather than sovereign from the actual location where the AI is happening because that will restrict you significantly. Does that make sense? KB [00:37:13]: Yes, because my next question is, for these European companies, they start on this endeavor, but then you said things change. So then how do they then go, okay, well, things have changed. We now need to take a different course. Is that what's becoming really difficult for these people to continue down this sovereign path? Jurgen Hofkens [00:37:28]: It becomes... Yeah, that's interesting. It becomes hard. We're back to the discussion we had around, you know, what you do with your data, right? It becomes hard if the way you implement your AI projects is by copying data. Let's say I have this amount of data, and now I need that data in another location. Let's go and copy, create another copy of it. And now create another copy because it needs to move and create another copy. Before you know it, you have plenty of copies. They all need to be governed. They, you know, they're all, you know, if you want, almost like an attack surface from a security perspective, they all turn stale at a certain moment in time. Jurgen Hofkens [00:38:08]: It becomes really, hard to manage it. So what I'm suggesting our customers to do is make sure that instead of creating these copies, get a data platform in place that gives you the flexibility to actually do that move and to do that, you know, create that flexibility to go and use different models in different locations, in different legislations, you know, with exactly the same data. KB [00:38:33]: And would you say European organizations are at risk of becoming too dependent, like you mentioned before? Jurgen Hofkens [00:38:41]: Too dependent, For sure. If you're not flexible today, like regardless of, you know, a political definition of sovereignty, right? If you are too dependent, you know, because your data has too much gravity and you can't manage that, you significantly reduce your flexibility to go and grasp new technology capabilities or grasp a new neocloud that now suddenly has a different model available in a different location because you can't use that benefit because you can't have the data move. So if you implement that flexibility now suddenly, you know, your options become, you know, to be sovereign become a lot, you know, broader and you can actually take strategic benefit from that. If you don't do that, you're very, restricted. KB [00:39:31]: Okay. So Jurgen, final question. This week, NetApp has made several announcements, a lot going on. Jurgen Hofkens [00:39:37]: So perhaps... A lot going on. KB [00:39:39]: For people who weren't in the room, couldn't make it unfortunately to Las Vegas this week, what's changed? Jurgen Hofkens [00:39:45]: What can you tell us? A lot of things have changed. So if you would take a step back, actually, a lot of things have changed for, you know, different, you know, type of customers. first, I would say our more traditional customer base, we've announced an additional hyperscaler, you know, that is delivering our, you know, our storage systems, our file systems as a first party, Oracle. To do that, we've introduced capabilities of a Keystone sovereign. So in Europe, you know, we didn't talk about that, but, you know, like a storage as a service product is now, you know, a fully sovereign product from a European perspective. Jurgen Hofkens [00:40:25]: So we do a lot around that. We do a lot around, you know, making sure that we've introduced actually, console and an agentic way of managing your storage estate. So a lot has changed in that, I would say more traditional customer base. So that's number one. A key innovation we've done for customers that, you know, like neoclouds and AI factories is Novus. With Novus, we've basically, in my opinion, we've created a new standard in terms of in terms of performance that you require to train or to, you know, to train or to inference on large AI factories. I think we are about 10x the throughput of, I would say, you know, what was the previous benchmark. Jurgen Hofkens [00:41:23]: So for me, with Novus, We've set a new benchmark in that industry in terms of performance. So really, important for us there as well. And the third space that we're innovating and where we've done a lot of announcements is around, so back to the discussion we had on how you use the data, how you discover your data is with AIDE. now, you know, you're able to kind of use it across your full NetApp estate, but also on top of that, your full data and storage estate. outside of NetApp as well. And So with that, people can discover which data they have And then they can, you know, the governance we talked about, they can start using that data for their agents in a much more governed and efficient way. Jurgen Hofkens [00:42:12]: And so lots of announcements have happened there as well. So super exciting. KB [00:42:16]: Jurgen, thanks so much for your time. Jurgen Hofkens [00:42:17]: Thanks for yours. Really, thanks. KB [00:42:19]: Thank you. And there you have it. This is From The Field. Stay tuned for more.

Other Episodes