Episode Transcript
[00:00:16] Speaker A: Welcome to K beyond the go. This week I'm on the ground at NetApp Insight 2024 conference at the MGM grand in the heart of Las Vegas. For this bonus series, we've been lucky enough to have lined up conversations with the selection of NetApp executives and other guests exploring the future of data and Aihdenhe. Stay tuned for the Inside track from some of the world's leading authorities, presenting at Insight 2024 as KBI Media brings you all of the highlights.
Joining me now in person, it's Prabhji Tijuana, senior vice president, Angie and cloud storage business unit at Netapp. And today we're discussing cloud storage and cloud providers. So Prabhup G, thanks so much for joining the welcome.
[00:00:57] Speaker B: No, thank you for having me over here.
[00:00:59] Speaker A: Okay, so I've been at Netapp Insight now for three days. I know that NetApp is very big on building relationships, especially for hyperscalers. So maybe tell us a little bit more about what is building and maintaining relationships with major power providers sort of mean? And why is it important to you?
[00:01:15] Speaker B: One of the commonality which we have with the hyperscalers and how we do things at NetApp is all about working backward from the customer requirements. The reason why we are having a good relationship and have been a successful joint venture between all three major hyper scalers. Remember, we are the only first party cloud storage available in all three major hyperscalers, AWS, Azure, as well as Google. So when we started this journey, it started with like, hey, what are the capabilities we jointly want to solve for our customers? And that's when we look jointly into all the on tap capabilities which we have, from our security to our data replication to our how we manage data, all those aspects. And they resonated with what customers were asking hyperscalers for. And that strategy is still going on, right? Like, we work very closely with them, look into what are the customer problems we want to solve and then work backward from it in terms of like, how do we solve it together? Some of engineering works and falls on my organization, some falls on hyperscaler organization. So there is a lot of planning and working together, which happens in that. But the whole thing here is as long as we continue to stay focused on if we are solving the right thing for customer, I think that partnership is, that's the reason why it has been so successful and by yet, that's why we are unique also in the whole industry. There is no equivalent of how our first party cloud storage services are available. There is no parallel to that in the industry, we are the only one who provide this one. And running clouds services, there is another aspect which is like the core fundamentals, be it security, availability, reliability, performance. So I think since hyperscalers work at a very high scale, these are also very important aspects to them. So we work very closely with them on these four things also. So combination of our and their four focus on customers, four focus on service fundamentals. It makes the relationship really go well.
[00:03:15] Speaker A: So you said the main thing is to stay focused on solving customer needs.
[00:03:19] Speaker B: Yeah.
[00:03:19] Speaker A: How would sort of someone know if they were focused on solving customer needs on it. How do people ascertain? Yes, we're on track and we are solving those needs.
[00:03:26] Speaker B: Yeah, it takes a lot of effort.
[00:03:27] Speaker C: Right.
[00:03:28] Speaker B: Like all the time we are talking to customers either who are using us in on premise or using us in hybrid or using us in multi cloud.
[00:03:37] Speaker D: Yes.
[00:03:38] Speaker B: The most important thing is to stay in sync with customers. Right? Like I'm like 90% of the roadmap which my organization deliver is actually controlled by customers because we are so our product management organization, even our GTM organizations, they're constantly talking to customers on figuring out what their pain points are, what is working, what's not working. And so that is the key thing which we do, and we spend a lot of energy and effort on. And the results are also equally reciprocal. When you spend so much of energy on customers.
[00:04:08] Speaker A: And so you said before, always constantly speaking to customers, how do you sort of like gather all of that intelligence to be able to say, hey, this is the path forward, this is what we should be doing.
[00:04:17] Speaker B: Yeah. So that's where we are. Like the product management process kicks in. So basically we talk to so many customers, we look into the patterns which are emerging.
[00:04:26] Speaker A: Right?
[00:04:26] Speaker B: Like what we really focus with customers is what and why, what they really want to get solved and why is it important to them. Right. Like theme. Then patterns start emerging. And then my engineering organization and product management organization sit together and figure out how we will build it. And then when we build it, like, our focus with customers is less about power and run. It's all about what and why and then using that what and why is to understand what are the patterns which are coming and then prioritizing based on like how much value they bring to customer, how many customers meet that. All those dimensions go into that. And it's a continuous iterative process. It's not like we do it once a year or something. It's like pretty much every three months we look into in our industry, the trends change very fast. The apartments change, especially when it comes to governance, compliance, security. These things are changed constantly evolving. I'll give you an example. This year Microsoft started this whole security initiative where all Azure teams are working on security sprints.
So we got this requirement, we sat with Microsoft, we understood what are the problems which we want to solve on security space. And we basically delegated pretty much all of our organization also to work hand in hand with Microsoft and work in same sprints to deliver the same initiatives which they are working on. So these kind of things happen, but they happen right now. The good thing is we have built this security over the years in OnsVAP, which we believe is bar raising. And now we are combining it with the security practices and security controls which our hyperscalers has mandated and asked us to do. And fomination is a really good victory for customers because they get high bar.
[00:06:14] Speaker A: Talk about perhaps, what do you think customers get wrong about cloud storage from your perspective?
[00:06:19] Speaker B: I wouldn't say that they get wrong, but there is lot of things when you are doing into putting your strategy in place or moving your workloads or business mission critical, business workflows into cloud. There's a lot of planning which is involved.
[00:06:35] Speaker C: Right?
[00:06:35] Speaker B: Like starting with. Right? Like hey, what are your organizational outcomes which you want from going into cloud store? Not just cloud storage but in cloud and in particular, right. Some customers, what we have seen is customers start with the things like disaster recovery, right? Like they will start with disaster recovery use cases and cloud, they will start using cloud for elasticity, they will start using it for instant capacity covering, they will start using it for bursting of traffic and so on. And that's where they start getting familiar with how to use cloud, how to use the different tech stacks and so on. And most importantly, understanding cost of cloud, mostly understanding security, governance, data residency, all those means, all those required a lot of rework, right? Like I think where we have in few instances seen is like customers just not doing that pre work to the hundredth and to the nth degree and just moving a mission critical work quickly. I think it's a once you have done pre work from everything from assessment to your cost to your security and outline it with the outcomes that you want for moving to the cloud, I think usually customers do really, really well. And then combination of our cloud services and hyperscalers, there is a well defined architecture framework for defined paths, how to do that which they can use.
[00:08:01] Speaker A: So you say well defined architecture, give us a high level. What does that look like?
[00:08:07] Speaker B: So like, right, like if you want to have like cross zonal or cross regional availability, high availability and those kinds. So there are standard practices and patterns which you can use, right? Like how do you set up your ha in cloud? Right? Like do you want to do it at an easy level? Do you want to do it a regional level?
[00:08:25] Speaker C: Right?
[00:08:25] Speaker B: Like so based on your business outcomes you want and what kind of workload you are running, which is the most important thing you can. There are practices and patterns which are defined by combination of Netapp and hyperscalers which they can read, understand, learn and work even with apps, solution architecture teams and so on to find the right workflow for that.
[00:08:46] Speaker A: So I asked you before around a lot of people get wrong, but perhaps I should rephrase it and say what do people perhaps miss?
[00:08:52] Speaker B: I think the missing part sometimes is they don't sometimes spend enough time to understand the cost very well.
It's not about that it is expensive. It is about how to govern the costs. You can turn on things and then if you don't have a best practices around, how to control how many instances are running, how they're governed and so on, sometimes cost and Boolean. But if you do that with a good planning and good oversight, then those things don't happen. So I think you also have to look into who are the resources, who inside your organization will be responsible for making sure items like security, government costs, migration. There are multiple ways to migrate to cloud and then also making decisions like hey, does your organization want all cloud or do you want hybrid cloud or do you want multi cloud? We have variations of those, right? Like we have lots of customers who use in multiple clouds and we provide capabilities, enough cloud storage, how to manage them in multiple clouds. We have two out of five customers who are onboarding tools cloud storage with us. They are onboarding us in a hybrid mode. So we have capabilities like Blue XP workload factory which helps them manage those things. And getting familiar with those tools and capabilities is unique, a good thing before you start doing those migrations.
[00:10:11] Speaker A: So you mentioned before having good planning, good oversight. What does good planning look like then?
[00:10:15] Speaker B: It's basically looking at fast looking at them. What are the workflows that you want to modify? It has to be all workflow driven, understanding what performance you want for your workload, what kind of availability you want for your workload. What are the data governance like? Probably best is to document them all together in one place and then looking into, right, like hey, I what will be the right choice for implementing those choices in the cloud? You can do anything in software engineering in multiple ways, like we can, based on those requirements, your footprint and cloud will look different. So doing that three assessment and all these dimensions is very, very important.
[00:10:52] Speaker A: So I go back around the governance side of things. I've interviewed a number of your executives the last few days and that's obviously been something that's been raised a lot in terms of governance. However, you have to know what data you have and where it is to be able to govern it properly. So what does that sort of critical path look like from your perspective?
[00:11:09] Speaker B: Yes. So we expose a lot of these capabilities to our manageability controls, right? Like APIs and so on. So being getting familiar with those sites, like, hey, having a good understanding of like, hey, what are the security and authentication schemes you have and so on, understanding who can access your data with what permissions and so on. I like that's all the free work which you have to do, right? Like who can get access to the data, who cannot get access to the data. So those are the kind of governance. And then there are broader shades of governance also, right? For some workloads, they want data governance data being in a specific country, right? Like, so we have capabilities of for that. Then there are governance items like federal or government workloads, which has to be done in specific type of regions that are hyper scalers. Then there are some air gapped regions which people want to use for very, very mission critical workloads, especially in the petrol sector and so on. So based on that, your governance is 100% controlled by what kind of workload you have and what you earn, what is getting used for, who's the end customer, all those kind of things.
[00:12:14] Speaker A: So I follow us a little bit more. So I'm based in Sydney, Australia, flown here for NetApp Insight 2024. And one of the big things that people often speaking about in Australia is software capabilities. So the data being stored.
[00:12:26] Speaker B: Yeah.
[00:12:27] Speaker A: Talk to a little bit more about your thoughts then on this.
[00:12:29] Speaker D: Yeah.
[00:12:29] Speaker B: So if you see cloud storage with all three major hyperscalers, we are available in hundred regions total, like combination of all three hyperscalers, right. Many customers of ours work on the data software and part of their where they want data to work in a, let's say in a specific country. And we do provide those capabilities. When you provisioning your cloud storage with us, you can define that. Hey, if I'm in a country x, keep my data in country x only so that controls are available. It's up to customer to use them if they want to, if they don't want to use, that's fine too. But those controls and capabilities are there in our hexadecim.
[00:13:03] Speaker A: What do you think is moving forward? What do you think is sort of going to be a common trend or thing that we're going to start to see emerging over the next twelve or so months?
[00:13:10] Speaker B: The world is seeing a very interesting intersection of likes. On one side there is a massive data growth happening, zettabytes and zettabytes of data is being produced every week. Then there is a migration to cloud happening. The cloud is an integral part of the strategy. If you see broadly, every organization has some cloud strategy. And then there is this advent of AI. Think this intersection is becoming very, very interesting. You will see lots and lots of customers using artificial intelligence to build, to accelerate their own businesses using that data. Power lies in the data and that's where our intelligent data infrastructure really plays into. For example, right, like we are, we have a capabilities plans for all three hecat scalers. If you're using FSX for on tap today we are integrated with Amazon Bedrock. So you can use the foundation models available in bedrock with your data stored in FSX for on Pap very soon. If you are using, if you are stored storing your data in Google Cloud to us, we have a similar plans for using in GCNB, which is a Google Cloud. And you can use Vortex AI and bigquery with the data stored in your Google Cloud NetApp volumes. And you will be able to use Vertex AI agent builder to build low port, no port applications. And if you're going use an Azure cloud of ours, right, like then we have integration which I demo did in the keynote also yesterday. You can use as you can make your data available to Azure one nate and doing that, even use wide array of AI and analytic services available from Azure. So you will see that our customers will start using lots and lots of AI with the data in the coming months and coming years, I think it's going to definitely accelerate. I think also we will see more and more workloads start getting migrated to the cloud. We have already seen that. That's why we have capabilities like workforce factory, which basically helps you to deploy, operate, migrate, protect all your workloads when you migrate them from on prem to cloud. And then you will also see a lot of focus on hybrid. I think hybrid is becoming normal. So that's why we are investing a lot in data migration, data mobility, disaster recovery, using capabilities as snapdragon, flex cache and so on. So these are the key patterns which you will see but we do expect lot of growth in terms of our cloud storage today. In last twelve months, we have grown our footprint in cloud by 114%. And I think it will be much higher in the coming extra 24 months.
[00:15:52] Speaker A: Julie now in person is Jeff Baxter, vice president product marketing at NetApp. And today we're discussing NetApp production announcements. So, Jeff, thanks for joining mobile.
[00:16:01] Speaker D: Thank you. Great to be here.
[00:16:02] Speaker A: So, Jeff, let's start with, for those who are not familiar, explain Ontap. Now, over the last few days, I've heard a little bit more about the announcements, what's happening? But maybe explain what does this mean?
[00:16:13] Speaker D: Yeah, so OnTaP is our unified operating system that NetApp invented over 30 years ago. Now, they really helped unify customers data state across all their data centers and into all the major clouds. And we're uniquely partnered with Amazon, with Microsoft, with Google to be present in all of those as a native offering from mem. And so we can really span the gamut for all applications, all workloads, any type of data for our customers, anywhere they need it.
[00:16:41] Speaker A: Now, we're obviously a media company. We've done a range of announcements today that's gone out on our site or known over the last few days as well. So maybe talk at a high level about some of the new product announcements and then we can get into a few more specifics.
[00:16:56] Speaker D: Yeah, so we certainly, we kind of broke down the announcements into four different areas right around enterprise storage, cloud storage, cyber, and AI. On the enterprise storage front, there was really a focus on our continued focus on building block optimized solutions for customer San storage. And so we introduced the new ASA, a series models that are simple, powerful, and affordable for customers to be able to consume. For block storage. That was on the enterprise storage side, we announced a range of different cloud storage initiatives, innovations across all of the different cloud providers, and then multiple different announcements in cyber and AI that I think we'll probably get more into.
[00:17:32] Speaker A: Okay, so let, okay, let's get into that. Now, in Tanzania, AI, where do you think it's going? And I know from a media perspective, a lot of people are talking about it. However, when you start saying AI and artificial intelligence, people's eyes start to glaze over a little bit more. So what's your view?
[00:17:47] Speaker D: I think this is the year that we finally start to get serious about AI being put to use in enterprise use cases.
[00:17:54] Speaker C: Right.
[00:17:54] Speaker D: We've seen sort of the consumerization of AI has been a huge theme from chat, GPT, sort of kicking it off to everyone from Apple and everyone else introducing you to consumer devices. And then we've seen very large corporations training large language models and doing intense data science. But what we really haven't seen is this full democratization where every enterprise is able to do it. And so this year is really the year of being able to take all this innovation that's going on in the ecosystem, being able to take all these large language models and apply companies actual enterprise data and actually bring AI to their data to start actually generating business results. And I think that's really where the next sort of epic of AI moving forward for the rest of this year.
[00:18:32] Speaker A: Ok, I want to get into a little bit more. So obviously, large language models, for example, one of the risks would be the hallucination of AI. So what are your thoughts then on that?
[00:18:41] Speaker D: So I think there's a couple of different issues there. One, we acknowledge that there are tremendous risks to AI. Our CEO, George Kerry, on day one talked all about the risks and the perils that we have from AI, both when it's used properly, and even then it can hallucinate, as you say. And obviously there's improper uses of it that are present with any technological advance. I think from our standpoint, introducing data governance and introducing sort of sanity into data as much as possible is what we can do to help resolve that problem. I don't think. I think it's going to be a collective group effort, and you're never going to eliminate hallucinations entirely. But the more you train AI on the right type of data, the more you ensure that that data is governed and sanitized ahead of time, the more likely you are to get accurate results from AI.
[00:19:25] Speaker A: So one of the things I'm hearing as well in interviews is people talking about, like, small language models. That way you're controlling and the data is what we say in Australia, I'm seeing that, especially if you're working in a government organization, for example. So do you have an opinion then on that in terms of slns?
[00:19:41] Speaker D: Yeah, I think our goal is that you can bring the same tools available, whether in the cloud or whether on prem, to any sort of data.
[00:19:48] Speaker C: Right.
[00:19:48] Speaker D: So whether it's a small language model, whether it's using large language models, but doing training and fine tuning on them with retrieval, augmented generation, to be very focused on your specific private data set. I think that's the future of enterprise AI.
[00:20:01] Speaker C: Right.
[00:20:02] Speaker D: These large language models are useful for sort of consumer applications, but for enterprises, you need to either build your own right, in which case you need to really build your own AI factory, whether large scale or small scale.
[00:20:11] Speaker C: Right.
[00:20:11] Speaker D: We hear Nvidia talk a lot about AI factories, and we help build them, or you need to be taking these large scale models that others have built and really fine tuning them for your needs, as you would find small language models.
[00:20:24] Speaker A: What are the perspectives? What do you think people feel overwhelmed by when it's. When people hear AI? Just generally speaking, I think it depends.
[00:20:33] Speaker D: Upon the type of person. That's a great question.
[00:20:34] Speaker C: Right.
[00:20:34] Speaker D: So here at NetApp, we have a lot of people. We span the gamut in the customers we serve, but a lot of our customers are at the infrastructure level.
[00:20:42] Speaker C: Right.
[00:20:42] Speaker D: NetApp is the intelligent data infrastructure company. And a lot of our customers, all the customers here at the event we're at NetApp Insight 2024, are storage administrators. They work in the data center. And so when you start talking about AI and you start talking about data science, there's this huge sort of data gap. And the question is, how do we get from here to there, right?
[00:21:00] Speaker B: How do I.
[00:21:01] Speaker D: Do I need to upskill myself? Do I need to become a data scientist? Right. How do I become a data engineer? And how do I really take everything that sounds like a hype level and everything that I'm using, you know, on my iPhone as a consumer device and actually take the steps to make it practical? And I think that's. That's sort of the fear and that's sort of the challenge that I think a lot of the customers that I top wins today.
[00:21:20] Speaker A: So then how do you find that equilibrium, then, moving forward? What, what do you think's on the horizon on that front? And then we'll sort of flick over to what's on the horizon for NetApp.
[00:21:27] Speaker D: Sure. Yeah. I think, you know, for us, it's all about companies. And certainly NetApp, we hope to be the leader in this about how do we abstract portions of this process, right. Today, AI is very much like the dawn of many other technological innovations, where you're stringing everything together, right? You go back to the invention of the World Wide Web, and we're writing HTML by hand. Right? I mean, you, literally, everything that you do at the start of any technical innovation is by hand. And then you introduce abstraction layers, you introduce automation. And that's really what NETF and other companies are trying to do in partnership with companies like Nvidia, where you're not having to reinvent the wheel, you're not having to move data seven times every single time that things become automated, that they become standardized. And so you can do plug and play models to do your AI on an enterprise level. And that's really what I think that this next is. The future of AI is moving from sort of the kitchen sink experiments to constant prepackaged meals of AI, let's just say, and being able to constantly do things in a repetitive fashion with ease.
[00:22:26] Speaker A: Well, just like, as well, when people think about the word automated, do you think people start to feel a sense of control over their data, for example, would you say?
[00:22:34] Speaker D: Yeah, I think that's vital. We, a lot of the announcements in this event focus on is completely understanding your data state, because if you don't understand your data state, you're already well behind at the start of the AI game. So understanding your data state, of being able to make, you know, we announced at this event a global metadata namespace, which you don't need to bring all your data together. I think one of the interesting things that's happening with AI is we're recognizing that this concept of a traditional data lake or data warehouse, while they're still viable for traditional sort of recording, they don't operate at the speed, and they're not cheap up with all the data sources that are necessary to truly inform enterprise AI. So you've got to get away from this concept of trying to move all of your data together into one place and be able to bring the AI to your data and to be able to represent all of your data in sort of a synthetic global namespace, where you can see all that data, you can access it, you can feed it all the AI engines, but it doesn't need to be all brought together in a traditional ETL process.
[00:23:28] Speaker A: I have to ask perhaps maybe a rudimentary question in terms of, do you think companies really know where all their data is stored?
[00:23:35] Speaker D: No, I think most companies don't, and I think most companies would admit that. Right. And whether it's through shadow it or whether it's through data sprawl, obviously the move to the cloud, although it hasn't exacerbated it, certainly made it easier to fray data silos, right, because you're moving, in a lot of cases, to a software as a service offering, right? So if you're buying a service, there's data behind that service, but you never intrinsically move the data. It just happened as a result of moving your service there, right? So truly understanding what all those data sources are and unifying them, and more importantly, knowing where your data is, but then do you have the right governance on top of it? Do you have the right compliance, all the privacy implications. I think we, it varies company to company, but it's definitely a challenge for just about every company I talk to.
[00:24:17] Speaker A: So now I want to sort of switch gears a little bit. And we're cyber studio podcasts, as you know.
[00:24:21] Speaker D: Yes, of course.
[00:24:22] Speaker A: So I want to focus a little bit more on the Netapp cyber resiliency enhancements and announcements that you guys have made.
[00:24:28] Speaker D: Yes.
[00:24:28] Speaker A: But then also I want to sort of talk about perhaps and maybe addressing some of the security challenges as well, from your perspective.
[00:24:35] Speaker D: So from an announcement perspective, I think we made a couple of different announcements. One is we're incredibly focused on ransomware, right. We look at ourselves as basically the last line of defense. If you look at the zero trust security model. Right. We know that all of our customers have perimeter security. For God, I hope they do.
[00:24:50] Speaker C: Right.
[00:24:50] Speaker D: We know that they are the endpoint security. We know they have checks at every point. But the zero trust model, which has been implemented for a very important reason, is the assumption that everything down to the storage layer is compromised. And we operate in that assumption we used to a decade ago. So I've been in that for 16 years. And a decade ago, we said, well, if they can directly access your storage layer, you're already game over. Now you have to operate in the assumption that people are there, and people have been there for a long time, and people are setting up shop inside the network. So if you operate on that assumption, hey, you've got to have an incredibly hardened storage os. And we've gone through basically every certification you can think of. We're actually the only storage that's certified by the us government to store top secret data everywhere, correct?
[00:25:28] Speaker A: Nsaid.
[00:25:29] Speaker D: Yeah, the NSA commercial solutions for classified data. Right. So we have that one we signed onto the secure by design pledge that cis originally come out with. So we did all of that. But then we also assume that even if the system is entirely secure, it's going to be taking in intrinsically flawed data. And some of the intrinsically flawed data is accidental, but a lot of it is ransomware. Right.
We think that we have an opportunity and an obligation to actually address ransomware all the way at that bottom layer. So we were really one of the first to introduce, I think we're still one of only two vendors among all the different enterprise storage vendors out there to actually embed anti ransomware directly into our storage.
[00:26:08] Speaker A: Is that because you guys obviously consider security as important, and it's what is always being talked about now in terms of what's always coming up in the media, in the news, and I feel that perhaps data or storage, from my perspective, would be. It's something that people forget about and it feels a little bit relegated.
[00:26:25] Speaker D: Yes.
[00:26:26] Speaker A: Are you, as a NetApp has brought that to the forefront, is what I'm hearing.
[00:26:29] Speaker D: I think there's people look at NetApp and especially netappable, and say, oh, they're just a storage company. Right? That's certainly not true. We do a ton more than that. We call ourselves the intelligent data infrastructure company. But I will take a point of pride in saying we don't try to be everything to all people. So while people get distracted by shiny object syndrome over the years, all we try to do is focus on data infrastructure. So that lets us to be ahead of the curve when there are certain curves that intersect. So, like, ten years ago, when data efficiency became important, sustainability really started to come to the fore. We were already building deduplication, everything into our enterprise storage. We were the first to have enterprise storage deduplication, like, 15 years ago, well ahead of the curve. Same thing with anti ransomware. We started building anti ransomware technologies over five or six years ago into storage. So Giv was already ready and available to customers at the time when concern over it really peaked in the last several years, and when the constant, ongoing daily damages from ransomware attacks became front page news constantly, we already had a solution. So I think, you know, focus can sometimes be incredibly helpful, and that's really what NetApp has done, is be focused on the data infrastructure layer and how we can add intelligence into it.
[00:27:35] Speaker A: And I asked that question because when George Kirin, CEO of Netapp, as you know, flew out to Australia and he, I was asking him about this, and he's like, well, we're not a security company. I'm like, yes, I know. However, you are baking that into what you're doing, I think that's perhaps from a market perception. It's about changing that perception of we're just not a storage company.
[00:27:53] Speaker D: Yes, I agree. It's, it's about, and, and that's why we kind of coined the term data infrastructure. We're definitely not, I would agree with George. We're not trying to be a security company.
[00:28:01] Speaker C: Right.
[00:28:01] Speaker D: A security company is trying to control all the endpoints, is trying to offer solutions for that. We're trying to say for this very part of your, important part of your infrastructure, we think we have an important role to play in it, and so we can do that, like, with our ransomware. The thing we announced yesterday was embedding AI ML models directly into that anti ransomware technology.
[00:28:19] Speaker C: Right.
[00:28:19] Speaker D: The assumption is that more and more ransomware attacks are actually being crafted either with the aid of or entirely by AI. And the only way you're going to keep up with that pace, right. A lot of these threat factors are not even human understandable anymore. Things like prompt injections and others into AI engines.
[00:28:34] Speaker C: Right.
[00:28:34] Speaker D: They started with silly tech strings, but now they're becoming incredibly complicated ciphers that in some cases data scientists don't understand why they're working. The only way you're going to be able to conduct that is by having AI ML models that can actually protect your data as well as be the ones attacking it. Almost sounds like a Sci-Fi movie, but so we announced the ability to build it in and we've had it tested by SE labs and others. And if you've hatched 99% of ransomware, the ransomware that makes it past everything else and hits our system, we can still block 99% of that.
[00:29:01] Speaker A: From your perspective, what do they think when you're speaking to customers or you're out there in the field?
Well, what would you say is the main question as sort of asking, like nowadays, obviously, like as things progress and things sort of change?
[00:29:12] Speaker D: Yes.
[00:29:13] Speaker A: What would you have on their mind, just out of curiosity?
[00:29:16] Speaker D: Yeah, so I think there's a couple different things depending on where they are in their journey and where they are.
[00:29:20] Speaker C: Right.
[00:29:20] Speaker D: European customers are still very heavily focused on sustainability. That still comes up a lot. Right. In the US, it's still important, but I think it's always been a secondary concern, and I think it's especially been a secondary concern recently with things in the economy and different things like that. So, I mean, the number one thing we hear from customers is they're trying to figure out how to constrain their costs in a relatively cost constrained environment. So how do I knowing that my data is increasing at a rapid pace? And more importantly, I was talking to a customer yesterday who said it's not just that the amount of data growth is there because that's been a constant since the time immemorial.
But it's that now in the age of AI, we're afraid to throw data away. We used to think some data is trash, we can get rid of it once we have the end products, but now we think of every stage of data along the line as potentially being viable for a training model down the line. So now we've got to keep all of it. So, Neb, you have to help us with ways to reduce the cost of the storage. And then the other two major things that I hear we've already talked about, which is how do I make use of AI in the enterprise and how do I protect myself against ransomware. And those are really the big three I hear in almost every discussion I have with customers.
[00:30:22] Speaker A: So Jeff, do you have any sort of final thoughts or closing comments?
[00:30:27] Speaker D: What makes NetApp special, and I've been here 16 years, which is an eternity in the tech space, it's because we keep sort of reinovating and we always have sort of the customer at the center of our thoughts and we are very focused on how do we take this data infrastructure, how do we add intelligence to it, as I mentioned, and yes, it's a marketing tagline and I'll say that, but it really is at the focal point of how we do all our product planning. So I run product marketing, but actually report to our chief product officer. And so every day as we're doing product planning, it's all about how do we build the best data infrastructure, how do we add intelligence to do it, how do we drive costs down, how do we increase security and how do we make data AI ready? And those are really every day how we wake up and how we think about innovating. So that's what makes me excited, like to sort of rejoin NetApp on a daily basis, is that we continue to innovate on a daily basis and just be maniacally focused on building the entire intelligent data infrastructure.
[00:31:19] Speaker A: Join me on person is Priscian Vital Dvara, senior vice president and general manager at Nanop and today we're discussing the future of intelligent data infrastructure and AI. So Chris, KV, thanks for joining. Welcome.
[00:31:30] Speaker E: Yeah, thank you KB, really excited to be here.
[00:31:33] Speaker A: Now I'm going to start with the first time I saw you on the main stage, you wore a shirt. Yes, but tell us what was written on the Shern.
[00:31:42] Speaker E: It says bring AI to your data.
[00:31:45] Speaker A: And it's exactly the premise of what we're going to discuss today. So on that note, what does that actually mean then?
[00:31:52] Speaker E: Yeah, so I think a lot of folks today think that to do AI, you need to have the data actually move to where the compute is. And it's actually incredibly hard thing to do because our data is everywhere and data has significant gravity. Data comes with where the geographically distributed, what kind of permissions are there on the data. So it's a very, very hard thing. And if you try to take data into AIH, you end up creating what we call data filers. So the way you solve the problem, or the data problem as we laid it out, is by doing two things, by making your data AI ready, and then by bringing the IQ data, meaning you can do inferencing in place, you can create your vector embeddings in place, and all your data can be turned into insights wherever and however you want as a possible.
[00:32:48] Speaker A: Okay, so a couple questions on that. So data silos, what do you mean by that specifically?
[00:32:52] Speaker E: Yeah, so the data usually gets created and it stays with wherever it gets created. And every time someone else starts a new project, in this context, AI projects, it's really easy for them to just start a new store in volume, or just starting new directory, or, or create a new storage system where the data was. And as you, over time, you end up with your data getting fragmented and being everywhere.
And also data gets copied over multiple times. So for AI, if you're copying data from its original data source to the new data source, you are essentially creating a silo. So for us, the goal is not having to create silos.
[00:33:39] Speaker A: So you've been something before. Chris shook around data insights. So historically, one of my roles about a decade ago was I was reporting illus, so I had an affinity to data and, you know, intelligence and driving insights. But do you think it's something that perhaps people are just not really doing in terms of deriving insights from their data because they're so busy trying to keep their head above the water, keep the lights on, they're not really focusing on, oh, what do I. In terms of insights, what am I getting from?
[00:34:07] Speaker E: Yeah, if you're a data analyst, you kind of really know this problem pretty intimately, right? It's not that you don't intend to generate insights, it's just a slower process to move from data to knowledge and knowledge to action. So the process takes longer than it usually should.
And when the process takes longer, what you're actually wasting time is expensive resources like yourself, like the data scientists, and the idea is make it as seamless, as friction free as possible for the data to move so that you get your insights fast.
[00:34:44] Speaker A: So talk a little bit more about lifecycle on data. So for example, as you would know, even like a decade ago, I worked in a bank, people talking a lot about gathering all of the data, all of the intelligence cloud, Dara, for example. And now we're sort of seeing the trend of going the opposite side of, well, perhaps we don't want to keep all of this data, we want to get rid of it. I think that's one of the stats from a sustainability perspective, 68% of things aren't even being used ever again. So what are the things going to happen now? Are we going to sort of pendulums, they go swing from one end of the spectrum to the next? Are we going to start to see it come in the middle now?
[00:35:20] Speaker E: I think there are two parts to the question that you had. There's a part of it where the beta is actually core for what you.
[00:35:27] Speaker B: Do as a business.
[00:35:28] Speaker E: If you're a bank, understanding your customers.
[00:35:31] Speaker B: Understanding what the risk profile of those.
[00:35:33] Speaker E: Customers are, that's a meta you want to actually keep and maybe keep forever, because you know that data is very domain specific and that is how your organization is going to be extremely successful or not based on that data. But then there is also a lot of the other kinds of data, like the operational data, the day to day transactional data. But over time, if you live in.
[00:35:55] Speaker B: The regulator industry, once you convert it.
[00:35:58] Speaker E: Into insights, you don't need to keep it around. So I think it goes back to once the data and how you as an organization, which data do you value the most?
[00:36:10] Speaker A: And from your perspective, what do you think people commonly value the most?
[00:36:14] Speaker E: They commonly value the data that actually lets them make the most, you know, create the most shareholder value so their customers extremely low.
[00:36:22] Speaker C: Right.
[00:36:22] Speaker E: Anything that's not, that is actually not as interesting, like a good example of it is, let's say you're in healthcare industry and you're using a lot of public domain data. You might actually bring in the public domain data to generate a particular insight, but you don't need to preserve that data. But if you have the patient data or a clinical trial data, that's propagating you. So there is significant value for you in keeping the data and preserving the data.
And when in the world of AI gets even more important because of things like monoliths, vainability. So you start, let's say you dealt with trial and you're in the drug discovery, and you want to be able to go explain why you made the decision you made or what kind of side effects did the drug produce or not produce in your studies.
[00:37:11] Speaker B: So you need to be able to.
[00:37:12] Speaker E: Preserve the data that you use to drive the models that created the drug. So from a model explainability perspective, that's the data you absolutely want to preserve.
[00:37:22] Speaker A: So perhaps then on that example, what do you think? What about biases? There's a lot of that sort of commentary floating around in the market around, you know, the biases on the data and the models. And I know I've spoken already to some of your other executives around hallucinations. I won't touch too much on that, but more around biases, or how, who gets to decide what is not necessarily wrong or right, but perhaps if you look at it, zero and one perspective, who gets to decide that? There's a lot of, that's a conundrum at the moment, but do you have any insights on that?
[00:37:55] Speaker E: Yeah, I mean, responsible AI is all the rage, right? Like, we all want the data to be unbiased, but at the end of the day, like, AI is being built by humans. And sometimes, knowing ye or unknowingly, our biases actually show up in the data as well as the conclusions you draw from the beta and the insights. Right. So the goal is, you know, how do you, you talked about hallucinations already. How do you anchor the data so that, you know, you can ground it as close to the truth as possible and not have the biases creep in? That's where explainability becomes really important, because even if there is a bias that shows up in the data, for you to be able to go back and identify how that bias made it into the insight and how do you eliminate that bias is a process that every organization is going to go approve in their models.
[00:38:48] Speaker A: Do you think companies be able to know if they have got those biases that are already injected in from the start, or will it be a retro set? Looking back to your point, the explainability side of things, we'll have to look at things retrospectively to understand. Oh, there was a vice there. Would you say that that's the case, or how do we start that off on the right foot?
[00:39:05] Speaker E: Yeah, I would say maybe like, you know, there are three interesting things for you to consider. The first thing is, as you're preparing the data and you're picking the data, it's important to make sure you have the right quality. And you're thinking through some of the most common biases so that, you know, you don't introduce them into your models, into your insights from the get go.
[00:39:24] Speaker C: Right.
[00:39:24] Speaker E: So this is a thing that you can be proactive about.
Are we gonna be comprehensive and catch every single bias out there? No, because even as humans, we're still learning, you know, things that we would take for granted turns out to be maybe not the right thing to take for granted.
[00:39:41] Speaker C: Right.
[00:39:41] Speaker E: You know, we were talking this afternoon about growth mindset and how all of us think about growth mindset is something that's an individual, but it depends a lot in the ecosystem that you're part of.
[00:39:52] Speaker C: Right.
[00:39:52] Speaker E: So that's a bias that they did not know before, but you could uncover that thing later and you're going to be uncovering some and fixing them. The thing that's also interesting is there are dynamic. As these insights get used for more and more valuable things, there will be active attacks on this data, like, you know, data poisoning. So people will be trying and Versailles actors will be trying to infinous biases into your models, into your insights. So how do you get ahead of that one? Because the first two you can argue are like things you can get ahead are things that are in a burt and that you could fix. The third one, the adversarial one, is even harder to deal with, like in all the misinformation we talk about, is actually one of those really good examples where the bias is not. It's an explicit bias that's introduced by, you know, folks who want to poison your data.
[00:40:47] Speaker B: Right.
[00:40:47] Speaker E: So I look at biases from those two lenses and then the posture are more easier. The third one, you need help of technology.
[00:40:55] Speaker A: Third one's really interesting because again, as cyber security practitioners, Dorothy myself, we're already still trying to keep our head above the water. And now you're saying potentially cyber criminals are going to inject biases into your model. So you may not even know.
[00:41:08] Speaker E: Yeah, imagine if you're a bank and all of a sudden a bias gets introduced into the model to be lenient to a certain population when you assess their wrist, or to be hard on another set of population as youre assessing risk and granting loans. So all of a sudden you did not want to discriminate the population out of the gate, but someone's introduction or poisoning of your data ends up making the models behave a certain way. So that's a challenging scenario.
[00:41:38] Speaker A: So what are you moving forward? How are people going to handle this? How are people even know? And are we going to live in this world of feeling constantly stressed around honor? Has someone injected in biases into my data?
People already getting burnout, especially in the cyber security space, and now we're saying we're going to add on more things to their plates.
[00:41:56] Speaker E: Well, the good news is, you know, everything that you are doing already as a cyber security person, things like protecting against ransomware, things like for that to make sure you have immutable copies of your data, things like, you know, right, security and policy posture that goes with the data, they're the same things that will actually help you from the examples on data pisin. In addition, there are companies like yesterday, we talked a lot about Nvidia and MIMO retriever. MiMO comes with its own guardrails, and the guardrails protect you from getting biased as to creep in.
So I think there's going to be technology investments to help us. And a lot of the things goes back to the data hygiene. So hopefully, if you had good data hygiene, if you had good security practices, that goes a long way.
[00:42:46] Speaker A: What if he died then?
[00:42:48] Speaker E: So I think if you actually, there's one thing that George said in the keynote that I think is really, really profound. The organizations that we see are making the most of Aon or organizations that are actually prepared for this data tsunami.
[00:43:04] Speaker C: Right.
[00:43:04] Speaker E: If you think about healthcare, or if you think about financial as a vertical or as an industry, because of regulation, they had to prepare and guard their data, and their data quality is much higher. So they are the ones who are actually first movers in the context of AI to take full advantage of the capabilities of applying AI to the data. If you're not prepared, it's going to be a longer journey for you, because data preparation is where 80% of the data scientists and data engineers time is spent before things become easy and you can convert them into insights.
[00:43:44] Speaker A: But would you say majority of customers are prepared? Did they start off being prepared?
[00:43:50] Speaker E: I think the preparedness of the customer raised a lot, industry by industry, based on the size of the customer. How long have they been on this journey? So companies like us, our job is actually to help customers prepare as fast as they can, to help democratize AI, to help them with this data preparation.
[00:44:12] Speaker A: With where you are in this moment, as of today at Insight 2024, what do you think is sort of worrying or perhaps customers concerns at the moment, just from your perspectives?
[00:44:22] Speaker E: Yeah, I think we talked a lot about.
[00:44:25] Speaker B: There are maybe like three of the.
[00:44:26] Speaker E: Real big challenges the customers have. The first one, when they think about data, is just the sheer magnitude of the data and the fact that data is fragmented. Data is everywhere. It's in the on prem, it's in the cloud. So being able to get a handle on the complexity of the data proliferation and the data sinus is maybe the number one thing all the customers are worried about. The second thing is you need to be able to take the data and convert that into insights, actionable insights, in a pretty rapid fashion here. And the two things that you need is technology and people with the expertise. Both are without technology. Use your friend, because the people in the expert peers, it's a custom industry and it's a scarce commodity.
[00:45:15] Speaker D: True.
[00:45:15] Speaker E: So that's actually a second place where I think the customers are thinking through, hey, what does this mean? How do I get ahead?
Because you're going to be successful on Mountbase and how soon you can actually get ahead on AI drive.
[00:45:31] Speaker A: Yelena's going to ask as well. You said convert into actionable insights. Can you give an example perhaps of what an actionable insight looks like?
[00:45:38] Speaker E: Yeah, me actually talked a lot about being, okay, I'll do an easier one.
[00:45:45] Speaker C: Right.
[00:45:46] Speaker E: We were talking about banks. Like imagine when we submit a loan application, someone behind some model evaluates whether your credit fees risk would be enough for the bank to do it. And once they analyze the Hornet risk worthiness, they also figured out pay for this risk. Watch from the rain D and if you think about it, both of them have really strong models behind it.
So those are actionable insights, because whether the customer gets a loan or not.
[00:46:18] Speaker B: Is really dependent on what the model.
[00:46:20] Speaker E: Decides with respect to whether you're credit worthy or not and what the rate of the loan should be.
[00:46:26] Speaker A: Who's the Turner focus announcer? Moving forward, what are some of your thoughts? And I know you don't have a crystal ball, it's just more so what you've seen the last few days and with your role in your experience, what are some of the things that you can start to see happening? And then as we I come and speak to you again and net up Insight 2025, what are your thoughts on that?
[00:46:45] Speaker E: Oh, I think from our perspective, everybody is going to have the infrastructure that meet that they need to work on their AI products, right. I think that's the easiest thing to get to.
The idea is can you also get to the same level of data preparedness, and can you actually bring these data services and the ecosystem and the ecosystem of partners together to stitch that end to end solution together.
So from our perspective, the companies that can actually put that solution together end to end, are the ones that are going to be really successful.
[00:47:22] Speaker A: And how do you see NetApp's role playing into the future of AI and intelligent data infrastructure, for example? What does that look like?
[00:47:31] Speaker E: Yeah, I think as a intelligent data infrastructure company, we really believe in helping you make your infrastructure and data AI ready. And when we talk about making your infrastructure AI ready, it's about giving you that performance and giving that to you in a way that's actually scalable, that sustainable, to give it to you with the right power utilization and efficiency. When we talk about making your data AI ready and to be able to bring that data, to bring AI to the data. Now you're talking about the role that we play in as stewards of more than 50% of unstructured data that's actually out there in the enterprises. We think about it as bringing that structure over the unstructured data, being able to actually bring compute closer to the data and being able to get to those insights in place. So that's how we look at our role, and we are pretty excited about that role.
[00:48:36] Speaker A: And there you have it. This is KB on the go. Stay tuned for more.
