Well, thank you very much, folks, for joining us. My name is Simon Leopold, Raymond James, data infrastructure analyst, and I'm pleased to welcome for our fireside, we have with us from Arista, we have Ita Brennan, CFO, retiring in a few months. So we're grateful that you're spending your remaining time with us, and Liz Stine, who's running the IR department. So format for this is fireside chat. I've got an outline of questions that I'll go through, but love to get questions from the audience. If you wave at me, I'll try to call on you, and we'll repeat questions, or, or, we'll check in towards the end to see if you've got any questions. Do you need to read any disclaimer or anything like that? Okay, awesome. I hate when I'm responsible for that.
That's-
We've got a major responsible.
I know. It's too much for me. You'd think I'd have it memorized by now. So I want to sort of reflect back on the recent Analyst Day, 'cause you provided an outlook for 2024, longer term. And so, you know, I like the timing of our meeting because we're, you know, December, everybody's thinking about 2024. So you've forecasted this 10%-12% outlook. Help us get a better understanding of what informed you in terms of whether it's thinking about the verticals, the trends, the cycles, back up what led to a 10%-12% growth outlook for the next year.
Yeah, so I mean, obviously, we're coming off of... Sorry. Is that better? How's that?
Nope. Might have lost our sound.
Yeah.
That one on?
I think it's on.
Okay, we're gonna try that again.
Yeah.
Yeah, we're coming off of, you know, 2 years of fairly aggressive growth and, you know, pretty elevated spending from cloud. So I think we've been saying for some time that we thought kind of 2024 would be a more, you know, moderated cloud year, right? And, you know, we are kind of coming out of the 400 gig cycle a little bit, and then you're starting to talk about and think about 800 gigs. So, it's been kind of our belief for some time that if we see cyclicality, this is kind of when we would see it. So that's definitely underpinning kind of the, the growth rate as we think about 2024. But obviously, the enterprise piece of the business, that has continued to do well.
It continues to grow nicely, a little bit accelerated in 2023 with some supply chain stuff, but even when you normalize for that, you know, it was kind of slower in 2022, a little bit faster in 2023, but when you normalize for that, it's still been growing very nicely, very consistently. So that's also an input. So taking kind of that and then thinking about that cyclicality, thinking about deployments that we know about and what we can see, discussions with customers, et cetera. I mean, all of that informs kind of that 10%-12%. And again, we like to have a couple of different ways to get there, 'cause it's early to be sitting here talking about the whole of 2024.
Having a couple of different models that can get you there is important, and then we'll see where we go from there.
So I appreciate that, that you're not guiding by vertical, but maybe it'd be helpful to sort of get some sensitivity around the three verticals you talk about, cloud titans, enterprise, and then service provider/other. Which do you sort of feel are above and below? Which sort of have the greatest variability? Which do you have the greatest confidence in?
Yeah, look, I think it's probably too early, honestly, to start to kind of pull it apart in that way. I mean, we do, again, have multiple different models that kind of assume different things about those verticals. That's part of kind of how you get comfortable with the view. But I think it's too early for us to start saying, "Here's a single path that gets you there." I mean, we've talked a little bit about the cloud just because it is—it's a driver of kind of the change in the growth rate. But other than that, I think we're just gonna go kind of quarter by quarter, and you know, maybe mid-year, we typically start to talk about a little bit more about kind of the customer and the vertical breakout.
And one of the common themes we've been talking about is the post-pandemic normalization. So companies sort of entered the supply chain constraints at a different point and built up different level of backlog. And so I've sort of looked at having backlogs as an insurance policy in the near term, but then it creates a tougher comparison later on. Where is Arista in terms of this cycle of adjusting to normalized backlog and normalized trends?
Yeah, so we don't talk about backlog and bookings, and we very much decided as we came into this supply chain cycle, that it was unhelpful, really, to talk about or to try to talk about bookings in a time frame where your planning horizon was going from a quarter to 12, 18 months, right? It's very hard to say that a bookings number has any real significance in that environment, right? So what we tried to do was to take the bookings and kind of turn it on its side and figure out exactly when these deployments belonged, almost, right? So where and when should they be deployed? You know, and we're continuing to do some of that still, right? Where we're making sure, you know, we prioritize shipments, obviously, early on.
Now we're making sure that we're kind of, you know, putting shipping to customers when they need it, when it needs to be deployed, et cetera. And I think that helps manage some of that adjustment that you're talking about. But there's still no easy way to go from, you know, 18 months of lead time to wherever we'll end up, 6 ±, right? You know, you do lose some visibility as you make that transition. We've been talking about that, the reduced visibility. But I think if you focus on the deployments, that gives you the best measure you can have of the business when you're trying to make that transition.
I do feel like talking as if there's a true normal is a little bit nonsensical... That is my preface. Are we closer to normal, or are we still sort of working through that?
Yeah, I mean, again, in terms of the business and the deployments, we've been, you know, we've been, that's the underlying business, right? So what you're seeing in revenue and, for us, is really the underlying business and when customers really needed to have that deployed, right? So I think from a business perspective, we've kind of, we looked through the bookings to kind of when do customers actually want to deploy stuff. You know, in terms of lead times, are we back to normal? I think we, you know, Jayshree had talked about it kind of improving, you know, 50% better than where it was coming into the year. I think by the end of the year, we'll be more or less back to kind of where we want to be from a lead time perspective.
Great. And I want to explore specifically your thoughts on the enterprise vertical. And really, the context of this question is broader around kind of the macroeconomy and how much that affects you, in that we're here debating, you know, are interest rates going down? Is the economy healthy? Did we dodge the recession? Did we not dodge the recession? You know, all these debates, and I want to get a sense of what's sort of your take on it. You have to have a view, and then how does it affect your thinking for forecasting in the business?
Yeah, I mean, it seems like we've been having this conversation now for years, right? I mean, really, since we came out of COVID, it's been a question of kind of what's, what's going to happen in terms of, of, of recession, et cetera. I mean, again, I don't know that we've seen that have a huge impact on the business, and part of it is, you know, we're, we're share gainers in that space. We're targeting accounts, new win accounts, et cetera. So we tend to be qualifying accounts on the basis that they have an intent to spend. They have, as Anshul would say, they have money to spend. That's the first thing you should qualify an account on when you engage with an account.
So there's some protection in that for sure, because you're not just exposed to the whole market. But having said that, if there were to be, you know, a serious macro impact, then I think it would impact us as well, right? I mean, there's no evading that if you had a serious kind of distortion in the market.
And sort of in light of that, the other angle is how you think about hiring plans. And really, as I'm sort of thinking about my group, I'm seeing many of my companies having to do layoffs 'cause they, they've overhired and things are slowing, where you're in a very different situation. Your revenue grew far faster than you could hire. So with the way you see the business evolving, what are you thinking about in terms of hiring objectives and sort of budgeting for that?
Yeah, I mean, obviously, you know, when cloud bursts in the way that it did, in 2022, 2023, we don't hire to that burst, right? 'Cause it doesn't really make sense. I mean, we have, we have a very kind of methodical, continuing hiring of software, headcount as we can find those, and that's become a little bit easier of late. And then on the sales side, it's really focused on the enterprise, piece of the business largely, right?
Mm-hmm.
So we, you know, we want to protect that regardless of what's happening with cloud to some degree, right? So that's, that again, that's more of a, you know, continuing hiring against that enterprise. As long as we see return on those hires, as long as we see productivity on those hires, on the enterprise side of the house, we'll keep doing that, right? So that's why, you know, operating margin sometimes will peak when you've got that accelerated growth, but then we might, we might call back a little bit of that as we continue to invest in sales and marketing, for example, for the enterprise, right.
I guess maybe a related question is, you seem to be under-penetrated internationally. How are you thinking about investment to grow your international presence?
Yes, we have been hiring. Probably a larger portion of the headcount that we've been done has been going into some of those international markets. And again, we'll continue to do that. It's been more developed markets, more of the kind of developed parts of the world, as our primary kind of target. So we'll continue to build out that, you know, enterprise footprint and add to that. I think we've been trying to target, like, a 30% increase in headcount, you know, every year. If we could do that, we think that's something of a sweet spot where you can do it and maintain productivity. So that's the plan. I think we never really got ahead of ourselves in terms of hiring.
Mm-hmm.
Like, we've always been very cautious around making sure that, you know, when we're hiring, we're being thoughtful about that. And, you know, thankfully, we haven't had to, you know, to really have a significant layoff.
So I'm going to pivot to the favorite topic, AI.
Mm-hmm.
'Cause what meeting-
That's right
...is complete without talking about AI? So you established this target of $750 million of AI-related sales by 2025. First off, help us understand what's included in that number. What's the composition, assumptions behind it?
Yeah, I mean, we're, we're trying to isolate kind of the back-end, AI revenues, so things that are connecting directly, directly connecting GPUs on the back end. Ethernet, obviously, 'cause that's, that's where we play. You know, it's gonna be—it's not gonna be perfect in terms of how you track that, but with, hopefully, with the help of customers, et cetera, we can do that for some period of time. I mean, I still think after, you know, I don't know if it's a year, 18 months, 2 years, you know, you won't care anymore what's an AI dollar versus a non-AI dollar. But I think for this transition period, right, where there, where there is this, these technology discussions and choices, it's helpful to try to give some indication of how we're doing there.
So we're going to hopefully work with customers to, you know, identify that back-end business, and that's what will be included in that, in that number.
And just to be clear, the opportunity that AI presents is not just about building AI clusters, but... and what you're referring to as the back-end, but there's also this front-end opportunity.
Mm-hmm. Yeah.
That's not part of the 750.
That's not part of the 750.
I think what you've said on the, on the last call, maybe it was at the analyst meeting, is, "Hey, we're not sizing that because we don't know how to ... create a dividing line?
Yeah. Yeah,
So-
It's the same products, right?
So is there a way to sort of think about, hey, here's the market forecast. If we subtract what we think is AI, what's left? Just, you know, what's your TAM, maybe, is the way to think about it. So AI plus non-AI?
Yeah, I mean, we talked about the TAM stuff at the Analyst Day, and you can see, you know, that the, if you look at the... What's interesting is to look at the TAM, say, from the last Analyst Day to this Analyst Day, right? We kinda increased the TAM by about $10 billion. About $5 billion of that is really AI and the AI opportunity in 2027, right? So that's a, you know, that's incremental from an industry analyst perspective in terms of how they're thinking about, and that's total Ethernet AI, right?
Yeah.
Now, again, that's not perfect today, for sure, because in some ways, we don't know what that split is, so it's gonna be hard for everybody else to track it, but it's at least starting to do a sizing on kinda what that AI Ethernet opportunity, full opportunity can be.
So one of the questions we're often getting is that InfiniBand is the choice for the back end today, and I think everybody in the Ethernet camp is singing the same theme of, in a few years, it's going to pivot to Ethernet. I'm an Ethernet-biased guy as well, so, you know, I like to hear that, as do my peers. But I think we're all struggling because the vendor of InfiniBand and the vendor of the GPUs really has a lock on that. They don't have a huge incentive to change. Now, they've said, "Look, we'll do Ethernet, too." But what sort of informs you from the marketplace, from the operators, that we're really going to see this transition by 2025?
Do you wanna take a shot at that from the technology perspective?
I think the benefits of Ethernet are really seen at scale, right? And as these environments get larger and larger, and you've seen, you know, some of the cloud guys put out, you know, some of these white papers now that are actually talking about their Ethernet environments and benchmarking them against, from a performance standpoint. And then, if you listen to them speaking, and the scale just keeps getting kind of larger and larger. You know, InfiniBand has some hard address limitations that limit the size of the cluster size, and when you think about Ethernet today, Ethernet is, you know, everywhere, right? And it can scale, and they're already running these Ethernet, large Ethernet cloud environments, and have the tooling and have the expertise, in order to do so.
I was joking with a couple of the investors we were talking to today. I was like, "You can Google how to troubleshoot an InfiniBand network and see kind of what comes up." But I think one of the other benefits, obviously, from an Ethernet standpoint, is it's open standards, right? And that's very important. It's also multi-tenancy, right? You think about carving up these large GPU clusters and wanting to share it among different customers, you kinda need a multi-tenant aspect. So I think that's- we see as these things start to scale, that's where we see kind of that tipping point le- favoring Ethernet.
And maybe to follow up, is sort of what's Arista's differentiation? 'Cause there's this sort of quandary of, well, Ethernet's a standard, so that means we've got several vendors that support it, so how does a vendor like Arista do better with Ethernet than the other people also doing Ethernet?
Yeah, I mean, Ethernet's a standard to date, too, right? So it's, you know, it's all about, you know, yes, there's a standard, but then there's also how performance and some products are better than others, right? I mean, for the same reason as we win at, say, 400 gig, if you think back to the 400 gig cycle, we had, you know, all of these discussions about how, you know, folks were gonna come and take share and take some of our position with some of these customers. But in reality, the product was capable, was delivered on time, it executed well. Those things are not for nothing, right? So, you know, everybody asks, like, "What if there's more Ethernet competitors?" A, it's not easy, and B, you know, we're used to competing for this business, right?
I would argue incumbency counts. If you're at an account with Ethernet, you're more likely to stay at that account with Ethernet.
Yeah, but there's no denying that every one of these product cycles, you need to bring... You have to bring your game, you have to bring your contribution from a roadmap perspective, from a technology perspective, and then you have to deliver really good products at that scale, right? And, you know, there's, that's, there's risk in that for us every time, but we're hugely focused on that. If you lose sight on that, of that, then that, that's your exposure, right?
Now, last year's analyst meeting, I think it was kind of an introduction, or at least an introduction to the analyst group, around this concept of Transit WAN, which was, like, I think, of Arista kind of spreading its wings, getting outside the data center. Can we get an update of where those opportunities stand and how that sort of fits into the forecast?
Yeah, I mean, we do, you know, we do quite a bit of routing revenue at this point, data center interconnect. We've reworked the software stack, if you like, the routing software stack for cloud, so quite a bit of routing revenue with the cloud customers as well. And then, you know, as we started to think about the enterprise space and, you know, the places where we were kind of, you know, had a gap in terms of the end-to-end solution that customers wanted and where customers were asking us to bring, you know, to bring an EOS solution to bear, and that was one of those, right? So we're early.
It's early days in terms of that deployment, but it's very important to customers to be able to see, you know, a product offering that will, that will fill that gap. I mean, to this Analyst Day, we talked about the NAC product. That's the same idea, where are you introducing something that you really don't need to introduce if you can do it yourself? It's core to the network. We think we can, we can drive a solution there, but it is early. I mean, I think when we had talked about it last year, you know, it was a concept, right? And so we're, we're kind of close to full product now, but it's gonna take some time to, to start to see revenue from that.
So, I wanna pivot to the campus part of the business. So you've reiterated your target, $750 million by 2025... which is easy to confuse with the AI target-
Mm-hmm.
'Cause it's the same. You should just pick 2 different numbers—so that we don't get confused. But we won't notice. But I want to reflect back a little bit on the strategy, because I think in the early days, the strategy was you were going to be more software-oriented and not get into sort of all these access points, and it looks as if you've sort of pivoted the strategy over time. You've kept the goals the same. Maybe before we sort of think about the outlook, I want to reflect what lessons did you learn in entering this business that's different than sort of your core data center business?
Yes. I don't know if that there was really a change in, a change in strategy or a change in fulfillment model, right? I think it was always our intention to, you know, sell the, sell the switches and sell the access points, and then, you know, maybe what's different is for like for an access point, really, all of the, the value, if you like, is in the, is in the management software, the management plane, which we run through CloudVision, right? And that's so that's still the case. You know, for the, for the campus piece, it was very key. Obviously, it was EOS, you know, all the way through, and again, CloudVision was managing.
If you had a data center, you could manage single image from your routing through your data center, through your campus to the Wi-Fi. So I think all of that has pretty much stayed the same. I think, you know, if I think about surprises that we've had, I think we were surprised how much of the business is coming to us as campus-first or campus-only, right? We had assumed that, you know, for the early stages, for sure, we would be, you know, really using our data center heritage and those customers to move into their campus solutions, and that it would be harder to win campus-only solutions. But we have seen a fairly balanced, you know, mix between new customers to us, so I think that's been good.
You know, that large enterprise part of that campus market is... You know, there's still a very dominant competitor there. So there are customers that are looking for an alternative to that, either because of the technology, because of their experience, because of the fulfillment model that they're driving. Lots of different reasons, right? But there's definitely not a lot of competitors that play in that space.
Mm-hmm.
There is a real opportunity there to, you know, to compete for campus-only. And it's nice to have a... You know, sometimes with a data center, you have to wait a long time for a data center opportunity, but now you can insert with a campus opportunity and something else as well, and one of those targeted accounts.
And can you talk a little bit about the go-to-market? You did mention targeting large enterprises, so I assume you win one deal, they're relatively big deals, rather than winning, you know, 1,000 sort of small businesses. So probably a little bit easier on the channel, but what's sort of your split between direct and where are you in terms of developing channel partners?
Yeah. So I mean, we're still, you know, our primary kind of near-term goal is to target that large enterprise customer, and that's the direct sale, right? And that's where, you know, we keep adding salespeople. We can continue to get more coverage, and add accounts there. You know, they don't necessarily pop in the same way. You do a data center, it tends to be a big spend. With a campus, you could do an initial campus and then another and another. So it's a little different from a spend perspective. You know, if you think about the market, it's about half of that, you know, including Wi-Fi, $40 billion market, about half of it is in that, maybe a little less than half, is in that large enterprise business part of the market.
So that's addressable that way, and then as you start to move down mid-market, that's where, you know, the channel becomes much more important. So I mean, we are adding channel partners. I think there's a relationship between how successful are you in the enterprise, because it drives the brand. Even though we don't necessarily have the channel to sell for some of those opportunities, they would be fulfilling some of those opportunities. They could be doing services for some of those opportunities. So the more that you are present in the enterprise, I think the more that channel becomes easier. But it is gonna take time. Like, there's just no easy way to...
You know, somebody was asking us today: "Can't you just turbo?" But it's gonna take patience because, you know, there's already some incumbents with large presence in the channel, so we have to kind of find those channel partners as we go. But we have... You know, we've got partners now who probably would never have engaged with us, you know, three years ago, four years ago.
I want to ask sort of a competitive landscape question, but I want to make it a little bit harder in that the easy answer is always, like, we're going to take share from Cisco. We get that. But when I think about the narratives from the bigger competitors, like in HP Enterprise with Aruba Central and Juniper with Mist, they seem to have very similar strategies to Arista's, where it's large enterprise, it's taking advantage of Cisco's weaknesses and complexity. So how do you compete against those kind of players?
Yeah, I mean, it's interesting, again, when you look at that, like, large enterprise piece of the market, there's not a lot of share that's going anywhere other than that key vendor in that large piece, right? There's still a lot of those players are still in that mid to maybe bottom of that large enterprise market, right? So our target is very much the big bet, large enterprise customers to start with. But we do, you know, we do see the other guys sometimes, but it's largely against that one incumbent still, right? I mean, that's where we're competing. Those are the opportunities that make the most sense for us right now.
And-
I don't know, product-wise, if you want to add anything.
I mean, I think value prop-wise, our campus story isn't really much different than the data center story, right? It is all about the quality of EOS, the operational efficiencies that you get with CloudVision, the visibility. And as the campus becomes more and more complex, and you think about campus isn't just a office building any longer. It is, you know, a hospital wing, it is a hotel, it is... You know, there are so many more definitions of campus, and those networks are becoming more complex. Customers need a quality operating system. They need the tools in order to run these larger footprints, and we give that to them with EOS and CloudVision.
... So do you have any certain long-term goals for this business? 'Cause my recollection is when you first announced it, there was sort of this aspiration of at some point being 10% of the market, undetermined time. But how do you sort of think about making it meaningful? Because at $750 million, you're clearly not done.
No, we would certainly hope not. Hope not to be. Yeah, I'm not gonna try and, especially now that I'm about to retire, to, to pre-
But you can say whatever you want now.
To pre-announce Jayshree's target for Chantelle's target for campus beyond that. I think we're, you know, we'll execute well against the 750, and then we'll go from there.
So, topic you and I have talked about before that I think is interesting in being different, is your, your thoughts on a software strategy. In that so many companies in my sector have sort of pivoted towards software because you wanna have a software model, you want to have recurring revenue, because that gives you a higher multiple. You don't need a higher multiple, so you've had maybe a different take, but maybe help people understand how you think of the software business within Arista.
Yeah, I mean, and we've thought about this a lot, right? 'Cause, you know, when you see everybody around you doing something, you have to really ask yourself, should you, and why aren't you? I mean, Arista is a software company. You know that because 90% of the R&D resources we have are software people, right? I mean, it is, it's a software company. That's what drives the capabilities of the product that customers are deploying. It happens to get sold with a switch, right? You know, we thought for a while, should we in some way take the core operating system, EOS, and say, "Okay, we're gonna sell this separately." But then what are you selling as a box, right?
You're selling a box that doesn't work, or you're gonna sell a box that, you know, the customer can use for a year, and then somehow it's gonna stop working. It was just a lot of friction with customers. Customers, large customers generally want to capitalize their hardware and software together. They wanna have a perpetual license, they wanna buy it, and to us, honestly, the economics of it are not that different, right? These customers are smart enough to realize that if you, you know, if you do a subscription and it costs, you know, 50% more, it costs 50% more, right? There was really no substantive benefit from a business perspective, and if anything, it was just causing churn and friction for customers, and so we didn't see that there was a need to do that, right?
But for offerings that are standalone, value add, like CloudVision, like the visibility tools, the security products, those we sell as a subscription license. Customers get to choose whether they buy it or not. It doesn't, you know, it doesn't hinder the switch from working, it just, it's their decision if they see value, they can buy those. So those are software subscription licenses, and, and we're selling those, and you see those show up a little bit in services and in the product revenue line.
How material is that within the business?
I think the last time we talked about it, it was kind of single digits. It has kept pace roughly with the business, which is not nothing, given how we've been growing, right? But obviously, it gets dwarfed again by the product, just because the product has been growing so fast. But yeah, so it's a contributor to gross margin. These are very important offerings from a customer decision perspective, from a strategic discussion with the customer, et cetera. But we still believe that, you know, taking EOS and trying to somehow artificially separate that is a bad idea, and really, recurring revenue means I win at 40, I win at 100, I win at 400, I win. That's recurring revenue in our mind.
Sure. Let me check with the audience, see if anybody's got questions. If not, I've got plenty.
Sure, we'll repeat.
Earlier, coming back to... Will that probably be going to next year with backlog as well?