Great. Thank you. Thanks, folks, for joining us. My name is Simon Leopold, Raymond James Data Infrastructure Analyst. We are here at day two of our TMT and Consumer Conference here in New York, and I'm pleased to introduce for our next fireside chat from Arista. We have Chantelle Breithaupt and Ashwin Kohli. Chantelle is the CFO, and Ashwin is Chief Customer Officer.
Customer Officer. OK. For some reason, I thought you did strategy. Did you used to do strategy too?
Nope. OK. Just my imagination.
Ashwin's always strategic.
Yeah. Well, we've met many years, and this is our first time. So I really appreciate you guys joining the conference. So fireside chat format, so sort of free flow. I've got some notes that we'll work from, and maybe I want to kick it off with the outlook for 2025. So that's sort of the great timing about having a December conference, as we're all sort of thinking about the next year. So Arista forecasts a 15%-17% growth rate. Can you maybe unpack a little bit the drivers and key assumptions behind this outlook?
Yeah, happy to. And thank you for having us here, Simon. It's great to be here today. So if you look at Arista's guide going into next year, the 15%-17% revenue guide that you mentioned, that scenario is based on what we anticipate to be more of a mix in cloud and AI, cloud AI, is how we're mentioning it. And so we're very excited about the opportunity as we go into 2025. But we are very excited about all pieces of the company: enterprise, data center, enterprise campus, Tier 2 cloud service provider, et cetera. So we're excited about the whole book of business. But going into 2025, our planned scenario for that guide is cloud and hyperscaler AI-oriented.
One of the things we think about is the market verticals. So you've got sort of the Cloud Titans. You've got sort of the rest of cloud and service providers. What are the different dynamics and different behaviors based on sort of that segmentation by vertical?
Yeah, I think I can start, and Ashwin, feel free to jump in. So I think from the cloud perspective, the one thing is we came into 2024. We started the year with a 10%-12% revenue guide. And part of what predicated that range was an anticipated or perhaps expected moderated data center growth cycle. We're pleased to see as we come through 2024 that not only have we seen growth from an AI perspective, but we've also seen the traditional data center spend continue, so it wasn't cannibalistic between the two. So as we go into 2025, one of the things related to the verticals you mentioned is to hope to continue to see that those spendings are together and accretive and not one versus the other. So that's on the cloud hyperscaler side.
I think from the CSP perspective, one of the things we're seeing in enterprise conversations is, as enterprise AI starts to be a discussion that's taking hold, the conversations they're going to do their training in the cloud service providers, depending on the vertical. And so that could be a great opportunity for the CSPs, which we will continue to support in a great way from a quality perspective. And then I think we could talk about some of the traditional service providers if you want, Ashwin.
Yeah. I mean, and Simon, just to add to what Chantelle has actually said, what we've seen is our core strength, not only from data center, is expanding our data center core strengths into routing customers as well, both Tier 1, Tier 2, Tier 3 service providers, and then on the campus side as well. So both Chantelle and Jayshree have mentioned our strength and our glide path into campus side as well. So yeah, there's a good diversity of sectors, use cases, depending on all our product portfolios.
And I guess that sort of leads nicely into kind of my next line of questioning is around where are you investing in terms of hiring and staffing? Because when I think about this vision where the product portfolio is getting broader, but you're moving into different or more into enterprise, which typically means a much higher expense ratio. So what are you doing in terms of hiring ahead or investing ahead of growing in newer verticals, expanding in the enterprise for campus and direct sales as well?
Yeah, I think that we've been very supportive as a company to ensure that we continue to grow enterprise, both data center, campus, routing, et cetera. So specifically, some of the things we've done, if we focus first on, call it rest of world non-Americas, where we absolutely have opportunity to grow, we've done a few things. We've entered into some preferred partner elite programs in the sense of looking at the channel more specifically for data center and campus and the enterprise. We have definitely hired talent to lead the regions that have experience in kind of this Arista 2.0 enterprise, so great leadership perspective. And we continue to ramp sales headcount growth below double digits. So we continue to do that, as well as invest in the systems required for the back-end processes. Anything else you want to say?
Yeah. I mean, Simon, I would say I would simplify on my side, simplify the answer, which is most times when we're reaching out to customers, we want to sell them what we do and then do what we sell. And so from my perspective, the investment's coming both from the presale side and from the post-sale side. And we always want to make sure as customers come on board with Arista from the different product portfolios or from basically the different use cases, they get a good support and a customer success both from our presales and a support perspective. So I've invested in our strategies to help them migrate away from the incumbent vendors and then actually be able to go support them from our tricky scenarios as well, where recently we actually got a phenomenal NPS score of 87, very high, recognized in the industry.
That's basically customers recognizing Arista's value from that perspective.
Sorry, and just I remember the first part of your question we haven't addressed specifically, in the sense of the great thing about being perhaps the only pure-play networking company is that all 4,200 people wake up every day to focus on networking, so yes, we have data center. We have campus. We have routing. But from the sales perspective, it's one go-to-market sales motion. We don't have a distraction where do we need to invest in sales, nor is there a large overlay contingency. We don't have overlays, so I think that also helps the margin conversation and the go-to-market perspective.
Or different BUs, right? A small BU, Arista.
Right. But I think you talked about being very selective, maybe surgical in partners. And I want to unpack this a little bit because I think the conventional wisdom is for other IT-exposed companies that I follow, they pour a lot of money into sales and marketing, into the go-to-market, and it's channel partners and VARs are big expenses and viewed as really the route to market. Why is that not the case for you? What are you doing differently? How do you think about that?
Yeah, I think that a few things, fewer expectations and a few are just our market, our model, excuse me. So I think we're very curated where we use the channel to ensure that we understand what the margins will be and to understand the inventories within something that we control our destiny of. So those are the two things we're watching. We absolutely use channel partners for fulfillment combined with a direct coverage model. But as we look to go rest of world, we have talked about investing in the channel. So I think it's a mix of the two, and not to get too far ahead in the sense of just doubling down and having less control over the way we're going to market and what that means from a financial outcome perspective. So it's not no channel, but it's very strategic where we use them.
And then I can add to that as well, Simon. If you think about the customers that we actually touch on a daily basis, they're typically the Global 2000 customers. So all those customers are, I would call them, sophisticated buyers. They want a very high-touch model. And so that's where Chris and I have actually invested in go-to-market strategy for those sophisticated buyers. As we start to go down the food chain, which is Global 2000s from Global 2000s to Global 5000s, to expand that strategy, that's where the channel really comes into play.
Is there a way to quantify it, how many channel partners you have? Because we've heard sort of some of the numbers from your peers. They're tens of thousands. Do you know how many you have?
We're not in that range. I don't think we disclose the number. We very much appreciate the ones we work with because we see value and we see partnership. We're not in that range.
So now I'm going to ask you to reveal the key secret, which is your gross margins have consistently been better than peers. Your operating margins tend to be much better. What are you doing to achieve that?
I think there are a few things that afford us those margins. The first one is we do have benefit from the leverage of the mix of customers we have, so those being hyperscale, cloud sort of customers, and enterprise. But I think some of the real tangible things that are just inherent, if you look at the business model, the hypothesis is because we have high quality and high innovation, if the product works, you don't need a high cost to serve or a high cost of quality. And those things can add a lot of cost to your P&L. So if you were to ask me percentage of time and money spent on addressing a product that doesn't work, we don't have a high ratio there, which absolutely falls down to the bottom line. So part of it's the quality we deliver. Part of it's the go-to-market model.
I think part of it is just in the sense of the mix that we have.
Now, of course, we can't have a fireside chat and not talk about AI. So you've put out the target of the $750 million of AI back-end revenue for next year. We hear some level of debate, but I think the impression I have is that many on the investor community feel like that number's too low, that it must be much higher than that. What's sort of informed your establishing that target, and how do you think about it today?
Yeah. So the $750 million back-end cluster number that we originally came out with is in the sense of what we can see and know and not get too ahead of ourselves in a market that wants to be very big and very frothy. So I think from that perspective, it's kind of trials, pilots, and productions as they work through those scenarios. What do we see happening in 2025, and so that informs our $750 number. Opportunity is things close faster. They go bigger, and perhaps that's upside. The risk is these are some of the most complicated endeavors our customers have been in, rolling up their sleeves and that they don't slip outside of 2025. So I think that that's kind of the $750 back-end number. For the $750 front-end pull-through number, we opened up the aperture.
I think Jayshree had the right thinking, saying, "OK, there are a lot of other conversations in the industry with much bigger numbers. Hey, Arista, why aren't you guys talking about that? OK, let's talk about the front-end inference that comes through, and Ashwin can talk about the scenarios where that comes through," and so we said, "OK, let's go after that," and then once we have some of these closed behind us, we can talk about 2026 and 2027, what we see. We just don't want to get ahead of ourselves in case other things move and change outside of our control, cooling, power, cabling, space.
Yep, exactly. Yeah, and well said, Chantelle. Right now, I would add to what Chantelle is basically saying, talking about AI and enterprise using the same modeling of trials, pilots, and productions. Typically, cloud is a little bit more quicker in order to go from trial to production. From the enterprise side, they're still in the trial phase right now. So it'll probably take another three to five years before they get from trials all the way into production. So there's a nice glide path in the future as well as they transition away from that.
The only thing I'd like to end on that point, because I think it's important, is so that's our thought in the sense of how we get to those numbers. I don't think it's at the expense of deals that we've lost, and that's why the number's not bigger. I don't think it's that we're only saving $750 because that's all we see. I think that's all that's gone out to RFP and that we've had a chance to win. I think that's the other way to look at it.
Yeah, and I guess the other sort of nuance is you're not at zero.
No, we're not at zero today.
And I guess you have not offered us a metric on 2024. I'm guessing you're not about to tell us today. But how would you encourage analysts to think about what the 2024 number is? And let me explain why I feel this is important. Is that people are doing this arithmetic and thinking, OK, if I ignore AI and I ignore the campus, then the rest of Arista is really not growing. But that sort of neglects the fact that you're doing campus and AI today.
Yeah. So two things. The first one is I would say if you figure 2023, AI wasn't really in the conversation and we're going to $750 in 2025, you can assume somewhere in that midpoint is a safe on the both front end and back end. So I think that's the general vicinity that I would encourage you. Now, from the perspective of my style, Jayshree's style, the company style isn't to put out a guide that assumes everything hits 100% and then hope we hit it. Our style is to put out a guide that we have optionality for. So inherent, if you take a set of assumptions, you're going to assume other things aren't growing. There's opportunity for many paths to get to this 15%-17%. We're very excited by enterprise, data center, and campus. We're very excited about our security offerings, et cetera.
I would say we're encouraged by the all, and the guide is representative of where we're starting from.
I feel like in 2023, the debate in the investment community was InfiniBand versus Ethernet. I feel like we've gotten past that. What's sort of your take from sort of the real world, not the investment community, as to where the market is between employing InfiniBand as a cluster protocol versus using Ethernet?
Yeah, maybe I can take this one up. The best way to think about it is the size of the AI cluster. If you're thinking about it, I'll break it down into three simple models: small, medium, and large. Typically speaking, a small AI model is anywhere between 0-10,000 GPUs. Medium is between 10,000-30,000 GPUs. Above that, from 30,000, 100,000, and a million GPUs. Typically speaking, what we've seen is customers have deployed InfiniBand in a very small kind of fabric because the technology really doesn't scale beyond that. Once you hit the barriers of InfiniBand, you want to make sure that the technology can scale. You want to transition away from InfiniBand to Ethernet.
And then the other thing customers want to think about is, OK, do they really want to be locked into a proprietary technology, which is what InfiniBand is? They want to have an open standards-based way of deploying technology in a data center. So they will automatically transition to Ethernet because it's not only applicable on the back end, but also in the front end as well.
So just to be transparent, my bias is I'm Ethernet biased because I don't cover NVIDIA. But Ethernet biased. And I certainly think that that transition's happening. But I think one of the questions I'm struggling to answer myself is NVIDIA's ability to bundle. So NVIDIA goes to market with this platform narrative and says, we're going to sell billions of our Spectrum-X. And to some extent, that doesn't mean Arista can't do $750 million or more of back-end revenue because the market's big enough. But how are you thinking about the industry landscape where there's a player that bundles and a player that's offering sort of best of breed? How do customers think time to market versus the other trade-offs?
Yeah. No, you want to take the chance.
I can start. I'm sure we'll tag team. So I think the way there are a couple of things, and thank you for that question because I think it's an important one. So when RFPs come to market where there is choice, best of breed, we do have, I would say we have quite a bit of optionality in our architecture. We have a fixed form factor. We have Virtual Output Queuing and a chassis design. We have Distributed Etherlink Switch. All of those things provide optionality in the architecture versus the competitor product, which I think I understand is not as flexible in the architecture. So I think it depends on what the customers would like to do and the choices that they have.
Yep, absolutely. And Simon, thank you for recognizing the best of breed. And putting Arista in that category over there. So I really appreciate that. But at the end of the day, customers are always looking at basically being able to lock in into a specific vendor or a technology versus having an open architecture. And so when they're looking at open architecture, they always want to look at, OK, are they moving to a best of breed architecture anywhere in the AI stack, either in the XPU or in the NICs or in the enclosure or in the networking side? They're always looking for the best of breed over there. They always want to make sure even on the networking side, if they do go with a networking vendor, they can actually connect any XPU from any vendor over to the networking side as well.
And then the last thing I would just say is I always call this what I call the QRA factor, which is quality, resilience, and availability. You always want to make sure, and this is what's true to Arista, quality, and this is why we call best of breed. It's all built on our software. And so we've been providing the value from a quality perspective in our software, resiliency. So we want to make sure that the network is always resilient from an AI fabric. And then lastly, highly available, the SLA, the five nines perspective. They always want to have QRA from that perspective.
And I guess one of the other aspects that this topic brings up is Ethernet is valuable because it's an industry standard and it scales well. But because it's a standard, there are lots of companies that provide Ethernet solutions.
So maybe you've partly tried to answer this. But I think about if we open it up to Ethernet and we have the Ultra Ethernet Consortium defining essentially the criteria. Now we bring in Cisco, we bring in maybe Juniper, maybe HPE with Slingshot, NVIDIA's platforms. So it goes to be a more competitive environment. How do you compete when it's commoditized? I'm doing air quotes for the webcast.
Yeah. So great. Listen, these competitors that you've mentioned, we've been competing with them for a very long time. This is not a forte, and so we welcome the competition, and come back to my previous answer, I would always say where the baseline is Ethernet, but then you always want to compare from a hardware perspective. Are you a single ASIC architecture or a multi-ASIC architecture? Do you have power, space, and cooling from a product side, and then when you're looking at the software side, what customers are typically comparing us on the software is the software, does it have quality? Do I have many bugs? Do I have security advisories? When they're comparing software from that perspective, they will always want to make sure that the network is always up, so let's say, for example, a switch can fail, an optic can fail, a cable can fail.
They always want to make sure that the network is completely resilient, and then from an availability perspective, once the network is available, they want to make sure that they get highly supported, and that's really where my support team comes into play, and we give the best of quality from a support perspective as well.
So I want to move past AI and maybe get back to what I alluded to earlier, the idea of trying to figure out growth x AI, growth x the campus, where we've made our own assumptions. Of course, there's a chance for error. But within that 15%-17% overall growth, it implies that sort of the traditional Arista business is maybe growing just below 10% or in that neighborhood, which is a significant deceleration from where the company's been historically. What would explain that? Is the math way off? Or if we're in that ballpark, why would the company's growth for non-AI, non-campus, core Arista data center business, why would it be decelerating that much?
I think we're still very excited and optimistic about enterprise data center. I don't think that's what we're trying to convey. Again, I go back to guidance philosophy versus the piece parts. For enterprise data center and campus, if you take the data center side, I think if you look at the Fortune 500, kind of Russell, Global 2000, we're 25% penetrated on logo names. And so we have many new logos to go after. We have very many verticals. We're in quite a few great verticals, but there's still more verticals for us to go get. And then there's also the whole rest of world conversation. So we feel there's much more share to go gain on enterprise data center.
Yeah, remind us international is like 24% or so.
It's about 20% all in, so we think there's definitely room to grow there, and we just have to have the intention to go do that, which we do, and then the campus side, if you look at the campus switching TAM, it's $20 billion, if you take the large enterprise, it's about half of that, which is $10 billion, we're saying $750 million in 2025, I think we're just getting started there, so those are two really great growth opportunities, they're higher volume, lower dollar, so it just takes time, but I think Arista has demonstrated once we have the intention with the portfolio, that's something that will go with all forces intended, and we're seeing some tailwinds from some of the competitive actions in the market that have some of the logos coming to us, which is great.
We have some campus-first wins where we've won in the campus based on recognition and product quality where we don't even have the data center. And I think that's a great testament to the team, Ashwin and Chris, and the global team getting that out there. So there's absolutely opportunity to go get. But I would differentiate those opportunities with the guiding philosophy. And then we'll see what we can do through 2025.
So from our vantage point as analysts, the campus market seems like a soap opera. It's sort of this cast of characters, heroes, villains, some on the ropes, divorces and marriages. And so I feel like the dynamics there have made it more complicated than it has been traditionally. It's generally been thought of as traffic growth, evolving standards. But it's sort of we've got suggestions that some people are dumping Wi-Fi 6 on the market ahead of Wi-Fi 7 cycle. We've got the consolidation. What's sort of your view on sort of getting yourself involved in a market that's got these kind of social dynamics?
Yeah, I love your analogy over here, Simon. So using your analogy of marriage and divorces, the reason why there are so many vendors out there primarily because if you look at the value that they actually provide our customers, they're always looking at providing different hardware and different software for the different use cases. So whether it's data center, if it's campus, if it's routing, for all the traditional networking vendors, there are always different hardware and software for each one of these use cases. So using analogy, yes, there was a divorce in all these solutions. Compared to the Arista side, there was great marriage. And so the marriages, it's different hardware platforms. The marriage comes in from the software side. And this is really what customers love.
What they want to see is ease of use for deployment, ease of use for provisioning, ease of use for troubleshooting, security, automation. And so all the value that we've provided our data center customers for a long, long time, that moat is what we're providing our campus customers as well, which gives us the opportunity to go into what Chantelle calls really well, which is this $20 billion TAM. We focus on the top $10 billion. The sophisticated customers recognize Arista's moat from the data center. And they want to have the same value right on the campus side as well.
I guess in the past, the sort of campus market had been divided a little bit where one part of the market was sort of large corporate campuses, and the other was sort of the branch office, the retail storefront. Is there a difference in terms of where you fit better or both those are addressable for you?
Yeah, so today we're going after the very large campus sites. So for that, because that is a very easy transition from the large data center customers. But we have the portfolio today both on the wired and the wireless side for customers to deploy, not only in the large campus sites, 3,000, 5,000 employees, but all the way down to a small branch where we have got maybe 30 or 50 employees as well. So we've got the solution that covers everything. And then once again, they can go manage that with the same solutions they have today.
Now, I feel like there are a lot of interesting, exciting things going on with AI and campus, which has now overshadowed the wide area network opportunities. I don't feel like those have gone away. Could you maybe update us or revisit us where Arista stands on the WAN opportunities?
Specifically for AI or just general WAN opportunities?
In general.
Yeah, no. I mean, listen, we entered into the WAN opportunity almost, I would say, seven or eight years ago. So very similar as we entered, as you know, very, very early on, we entered the space with cloud providers. And then we entered enterprise data center. We entered the enterprise routing space almost seven or eight years ago. And so we've been present in that space for quite a while. Our goal was to first basically attack the Tier 1, Tier 2 cloud providers. Then it was to transition over to Tier 3 cloud providers, which we have done. And then today we basically provide, excuse me, routing solutions to a lot of enterprise customers.
So if they want to connect, to your point, campus and branch back to the data center or campus and branch back to the cloud or connect cloud back to the data center, they can go do that seamlessly. Once again, back to our moat, same hardware, same software, ease of use, ease of provisioning over and over again.
That market seems like it might be a little bit more difficult given that you've got sort of three entrenched players in the Western world. I'm just putting Huawei to the side, but HPE, Cisco, Juniper. Is there a way to differentiate or is it more of the sort of integrated software and working with your data center solutions? What makes it different?
Yeah, so it's actually two-factor. So basically, typically what happens over there is, yes, it is software. But most large service providers, they have typically gone and they've used legacy protocols and legacy hardware. What they've recognized is they need to move to what I call a modern operating model. So they want to move to more of a cloud-like routing model. And so as they do that, they've basically decided to say, OK, fine, which is the best cloud provider that actually does that for a routing solution? And so they've transitioned to Arista. So many service providers have actually used us in multiple use cases for their routing solutions.
So I want to get back to one of the key topics. We sort of touched on use of cash as it refers to hiring and staffing. Even with the expenses growing, our balance sheet keeps getting bigger and bigger. So you've had some stock buybacks. But what's your philosophy on how you want to employ cash?
Yeah, thank you for the question. Nothing's really changed, I think, for me through my first year from a capital allocation perspective. We have some great investments that have some great yields. So we'll continue to do that. I think that's a very important part of our portfolio. We do look at the share buybacks when it makes sense from a timing and market capability perspective. We will look at some tuck-in acquisitions as we need to. But the one thing I think that Jayshree and I are very aligned on, we won't do M&A for the sake of using cash. We're very clear on what our business model affords us and what would break that or what would make that accretive. And so I think until we find something that makes sense, we won't use cash just for that.
So buybacks, organic investment, and I think the investments are the top priorities at the moment.
What's your philosophy on introducing a dividend? How do you think about that?
Yeah, I think that my philosophy, having spent some time on it because I was expecting this question as we kind of come into the end of this year, is we're a high-growth rate company. And I don't think that's the right philosophy for a high-growth rate company. I think there's a time and place for that. But I don't think that's Arista's journey at the moment.
And what would it take for Arista to become more acquisitive? So you definitely have done your handful of deals, but they're generally small technology tuck-ins. What would cause you to rethink that?
I think if you think back to the things that Ashwin and I have discussed in the sense of what differentiates us, it's the OS, the cost of quality, the ability to serve, the quality, resiliency, and availability. I think those things cannot be broken, so if there's a great acquisition out there that's accretive in the margin space, in the realm of sense getting it to decent margin, and it doesn't break what makes us a pure-play software company that people choose, then I'm open to it.
So believe it or not, we've pretty much run out of time. I always like to close with this sort of same general question, which is, what do you think is the least well-appreciated aspect of Arista's story or for the stock?
Do you want to go first?
Yeah, I would say, and I'll keep it technical on my side, it's basically our software. I think most people tend to know Arista as a best-of-breed vendor. But many, many times, I always try to go emphasize on why Arista. And I always say, it just works. And the reason it just works is because of our software story. It's a nice, simple story that customers can always understand. And it relates back to them from an operational perspective as well.
Yeah, and the only thing I'd love to close with, and thank you for the opportunity, is all 4,200 employees wake up every day to focus on networking. And that's what we do focus on. We have tremendous opportunity just in what we did before 2024 started, enterprise data center, campus, and cloud data center and campus. The fact that AI has come in is a tremendous opportunity. If you feel AI has the growth potential it does, and you've heard us describe our portfolio, then hopefully you're convinced that we absolutely have a chance to be a great player in that market.
Great. Chantelle, Ashwin, thank you very much for joining us, folks. Thanks for joining us in this session. Signing out. Thank you.