Enterprise Grade Artificial Intelligence with NetApp

Thanks everyone for joining. Uh, thank you delegates for joining us here today at, uh, NetApp Insight here in fabulous Las Vegas. This is obviously our annual, um, user and partner event that we bring together, uh, data professionals from around the world to learn about the latest in, uh, intelligent data infrastructure.

So what I'm gonna do today, um, just have a very short intro for you that almost is a recap. I know many of our delegates got a chance to go to our fabulous keynotes. I know for some of the people watching on the live stream, some of you may have watched the keynote, some of you may not.

So consider this, uh, me doing an hour and a half keynote in about nine minutes with delegates asking me questions in the middle. And so that's, that's the test that I've given myself. Um, so obviously what we talked about and what we introduced here at NetApp Insight was this whole concept of really talking about enterprise grade artificial intelligence.

Um, I should note throughout the day, um, we may talk about things that are going in the future. We won't talk about things that are non-public, but as we're talking about our vision, obviously our vision is subject to change. So just, just so that we're aware of that for the audience out there.

Um, so obviously everything we do at NetApp is around data. One of the things, one of the things that's kept me at NetApp for almost 18 years now, um, if you can believe it, is that we have what I think to be a very important focus, but it is just that a focus on data and around data infrastructure. We're not a, um, large conglomerate holding company that does everything for everyone, and there's nothing wrong with that.

But at NetApp we focus just around data infrastructure and specifically around making intelligent data infrastructure, which we believe to be at the heart of really major customer imperatives around modernizing their data center, transforming to the cloud. Um, and then some of the major imperatives of the time, both adopting artificial intelligence, which we'll talk a lot about today, um, as well as promoting cyber resilience and protecting your data. You can't have good AI if your data is, um, suspect, right?

So both work hand in hand. Um, so those are what we call our four customer imperatives that we've built out an entire data platform, the NetApp data platform to help support. And one of the things that we've really talked about at this conference is unveiling the, um, NetApp data platform, unveiling the NetApp data platform.

And it's not new technologies in most cases. There's this new thing called the AI data engine that I'll talk about in a second. But this is really about how do we bring order, how do we bring sense to what our offerings are and start to talk about ourselves as truly a data platform company.

So we obviously build all of it of unified storage, having a single data plane built on ontap, right? Which is the same operating system that we've been building and hardening and making more resilient since 1992. It's obviously been wholly rebuilt.

It's obviously, you know, been rebuilt from the ground up, um, about 10 years ago. But we can continue to enhance on it and build out a digital estate that can go across on-prem storage. It can go across public cloud storage, it can be built into neo clouds to help build AI factories.

It can be built directly into sovereign clouds. And obviously we're the only storage offering with ONTAP that is native on all of the public clouds. Uh, and we think that's a huge advantage for us.

Obviously, we spend a lot of time building protection on top of it and then making it available as a service, making it available through open source and open APIs, making it available through a broad system of partners and ISVs that we integrate with, as well as building a complete open API package so that you can integrate with it, uh, directly through your automation, through multiple different automation stacks, uh, or through the control plane that we offer, um, including the GUI that we newly renamed the NetApp console. Okay. Are there questions from the delegates real quick?

I know this is just foundational, um, and we got a lot to get through, but are there questions just about this platform in general? I don't have a sense for the extent of the public cloud storage Okay. Abilities if that's done through partnerships.

Obviously we know about the major three. Yes, there's a host of not even specialty, but enterprise clouds. Mm-hmm.

Yep. So there's, there's a, there's a broad list of that, obviously. Yeah, there's the major three, right?

So there's the starting with Azure NetApp files, then AWS, um, FSX for NetApp, ontap, and Google Cloud NetApp volumes. Um, but then there's also, you see multiple different clouds that are built on NetApp, um, and use NetApp as the underlying storage. And then you see others that expose ONTAP as a service, right?

So, um, I won't go into all the different names here because, you know, I'd have to get permission to say each logo, but there are, you know, hundreds of clouds out there that expose ontap either indirectly or directly to their customers as a service, have it available on their marketplace, or have different services that are built as a service on top of ontap. Um, and that continues to extend over time. Great.

Yeah. And then from the sovereign cloud perspective, we're working with multiple different, um, regional and, um, nation state based clouds, as well as obviously the sovereign cloud domains within all the major public clouds as well. Okay.

So I want to kind of quickly cover the major challenge that we're trying to deal with on ai. We're not trying to solve every AI problem in the world, right? We're not trying to solve, um, we're not gonna be providing an LLM anytime soon, right?

We're not doing any of that. The problem we're specifically focused on is the challenge of getting AI ready data, right, of solving the garbage and garbage out problem. It doesn't matter if you have the best data scientists, it doesn't matter if you have the best models.

All of those are challenges to overcome, but we know there's multiple stats out there. George, I think gave one, uh, our CEO gave one about, um, you know, 80% of projects having challenges with AI ready data. Gartner has another one that says 60% of AI projects will be abandoned through the next year just due to lack of AI ready data.

They have the models, they have the scientists necessary, or they're using the public cloud for that, but they don't have their data in a ready state. And by ready, I mean both accessible across their entire data state and also in compliance with all of their standards as well as governmental and regulatory standards to be able to present that data to the AI model of their choice. And so they have everything ready to go, they've got the car all built, but they just don't have any gas to put in the engine.

So the announcements that we basically made are all around this idea that AI is now an enterprise workload. And if AI is becoming an enterprise workload, especially with the rise of agentic ai, if AI is actually taking action, then AI goes from being a science experiment to being something that needs to be available 24 by seven by 365. If AI is starting to independently act, if it's starting to act as an agent of your company, whether interacting with customers or performing backend maintenance, then it can't simply go down.

The analogy we give is you can't suddenly have, you know, if you have a company of 800 people, you can't have 400 of them blink out of existence for four hours and have no impact on the company, right? That's what agen AI looks like. So it's gotta become an enterprise workload, which means you need to have enterprise grade availability, cost efficiency, you've gotta be able to actually demonstrate return on investment.

We're past the science experiment phase of ai. If you're gonna be deploying AI at scale, then you need to be able to demonstrate that you get ROI out of it. And obviously you need to have things like built in ransomware protection and unified up-to-date assets.

So throughout today, we're going to have experts that come in to talk to you at a high level about our portfolio as well as around, um, two very specific announcements that we made at this event. Um, one was really around the foundation for enterprise grade data platform for ai. And by that we mean having all that enterprise resilience and performance.

We built up with ontap built on a unit unified data foundation, and then building on top of it. And this is the newer part that we've launched, is an accelerated data pipeline built directly into that. So the two things we launch, and then I'm gonna pause for a couple minutes and take all your questions and, and then we'll move on to, to our, our next speaker.

Um, two things. We launched NetApp A FX, which is a new disaggregated storage system paired with the NetApp AI data engine with, which is the streamlined AI data platform that works on top of it. And together we call them the NetApp A FX AI portfolio.

And that's a combination of this new disaggregated storage system, which at its core runs ontap. And I think if I could get one thing across about the A FX is, it's the same ONTAP that runs on all of our different systems, but with specific enhancements to allow it to be fully disaggregated into a modern architecture built for the largest exascale AI workloads. So you get all of the data management capabilities, all of the data mobility, all of the data security, all of the data efficiency that you normally would get with ontap, and you get all of the resiliency we built by running into all the sharp corners over the last 30 years, which is the only way sometimes you learn on some of these things, but you get it on this disaggregated architecture that can granularly scale both performance and efficiency.

And we'll have some experts on the a FX coming in a little bit later who can really deep dive with you about how each one of those features works to be able to do that just Granular granularly scale. So it means that you can scale out in increments of, you know, you just add controllers when you want to add performance and you add enclosures when you add want to add capacity, right? So in a fundamental disaggregated architecture, so it's on a node basis, mark, it's on, yeah, it's on a node basis, and you can just add individual storage enclosures.

And on the back end, and we'll go more into the architecture throughout today, but on the back end, it's all connected through a high speed RDMA offloaded 400 GigE network. So every storage enclosure can see every controller, um, in our number, you know, NetApp recently became number one in Flash, and a large part of that is our NetApp a FF product that's gonna be continue to be our absolute lead and lead the market in unified in that particular environment. Um, every ha pair, although it's scale out, has its own dedicated stack of storage.

So as you add ha pairs, you have to add some additional storage to it, and that works for 90% of today's enterprise workloads. But as you move to ai, you have AI requirements where your amount of capacity is tiny, but your performance requirements are through the roof. And then you have the opposite where you have a massive sort of data lake, but you've got a pretty nominal, um, sort of inferencing or rag workload working on top of it, right?

So that's what we mean by being able to be, um, both granularly scalable as well as linearly scalable. As you add additional nodes onto this, as you go into the hundreds of nodes, you increase performance. Okay.

One nit pickety follow up, which is you used the phrase scale out. Mm-hmm. Um, I am super annoyed with the industry over the last years.

Okay. Uh, because, uh, if you talk to data center managers or whatever, scale up is in the rack, scale out is across racks, you know, and then if you talk to some, uh, vendors, you know, uh, scale up is within an application or a stack scale. So, so I I, are you or is NetApp thinking of scale up versus scale out?

Is in scale out is more enclosures or scale out as more racks or, or is, are you, I hopefully just using one phrase to cover everything because I think we're using one phrase to, well, what we're saying is I don't use the rack constraint, but what I fundamentally say is if everything is within, uh, a single node or ha pair, and you just add additional capacity onto it, and if you want to increase performance, you switch out that ha pair, that's a traditional scale up model. Yeah. If you can add additional, you know, pairs into a cluster and then be able to move workloads across it, like almost like a VMware vCenter or vSphere store thing that's scale out disaggregated is when every asset can see every other asset and all of the workloads on it become essentially ephemeral and can move across all the different individual controllers or storage enclosures, and there's no more primacy or locality to your workloads.

Does that make sense? Yeah. Okay.

So that, that's how we talk about Yeah, that's, yeah. You could call a true scale out. We use disaggregated because scale out as a term, as you pointed out, has become somewhat polluted, right?

So, but I think disaggregated people fundamentally understand that concept. Okay. I Have a question.

Yeah. Um, I'm trying to find it in my notes. You said that, um, the new, the A FX is on tap with enhancements.

Mm-hmm. So can it exist with, um, uh, the NetApp without enhancements? Yeah.

Yeah. So the enhancements are specifically on handling the disaggregation on the backend. If you think of ontap, and this is fundamentally how it's architected as well.

There's this top half that is all of the protocols, all of the data services, all of the efficiency, and that is essentially unchanged. And then there's a bottom layer that is the more fundamental how you interact with your storage pools. And that's what's fundamentally different between our different, um, systems.

So with an A FF, it's built on the scale at architecture with a SA, it's built on an optimized small sort of pair architecture for san, when we go to A-W-S-F-S-X for a ontap, it's the same ontap, but we put a different shim on it that understands EBS as its discs instead of having SSDs. And so that's fundamentally a way to talk about, so the enhancements are to allow for disaggregation across the entire environment, but all the APIs are the same. You can snap near between them.

You can flex C between them. Everything that people are used to with ONTAP is there, but we just add these sort of enhancements to allow for disaggregation. Does that make sense?

Mm-hmm. Okay. So I'm getting the hook here very rapidly.

So what I'm gonna do is just tell you about the very last piece, and then we will go into this in much detail throughout the day. And that is the NetApp AI data engine. So you'll have speakers coming in in detail to talk about this, but just so people get the sort of 32nd pitch on it, the NetApp AI data engine is how we build an accelerated data pipeline so that our data engineers, our storage admins, can help provide that AI ready data, they can help curate it, use semantic search to find the right workloads, put in place data guardrails, and create a directly embedded vector database that's highly optimized sitting directly within an A FX environment.

What we're essentially doing is almost unveiling a new protocol. You could think of it, right? So instead of interacting with NFS or with Fiber channel, you're interacting with a secure rag AI endpoint that any AI application can go in and access a vector database.

That's always up to date, always secure and always compliant with your regulations.