Thanks for joining us. My name is Simon Leopold, data infrastructure analyst here with Raymond James, for our third day at the Tech and Consumer Conference here in New York. I'm pleased to welcome with us; we've got a session with NetApp this morning, and Phil Brotherton, who is the director of Solutions and Alliances.
Vice President, baby.
Vice President.
I've been promoted, yeah.
Awesome. So format's gonna be fireside chat. I'm going to read a safe harbor, so that I've been entrusted to do that. And then we'll get into the Q&A, and we'll be happy to take questions from the audience as well. First of all, today's discussion may include forward-looking statements regarding NetApp's future performance, which are subject to risk and uncertainty. Actual results may differ materially from the statements made today for a variety of reasons described in NetApp's most recent 10-K and 10-Q filed with the SEC and available on their website at netapp.com. NetApp disclaims any obligation to update information in any forward-looking statements for any reason. Did that. Awesome. So let's get into it. So we did clarify, you got a promotion title-
Yeah
... but for folks who haven't met you yet, tell us a little bit about your role at NetApp and your history-
Sure
... with the firm.
Sure. So, I joined NetApp—I'll give you a real quick chronology. I joined NetApp in 2004, so I've been here a long time. I've seen a lot of the evolution of the company. For those of you who don't know NetApp, we started in the 1990s as a dot-com darling. We made a lot of money and a big IPO and stuff, mainly on the dot-com bubble, for those of you who remember any of this. And when the dot-com bubble burst, the company was largely in technical workloads, is what we call them, so software development, chip development, things like that. And our file servers are perfect for that, and those remain... Some of those big Silicon Valley companies remain, not just Silicon Valley, but they remain our biggest customers.
When the dot-com bubble burst, the company pivoted. I joined around this time. The company pivoted into focusing on what's called the enterprise segment of storage, which is largely about running file services and databases and things in big enterprise workloads at big banks, manufacturing companies, things like that. And we built a big installed base over the two thousands on that focus. And then the next big thing really was the cloud. I'm skipping the technology evolutions. I'll tell you real quick what I do, but the business evolution went then towards the cloud, where customers are adopting cloud, and we put a lot of attention into hybrid cloud. And then I think you'll see that this next wave of AI, I think, is quite real and will be the next big focus for us.
So those are, like, the big mega waves of the company. For me, personally, where I work, I always work at the interface of our software. NetApp is fundamentally a software company powered by appliances, so people love to talk about the gross margins and the hardware and everything. I come from the software side, primarily, and I work mostly with companies like VMware and Oracle, and Microsoft, and Amazon, and Google, and NVIDIA, and all these companies where we connect up into their ecosystems and, and connect our operating systems to the big partners', products. The reason you do that is so that the customers, when... Like, if I'm buying a big AI system, I need compute from, say, NVIDIA. I want storage that supports my compute. It needs to work easily and seamlessly and plug into all the various tools.
That's what my team does.
Great. So I want to start out with some recent disclosures and then really get into kind of big-picture stuff. But I, I think one of the topics we, we sort of were surprised about is this strategic review of public cloud services that the company undertook-
Mm-hmm
... and completed. So help us understand a little bit about what happened, why it happened, and then maybe-
Yep
... help us understand what is actually going on with public cloud services today.
Yep, yep. So when I actually would say, in my career, the biggest change in the overall business model is driven by the cloud. So the reason I say that is, when you look at most computer companies from the Silicon Valley, say, they have a pretty standard go-to-market model, where if you think about it, they either sell direct or they sell through distributors and VARs. We sell boxes mostly or, you know, appliances, and we sell on capital. You look at the cloud model, it's much more of a consumption-based consumer model. It's more of a service than a product sale, that type of thing.
If you've watched this carefully, you've seen companies like Microsoft really work hard to make that pivot from selling EO, in their case, EAs to consumption, and it takes years. We started that journey back in 2014. I actually started our cloud port of our software and the beginnings of this back in 2014. We're 9 years into it. We are pushing real hard that direction. To get to the question you asked, we've made some acquisitions. If you look at our business today, it's grown. I remember how hard it was to get to $1 million ARR back in 2014, 2015. We're now towards $600 million. It's 60% cloud storage, so ONTAP is about 60% of that total, and we've done some acquisitions, a string of acquisitions to expand past that.
Some of those have played out really well, like Spot, Instaclustr, Cloud Insights is a homegrown one. A couple of the others we wanted to clean up and get our, you know, the ROI back tuned, basically, and that's the gist of what we were talking about on the call.
Great. So it sounds really like it was much more of a tweaking than a major-
... Oh, for sure. For sure. It lets us put more energy into where we know there's growth, which is, we call them the 1 P cloud services, cloud storage services. But we're the only guys who've got our software built right into the consoles of Amazon, Azure, and Google. And so NetApp file services are available directly through those guys. Not— You don't have to buy from me, you can buy it straight from Microsoft, straight from Azure, on your big purchasing contracts and things. And we're seeing really good growth in that part of the business, and then some add-ons. So if you think add-on cloud services on top of that is a way to think about other parts of that portfolio. And we wanted to put all our energy into those, the ones we think are hot and growing.
So I tend to find... I get questions about what actually is public cloud services, and I think you've kind of touched on it, where I think sometimes there's a misperception that the customer is AWS or Azure.
No, no, no.
Um-
They, in a way, they are. They're like a channel partner almost.
Yeah.
It's. Let me use AWS as an example. AWS is a platform, right, that you have tons of services you can get on AWS, if you're familiar with them, and one of them is a NetApp file service.
Mm-hmm.
They have two ways of delivering. This is true of Azure, it's true of Google. You can either have a marketplace product, which is, it's a lot like eBay or something, actually, but you have to go find it as a customer and stuff. The easier way to procure is actually if you're in the management console of the mainstream cloud services of Amazon or of Azure, and so that's what we've worked to move from the marketplace to being what's called a first-party service.
Mm-hmm
... of each one of those. That makes it super easy to do things like integration with, like, AI is hot right now. Gen AI is super hot, so there's tools like Bedrock on Amazon, and you can do integrations. It's much easier to do integrations with upper-level tools when you're in the, in the main consoles. And that's our core, that's the 60% of our, of our public cloud business. Most of the other piece—the other pieces are add-ons from a customer point of view. Some are really discrete, like there's an Instaclustr product that's pretty discrete, but that's how the... That's how our customers and our sales force look at our public cloud services.
And I think early on there was a perception, and I, I think I'm guilty of it as much as anybody else, of thinking: Well, it's just NetApp on-prem customers would be the customers that would embrace it in PCS. And you've seen-
No, I-
... somebody new to NetApp. Could you talk a little bit-
Yeah
about why that happens?
I'll tell you a quick story. I was at re:Invent last week. I won't name the customer, but we were talking to a service provider, so a guy who sells software as a service and runs on Amazon, doesn't actually has evacuated data centers, not a NetApp customer. And we were talking about, in this case, it was FSxN, which is the Amazon-based first-party service, and why they'd adopted it. And, in that case, that example, that customer probably won't go back to data centers. That's gonna be a cloud customer forever. And it was a good example of being in that first party, being in the consoles of AWS, was so critical to us. We've seen a lot of that kind of business.
It's great because there's a connection if you're an ONTAP user. There's a logical reason to use ONTAP on the cloud. That's kind of an easy; we call it lift and shifts, taking apps that exist on-prem and that. That's a good market for us. There's just basic value in having the world's best file services on the public clouds, and that's attracting these new customers that are... They could be Dell or HP or other customers on-prem.
So I want to pivot to sort of the hottest topic, of course, which is artificial intelligence. So maybe just at a high level, start out with discussing how you see the AI opportunities, generative AI, broader AI, affecting NetApp's business.
Yeah, as I said, we're thinking it'll be a big, big growth driver. The way to think about it is we've been working in AI for years now. We've had a program about growing our AI business for five years. We have hundreds of customers, mostly doing what's called predictive. I think the new term for ML is predictive AI. And then we have a handful of people really getting into Gen AI. What we continue to see is, when you think about this at the storage layer, not the compute layer, 'cause they're quite different, is at the storage layer, there's a whole flow of getting your data ready to be used in these models.
So that usually starts with, actually, object storage is the most frequent way it starts, and we have a cool product called StorageGRID. That's an acquisition we did years ago now, and it is really starting to see pickup because, I think, partially because of AI. But you have to get your data in order. That's usually in objects. And then, as you move towards the training models, you move towards high-performance file services. And we've got—this is where hundreds of customers use our file services against GPUs, and where the basis of our relationship with NVIDIA lies. And what we think is gonna happen now, the big change with Gen AI from a storage point of view, is these things called vector databases get involved, and you'll start hearing all these terms. And it's a wild, wild west right now.
There's, like, 20 vector databases and all this stuff going on and, and we're-
Those are basically where the data lives, that's used in training?
It's the conversion of... Let's see, you're gonna test my knowledge of Gen AI. You, you have to-
I'm not smart enough, so-
You have to convert-
Say whatever you want.
In text, you have to convert words. Computers work in ones and zeros, right? So at some point, you gotta change the words into ones and zeros, and a vector database is one of the tools they use to do that.
Okay.
So anyway, what we think is gonna end up happening is there'll be this notion of a data pipeline that's sort of AI-powered. And to be the data pipeline provider, you have to be great at unstructured data and cost and reliability and all the fundamentals, but you also have to be great at data movement. The other thing you're gonna have to be, a lot of these AI apps are starting on the cloud, as we've all seen, and we've seen from predictive AI that customers that do a lot of work on the clouds and are starting to see big success, often put up an on-prem infrastructure that mimics the cloud. And they do it mostly for cost savings, to get cost savings. There's other security and data...
Like, over in Europe, they're well more into data privacy or like, sovereignty. So there's different reasons, but we think it'll be a very hybrid use case in the end. And because it starts on the cloud, we're really putting a lot of attention into our first-party services and integration with the tools on the various hyperscalers. One last thing on this, and it's just a big brag. We demoed. In the last few trade shows we've been doing, we're demoing. We have a technique where you can cache, basically, you can cache one ONTAP system to another, and it was built for, well, it's actually built for, like, chip design and life science use cases. But in Gen AI, you can do this caching technique between on-prem data and the cloud, and your data doesn't actually leave your data centers.
You put a cache over there on the cloud, and then the cloud can suck that into their foundation models.
Mm-hmm
... things like that. And so those things, certain things are interesting right now, capabilities. They're interesting, but they're early. All this is very early days. It's very interesting, though, what... You know, we think, like I said, we think this is gonna be very big for us, but it's early days, and it's definitely Gen AI powered.
And sort of in the theme or spirit of that kind of bigness, it's been, I think, relatively straightforward to think about the size of the market for things like GPUs and the size of the market for things like switches to connect GPUs. I've personally found it very difficult to size the opportunity for storage because it doesn't seem like that's-
Yeah
... super well defined of, well, how much storage do you need per GPU? How are you thinking about developing market models and market sizing? What, what do you look at?
Yeah, it's a good question. You're right, I think it is hard. I'll start with, I agree with the premise of your question. A lot of the data today—so when you look at the Gen AI models that you're hearing about today, basically, that is data off the internet. So the storage already exists, in a sense. It's not net new storage. Then you have to compile it, and actually, the size of the file for a text, like for a text foundation model, is relatively small. It's insignificant compared to the total storage market. So when you get right down to the storage per GPU at the foundation model level, that isn't actually a lot of storage, especially in text. As you get into video, it'll grow.
The big numbers start to become, in our world, things like Hadoop farms moving out of DAS and moving into shared storage. And also, I think you'll see this—you'll get replication of data, 'cause you're gonna take these foundation models, and you're gonna bring data that you have that's your own and merge them. So everyone expects their to—companies to have multiple models of their own.
Yep
... built off a foundation. You'll pick a foundation model, but then you're gonna work on top of it, right?
Yeah
... eventually. This is all, all speculation, by the way, to go back to safe harbor.
Yeah, it's too early.
It's too early to know for sure. But that all ends up creating a lot of data movement-
Mm-hmm
... if you will, which we do. Some of our value is things like this thing we call SnapMirror. It's in the data movement area. And it will create replicas of data. How big that market is exactly is still... I agree with you, it's hard to size.
And so for any of the analysts following NetApp, you've announced a number of products. We've talked about sort of public cloud services, but from a platform perspective, which do you think are sort of the products that you're positioning for sort of the AI opportunities?
Yes. Yeah, so I mentioned the cloud services. So the cloud storage services on Azure, Amazon, and Google, those are very key.
Mm-hmm.
I think, and that's actually a main differentiator. None of my competitors have that starting point. So that's, because so much work starts on the public cloud, that's really important. In the on-prem side, the best price performance generally on, for the higher performance needs of the data pipeline is the C-Series. You've heard, for those of you who are following us, we introduced a new product called the C-Series about six months ago. Really good price performance for this use case. And then it connects back down into object storage.
Mm-hmm
... and object storage is StorageGRID, is in our product line. And I think that's the... Well, I don't think I actually get to spec this. The, that's basically what we tell our sales guys, and, and -
This sort of gets to, nicely, the next question I had, is that I think the marketplace perceives NetApp as sort of the file company, right? Classic enterprise-
Mm-hmm
... file management, and some of your competitors have sort of positioned themselves saying, "Oh yeah, file's kind of old school. We're the object company." And everybody says they do both. I really feel like I'm hearing this almost repositioning from NetApp about object, and maybe-
Um
... you know, for a financial audience, help walk us through this-
Sure
... a little bit of the nuance between file and object.
Yeah. I'm gonna get geeky. Well, if you haven't think I'm already geeky. The, so files, if you think, when you go files, what does, what do they mean? And the other one to put on your list is blocks, because most of my competitors, I'd say, are block companies, first and foremost. Blocks are used to talk to servers, so, like, databases talk in blocks, historically. We were the ones who pioneered that you use files for databases, but most of the market still, still, they all still uses blocks. Files are used. They're like your Word docs or your PDFs. So NetApp's biggest business is, like, for, just for example, is all the files that come off of, of financial companies that are provided to us as consumers. Many, many of those are being fed off of NetApp file servers.
The other one is, if you think about, like, writing software for a living, you create a zillion files when you write software. So those people who write software are our biggest customers. People who do a lot of observation, if you have a lot of cameras and things, those all create files. Those I won't talk about who those people are, but you could probably guess. Those people are our big customers. Objects is another level of that same idea, objects don't have the same speed and latency. They have some big advantages in scale. You could just put bigger object pools together. Files get complicated to manage at massive scale, okay? So what we've done is we've extended our file position, but objects literally are another word for files, basically.
So what we're doing is we've got our own object store, dedicated object store, and we've put all these techniques into how you connect our file services to our object stores. So, and, for example, the latest one we brought out is known as... This is where I'm gonna get really geeky. It's known as File-O bject Duality.
Okay.
So you can have an application do a file call to one of our systems, which the file actually lives in an object format. The system will reformat the object and push it back out to the app, and the app doesn't have to know that-
Okay
... whether it was a file or an object. So we're gonna obfuscate this difference between file and object, because it's how customers actually live. You've got file servers, and now you're bringing in objects. You've gotta. You don't wanna change your apps. You just want somebody who can help you ride the journey, and that's what we do.
Now, you mentioned-
By the way, that's our life is.
Yeah.
When you say we're the file company, I take a lot of pride in that, actually.
Yeah, it's-
I think it being able to manage unstructured data is a very hard problem.
Yeah.
That's why we get... That's why you see us have a 30-year lifespan and keep going.
So I wanna, I wanna talk about the NVIDIA partnership, 'cause everybody knows NVIDIA is sort of, you know, is the foundation of all things AI right now. So having an alignment with them, partnership with them, is important. But, walk us through, what is that relationship? What's it mean for NetApp?
Yeah. So, the way to think about the NVIDIA partnership from our point of view is, when you look at AI, again, AI is mostly unstructured data. And so when you look at, from NVIDIA's point of view, NVIDIA's primarily a chip and system company at the compute layer, and we have all this data sitting in all these enterprises that is available to connect to those. We have these big footprints and big users. So, like, the chip developers are big users of AI, software developers, life sciences. So these are all legacy long-term customers of ours. So coming together with NVIDIA had a lot to do with our market position when you get down to it, and our ability to work with enterprise, which NVIDIA is relatively new to the enterprise.
They're sure not new to selling GPUs, as you guys all know. And so that's been the root of it. And then there's technology integration. They have programs about that spec performance. We plug into all their MLOps tools. They have a long, big pile of software on top of their GPUs, and we plug into all that. And then we have built a good go-to-market program with their lead sales teams, and that's how we work with them. When you're talking about a big, important partner like NVIDIA, we're gonna work on integrating go-to-market, we're gonna work on integrating engineering, which that's actually my job, and we're gonna work on integrating service and support so that the customer gets a total solution when they buy NVIDIA and NetApp.
So I think you alluded to my next question a bit, but I wanna unpack it in that we've talked about the evolution of generative AI as being very hyperscale-focused right now, building these big training engines, and then eventually it sort of phases into the use of inferencing by enterprises.
Mm.
You sort of talked about this idea of bringing their data. What, what I'd like to get an understanding from you is your vision of really the timeline and these phases of when do we kind of shift to that market, and what do the different phases mean to NetApp?
Yeah. It's been, it's fascinating. The joke in AI right now, I'll brag again. I went to Tahiti a couple, for a month ago.
Well, that's enough right there.
Right. So I went for 10 days, and now I'm a month out of date in AI, because AI is really moving fast. But if you backed up even a year ago, AI was mostly in specific verticals where, like I keep mentioning, life science is a really good example of this. They're very successful, very well. You know, they know how to use AI. It's mostly machine learning. The ChatGPT was like: Oh, this is gonna go much broader than the chip development, the case, the markets we're seeing it in. And but everybody's like. You go, "There's a pause right now." Everybody's looking at it right now, let's say. They're thinking about what it means. They think that it's gonna be important.
There's a lot of legal implications still to be worked through. There's lawsuits all over the place about Gen AI. And if you—I'm not the world's best expert. So if you listen to, you know, leading people out in our industry, you hear a lot about this. So I, we're definitely seeing pilots going in today, is what I would call them. It's definitely demand constrained because of the GPU, you know, how many H100s and H200s Jensen can build. And so I, I think you're seeing pilots going in today. My suspicion, just based on history, is pilots will go on for even 12 months.
It'll take a while, and you're going to see people start to try things mostly internally first, kind of quietly, because of the legal implications of exposing Gen AI models to your customers. So we'll have to—we'll be working with our customers all through that phase, is how it's going to go, is what we're thinking.
Can you maybe address some aspects of the competitive landscape for you in these opportunities? So clearly, everybody's kind of talking about it. I think you've been out front in terms of-
Yep
various aspects, but what's the competitive landscape? What are the options customers have?
So... yeah, so the first thing, it's common in our industry, in the storage industry, the first thing that happens in AI is the data scientists will buy infrastructure for their problem, and they're not worried—storage isn't... You know, you don't wake up as a data scientist and go, "Gosh, I really want to buy storage." So they'll go, "Well, I want to buy the fastest GPUs I can do," because that'll make their job go faster. And they worry about how fast can the job run. And that first layer of the storage is oftentimes done in what's, in our jargon, a high-performance computing model. So it's often known as a parallel file system and just basically cheap, cheap disks underneath the parallel file system.
Why not flash?
It'll be flash-based, for sure. Sorry, I use disks in a generic sense, and I shouldn't say flash.
Okay.
Yeah, it's fast. It has to go fast, so it'll be flash. You're right. I'm sorry about that. Thanks for the correction. So the problem we have. So when you get into this one layer back, this is where the storage industry kicks in, and I'd say, like EMC, back 30, 40 years ago, pioneered this. You now need copies for... You got to keep copies of your training so that you can prove you did the right things. You need backups and replicas. Anyone who's working in a regulated world needs all these copy management underneath that. The Lustre file system is useless for any of this, and this is where ONTAP, where we live, is that kind of a problem.
So we end up walking in, often these systems are being purchased by departments, and then the IT guys will walk in. These are our, the IT guys are our customers, and they'll say, "Look, I can't even manage this thing. It needs all these things." And so we get involved. So my first answer to your question is, just getting people aware that, "Hey, there's this set of requirements you should do, and here's the reference architecture." That's where we are right now in this market. When you look at direct, once that, you've gone through that step, there's a few startups that are pretty interesting, I'd say. I think we're pretty interesting in how to do this. That's basically what the competitive set looks like right now.
Okay, I also want to ask you about the VMware partnership in that, in one respect, you've had, you know, a year to think about it, but now, the acquisition of Broadcom of VMware is closed. Does this change things for you? Is it too early to say? Do you have any predictions?
Yeah, I-
as to like what happens?
You know, I started the VMware relationship at NetApp back in the mid-2000s, and it's been. They're an amazing company, how they've grown and changed our industry. Oh, to give you how important this question is, roughly half our installed base talks to our controllers talk to an ESX server, a VMware server. So this is a super important relationship to us. The basics of it, I don't think anything. We actually upped our investment in VMware about 3 years ago. The partnerships always ebb and flow a little bit, but when Dell and Silver Lake took VMware into the model with Raghu as CEO, that actually. We're friends with Raghu. We've been working with Raghu for 20 years.
We upped our investment, and we have a really good pipeline of work going on right now. I don't expect, where we work with them, I don't expect that to change at all. We're certainly not changing our position one iota, and the teams I'm working with at VMware haven't changed either. There's other things going on in the industry, where people are modernizing, they're trying open-source alternatives. There's all kinds of things going on beyond just the VMware partnership, but it's such an important installed base to both of us that we're going to keep working together quite... I don't expect, as I said, I don't expect substantial change.
You know, we've spent a lot of time on sort of the whole AI opportunity. Before we run out of time, are there other sorts of developing aspects in the industry and the marketplace that are most exciting to you? What's... Or is it, are you just AI-consumed?
Well, I'll go get to it real quick. There's this continued evolution of the data center. This is a little bit of a detailed answer, but there's this continued evolution of the data center towards more of a cloud, more of the way the clouds are built, with Ethernet as the backbone of all the data centers and a more abstracted connection between the servers and the storage. And we pioneered this with running what were called Oracle grids back 15-20 years ago. VMware, we, a lot of the customers run us in a somewhat unusual way, with NFS as the connectivity. They abstract the server from the storage with NFS on VMware. This is what, how our big service providers do it.
I see the economics of doing it that way versus the old way, and I'm like, "It's got-... There, there's still so much just modernization of data centers to do. And then, the cloud connectivity obviously is an important piece of this that's going on, too. AI's like a whole new application stack. That's why it's-- so it's really brand new from... It's, it's not the 30 years of client- server, it- the Bell Labs we were talking about.
Right.
It's the new world of cloud-based app development. So it's truly new and exciting.
We got a question in the back?
Yeah. So you, you rolled right into what I wanted to ask, but we heard from one of the large VARs that they really need AI go lives in the enterprise, probably more in 2025 than 2024, the sort of data prep and management year.
We're-
How do you plan-
Yeah, we're seeing more-
Let me paraphrase-
Yeah, okay
... for the podcast.
Yeah, go.
So basically, the question is that we've heard that AI in 2024 is really about preparation, getting set up, and then 2025 is the year of enterprise adoption and sort of the big rollouts. Is your view-
Yeah, I'd agree with the VAR you talked to, basically. I don't know about the timing. The timing's anybody's guess, I think. But the, we're definitely seeing an uptick in projects that we would call analytic projects, not AI projects, that are in the data prep for AI kind of mindset. It's one of the reasons you'll hear me talk about data pipelines, because it, right now, like, analytics and AI are like two segments. They're being bought separately at the moment, but fundamentally, there's a flow of data from all your edge devices. So whatever your edge device is, right? There's a flow of that data into low-cost storage, and then there's a prep stage and a flow, and eventually, that makes it all the way up to the models.
Companies that want to use AI really aggressively are gonna want to manage that whole pipeline, and we think the big opportunity for us will be AI-powered, but the big money is gonna be in the pipeline.
Well, great. We're out of time.