Let's get started. All right, good morning, everyone. How's everyone doing? Thanks for coming so early. First session of the conference, it's always exciting. My name is Sebastien Naji from William Blair. I'm pleased to introduce Chantelle Breithaupt, CFO, and John McCool, Chief Platform Officer of Arista Networks. Before we begin, I am required to inform you that a complete list of research disclosures or potential conflicts of interest are available on our website at williamblair.com. And with that out of the way, I think you have a presentation for us?
Yeah, sure. Hi, good morning. Again, thank you for attending this very early session. We appreciate it, and thanks for giving us the earliest time slot, Sebastien. So you'll just see the Safe Harbor statement there. So these slides are to, if you're new to Arista, to give you an overview, it'll be five to six minutes, quick overview. And then for those that know Arista, fresh reminder if there's anything here. This is typical, IR deck material, so you can find a lot of these slides on our, on our website. So Arista, you know, positioned to be a leader in data-driven networking, and what allows us that, we have a, you know, I would say, a sizable opportunity through us, and I know you're gonna have questions, or Sebastien will have questions on that opportunity.
Proud to have a very scalable architecture, single operating system, which is well-received by our customers and allows us to expand through adjacencies. And then I think, too, you know, we like to feel that we have a highly differentiated operating model that our customers enjoy and find easy to deploy and to realize their benefits, so very proud to have that in the company. Quick, Arista at a glance, you know, just over $6 billion. I think you know our gross margin's around the 63%-64%, high operating margin that we like to deliver to our shareholders. You'll notice on the top right, the IPO of 2014. We're happy to celebrate our 10-year anniversary tomorrow at the NYSE event in New York, so hopefully we'll see some of you there. Very excited for that celebration. Not an analyst day, celebration.
Joined the S&P 500 in 2018. Proud to have over 9,000 customers across many different segments and verticals and industries, and a leader in 100 gig and 400 gig. Updated chart in the sense of founders and leaders now, you know, running what we call, excuse me, Arista 2.0. So some familiar faces there, Jayshree, Andy, Ken, Hugh. Myself joining four months into the company. With me today is John McCool, Chief Platform Officer, longtime Arista. And then the recent announcements of Ashwin Kohli and Chris Schmidt to Chief Customer Officer and Chief Sales Officer, creating those roles for the company. Their roles are effective July 1, but they're already acting in the role day-to-day now and delivering to our customers. And then General Counsel, Marc Taxay. So why do these 9,000-plus customers choose Arista?
You know, I think it's a fairly clear, concise, elegant model that we deliver to the customers. Hopefully, you agree, you know, delivering on innovation with the work of John, Andy, Hugh, Jayshree, et cetera, to ensure that we're innovative and keeping pace to being, you know, seeing around corners. Superior quality is one of our key fundamental kind of mantras. So in the sense of we ensure that there's quality before anything gets to the market and ensure that our customers experience that quality, and then world-class support, you know, getting great customer feedback in the sense of the way we support them. An outlook on our TAM. You know, I think that if you see since 2023, going from $37 billion to $60 billion, you can see the core adjacencies.
Data center cloud is in, like, kind of that green, the adjacencies, campus and routing, and then the cognitive network is the software and services portion, so $60 billion plus. Some ideas on the demographics of how we, you know, how we run the business and our results. You can see the upper right, I think is probably the most interesting to you at this point, is the market sector trends. These are our thoughts in the sense of where these percentages would be over a length of time. So 40%-45% for cloud AI, enterprise and financial, and then the service provider and Tier 2 cloud. Fairly weighted, you know, from a U.S. Americas perspective, mostly driven by the cloud titan business, and then you can see the product trends to your left.
And last but not least, we're proud, after all the hard work of the 10 years since IPO, where we've kind of positioned ourselves as high-speed Ethernet market leadership, 100 gig, 200 gig, and 400 gig. So I always like to kind of position this before we start our conversation.
Great.
With that, Sebastien?
Thank you. Yeah, that's a very helpful overview. I guess maybe just the first question, somewhat of an easy one, but you're a recent addition here to Arista. You joined in February of this year. Maybe if you could give us, after these first, you know, four to five months, your perspective on, you know, what's really impressed you about Arista's culture, about Arista's organization, and then where you think maybe there's room to improve or build upon the foundation.
Yeah, sure. Happy to. So I'm very excited to be with Arista. I thought I was excited when I joined, but since I've been here four months, I'm even more excited to be with the company. I'll tell you why. A lot of my experience has been with larger companies, you know, high matrix, matrix environments, lots of BUs, lots of different goals. What I really appreciate about Arista is everyone is clearly focused on the same goal, and that was the goal that you saw on the slide of being innovative, delivering quality, and having world-class customer support. There's no confusion why the 4,000-plus employees come in every day and what goal they're looking for. A very elegant business model that serves across cloud, enterprise, et cetera, so really excited about the business model, Sebastien.
I think it's very clear to see the goals and how to get there. The culture acts very much like a startup with $95 billion market cap. You know, I think the sense of everyone rolls up their sleeves, they're happy to do what needs to be done, not a lot of egos that get in the way, and so I think that's refreshing and great. I don't know if I'd call them improvements rather than just the next steps that are needed. If you think about where we are now, kind of the $6 billion-$7 billion range, the next role is to how do we scale to $10 billion, $15 billion? What talent, processes, and investments are needed to get take us to Arista 2.0? I don't know, improvements. I think it's just how do we do more of the same without clouding that elegant business model?
Mm-hmm. Mm-hmm. That makes a lot of sense. So, you know, you sort of laid out, you've got these three major segments that Arista serves, the enterprise segment, the cloud titans, and then the providers. I wanted to focus maybe the first question around the enterprise. That was one of the, I think, the unique takeaways from the last quarter, is that despite, you know, most networking vendors seeing softness in the networking space and in demand, you guys are continuing to see strength there. Could you maybe highlight why it is that Arista is winning within the enterprise? And maybe give us a little color in terms of, you know, what do these enterprise wins really look like? Is it broad-based? Is it, you know, major large customers? Is it competitive wins?
Yeah, no, great. I'll start, and then, John, if you have anything, of course, to chime in. So I think—so if you look back to our Q1 earnings that you're referring to, you know, seeing strength in the enterprise. So overall, we're not the bellwether for enterprise, but we're certainly seeing the ability to gain share. And I think, there are a few things that help us drive that. The first one is the sense of winning in the data center. You know, so how do we take what we've learned in other markets and apply it to the data center? So that's one, but two is working on the whole portfolio. So we can go into the data center and win across the campus. We can go into the campus and then get to the data center.
And I think that kind of horizontal play is what we're looking at and what we're working on from a technology perspective and a go-to-market perspective. And I think from the perspective of the kinds of customers, you know, we're really focused on the, you know, financial, the Fortune 1000, Fortune 500, Russell 2000. And so if you look at different segments and verticals such as banking, medical, healthcare, Fed, SLED, you know, so we're across very many different segments and verticals. So lots of opportunity there, Sebastien. And I think what, what's happening is we're demonstrating the quality, the value, the innovation consistently, and our customers are seeing the ease of the operating model. You know, their teams can be home on the weekend.
It takes days versus weeks, weeks versus months, and I think we're just seeing that expand and getting reference customers, and that's expanding through their—as they move, they take us to their new customers where they're working. So lots of different ways that we're working through that.
Mm-hmm. Mm-hmm. Anything to add, John?
No.
That's good. And I think one of the comments that was made last quarter is that you're also seeing some activity from, you know, large, maybe Juniper customers who are rethinking their Juniper deployments with this impending HPE Juniper acquisition. Could you maybe talk about what you see as Arista's opportunity there? Are there any low-hanging fruit that you think will be some of those initial wins, or what initial wins have you had?
Well, I think that, so, anecdotally from our customers, at times we're hearing, we're not sure of the roadmaps of some of the other choices they have as vendors, and so that roadmap, lack of clarity or lack of timing, I think, provides us an opportunity. So how we see it as we're invited to conversations, we're being asked to come in to take a look. We're getting callbacks from people we spoke to two or three years ago, and so there's lots of different variety we're coming there. So how does it look? It, it looks in the sense of we take the call, we do the proof of concept, you know, we come back and show them, you know, other similar industries where we've won or customers or peers of theirs.
So I think that it affords us the conversation, and the low-hanging fruit is just when a call comes to take it and go in and show our product and, you know, take the conversation from there. But there is lack of clarity in the customer base, I think, with some of these M&A activities going on.
Mm-hmm. Mm. That's helpful. And maybe turning to AI, which I'm sure everybody wants to talk about. I think maybe it would be helpful just to frame the AI opportunity for Arista, and the framework that I've sort of used is there's three layers typically in your AI cluster, right? You have the interconnects that connect all the GPUs together into a server, that's scale up. Then you have the back-end network that connects all these servers together, that's scale out. And then you have that front-end network that connects these AI clusters to, you know, the broader data center, the internet, applications, et cetera. I guess, across those three layers, can you maybe talk a little bit about what is Ethernet's opportunity, and then more specifically, what is Arista's opportunity to win each of those segments?
Yeah, I'm gonna clarify our 750 target, and then, John, hand it over to you if that's okay, because I wanna tie this. So we have stated, in case you're not aware of the Arista declaration, to have $750 million of AI revenue by 2025, and so that's, that's counted just by the back-end clusters in what you described, Sebastien, just to be clear, the 750, 'cause there's not a unified definition of what AI is across different companies reporting this number. And then, John, if you want to talk more, please.
Yeah, so of those three, today we participate in the front-end networks. That's connecting the clusters into other aspects of the data center across the internet. So that's here today, and I think that's uncontested for Ethernet. In terms of the scale-up architectures, people are using, you know, talking about CXL, NVLink, other technologies from memory coherent architecture within the rack or the server itself. What's really interesting is the new network that comes in when you go scale out, and you're connecting many GPUs today on the order of 30,000 GPUs. You know, InfiniBand has been positioned by NVIDIA as a option for connecting those GPUs.
I kind of view that as a vertical integration of their stack, and that has challenges once you start to get beyond 1,000 connections, and you start to have to invent technologies that have already been well suited for Ethernet—the resiliency, the failover mechanisms, segmentation. We see the largest providers looking at going beyond 30,000 GPUs, really driving the focus on Ethernet as they interconnect to the back end. And one of the reasons we're focused on the $750 million target is that's really a new opportunity, a new TAM for interconnect that we haven't seen inside the data centers before.
Mm-hmm. Mm-hmm. That's helpful. I guess, you know, while we're on the topic of NVIDIA, sort of the elephant in the room here, and to your point, they've, you know, they're very focused on NVLink and InfiniBand, but they've also talked about their Spectrum-X Ethernet class of solutions. And on the latest earnings, they talked about that's gonna be a multi-billion-dollar business within the next year. Can you maybe help us understand, you know, how does Arista differentiate from NVIDIA's own Ethernet solutions, and what are the advantages that you might have versus their ability to, you know, vertically integrate everything and sell it as a system-
Sure.
rather than having to build those individual parts?
Yeah, right. So a couple pieces to that number, the Spectrum number, isn't just the Ethernet switching piece, but also includes the InfiniBand, cable fiber interconnect, as well as the NICs that go into the server. So all of those are very important pieces of the solution. We obviously focus on the switching aspect of it. A couple of things I think that we're highly differentiated, our ability to do the front-end networks, interconnect routing, and the other technologies provide a solution that you can use for the front end or the back end of the network that's agile in both ways. Also being able to interconnect other types of devices, I think we've proven that we can interconnect pretty much anything in a Data Center, not just GPUs, not just NVIDIA GPUs.
As people are looking to source their own inference engines or have diversity in their GPUs, that agnostic capability and being able to, you know, connect multiple things becomes highly important.
Mm-hmm. Mm-hmm. That, that's really helpful. I guess as I think about the $750 million target that you've laid out, I think one of the investor concerns is, okay, maybe the switch chips are there, maybe the switches are there in terms of the speeds, but there's a whole ecosystem that needs to follow, whether that's NICs, whether that's optics, and that may be lagging a little bit and hurting, you know, the market of putting it together yourself versus buying that integrated solution. So maybe help us understand what are some of the gating factors right now that you see in terms of the rest of that Ethernet ecosystem being ready for prime time? And, you know, what is Arista's play, or what is Arista doing to maybe help improve the overall ecosystem's maturity?
Sure. I mean, it's, it's not unlike any other transition we've seen. If you look at that ecosystem around 400 gig Ethernet today, it's, it's there with optics and interconnect and NICs and, and the whole thing. The industry is moving to 800 gig, all the optics, as well as the NICs, as the switches, and I think we're in a good position to take advantage of that opportunity.
Sebastien, if I can also add to John's comments on, so that's one factor that could gate success or delay, right, in the sense of the ecosystem components and piece parts. The other two is, if you look at the actual customer base, and I think you guys will know this as well as I do, but just to state it anyhow, is, you know, you have different company decision methodologies, and so you have some customers that are in the mode of build it, and the ROI will come. You know, build it, and the model will come, the payback will come.
You have other customers that have, you know, board processes and financial processes that want to see a demonstrated ROI, and so those customers might have a bit of a delay or a bit of a wait and see how other companies are rolling out their, their ROI and their business models for the sense of the payback on AI. So I think it really depends on how quickly companies feel they can lean in, which will gate how quickly. We're ready to serve when they're ready. We're ready to delight those customers, but their internal processes on when they decide to go will be just as big of a gating factor, I think.
Mm-hmm. Mm-hmm. That makes a lot of sense. I think part of it is also, you know, the, the Ultra Ethernet Consortium, which we've talked about, before, and, and having some of those standards ready. It looks like in the second half of this year, which will flow into product. Do you think when those standards are ready, is that sort of a, a catalyst or pivotal moment for the industry, or are, are customers, you know, already working on these build-outs, and that's, you know, that'll maybe accelerate a little bit, but it won't have a huge impact in terms of demand? What's your, your view on, on the impact of the Ultra Ethernet Consortium standards?
I'll give my view, and then John can, can also provide one. So I think it's customers aren't stopping or waiting for, you know, the UEC to come to fruition in the sense of having a standard. I think, from our perspective, it won't be a binary event that, you know, once the standard's out, it's going to be a floodgate that opens. I think what it allows is some thoughtful design, some thoughtful choice in the sense of how to build out with the new product innovation to make the best decisions for the company and how to be smart about how to progress their deployments. But John, I don't know if you have anything.
Sure.
Yeah.
Just to add to that, I think if you look back where we were in our last analyst day talking about AI, we saw an 800 gig opportunity. We also saw that both InfiniBand and Ethernet were not purpose-built for AI. And one of the things that Ethernet has done well as a community is evolve to incorporate new technologies and standards that could be interoperable, and one of the areas was around the transport mechanism. Today's GPU interconnect, whether it be over InfiniBand or Ethernet, is built on RoCE and RDMA. There's an opportunity for new transport mechanisms that go into the Linux stack, to do a better job and do things around load balancing and really cater to this new environment of AI. So that's what the Ultra Ethernet Consortium is about and building in those new capabilities.
But as Chantelle said, we haven't seen people slow down in deployments of Ethernet today with the RoCEv2 capability that we've had for a long time.
Okay. And then maybe following this line of competition, I want to talk a little bit about Cisco. I think, you know, they will admit that they missed the boat on cloud, you know, five to seven years ago. That allowed Arista to really take a pretty substantial amount of market share in the high speed switching market. You know, this time around, they've tried to get ahead of it. You know, they say they're ready for AI. They have a very broad range of products. So maybe my question is: Can you remind us why it was that Arista was able to take so much share from Cisco, you know, during that cloud upgrade to 100 G?
What allows Cisco today to, or what allows Arista today to maintain its market share and maybe even grow it, as you go for this AI opportunity?
John, do you want to go first?
Sure. I think, you know, Cisco did a good job in the 1990s figuring out how to connect everything to the internet when that was all new. If I look at where, you know, our 10-year anniversary is coming up this week of our IPO, we focused as a company on how operators would scale the internet, focused on those cloud companies with a single operating system, the ability to upgrade, low security events, etc. And that scalability is what's appreciated today in the cloud. As we go from 3,000, 30,000 GPUs to 100,000 GPUs, that same capability in the operating system to scale, to be able to observe events, to fail over reliably, to be able to upgrade seamlessly without introducing new defects, is gonna be paramount.
You know, I do think there was probably a lack of focus on the 100 gig cycle, but when we went to 400 gig, we saw similar results, and we hope to do the same thing with 800 gig.
Okay, great. Maybe turning to the financial model a little bit. I guess one of the questions that I get a lot, and maybe just set the stage, there's been this dynamic within Arista where cloud titan gross margins tend to be lower than enterprise gross margins, and so any mix shift usually has an impact on gross margin. I guess as Arista, you know, has more and more of a software stack, they're able to to sell, particularly within security, as you have this new AI opportunity that comes in, what do you expect in terms of, you know, the gross margin dynamics here in this world of AI? Do you think there's room for, you know, upside to the gross margins from where they've been historically? And if not, you know, why not?
Yeah, I think that, my view on that, Sebastien, to your question, is margin always comes down to volume with the customer. So I think there's a couple of dynamics here. One is, you know, obviously, the Cloud Titans have leverage due to the volume, so they're naturally going to be a different, a different trajectory from a margin perspective. But even with the conversation on AI, you're going to have enterprise and AI titans, so they're still gonna remain a mix. So I say, regardless of whether it's AI, cloud, software, there's always going to be a margin shift based on volume. So I think from that perspective, it'll maintain in the sense of it'll bounce around that 62%-65%, 64% range, depending on the mix.
Always aim to do higher and better if we can, but it'd be-- it would be based on mix from that perspective. You know, and I think from the software perspective, a lot of the features are on EOS now today, and it's not necessarily a separate software SKU, but delivering more functionality within that pricing model. So there's a lot of the go-to-market there that I think will continue to afford us mix, but again, it comes down to volume of which customers are buying in which quarters or which fiscal years.
Mm-hmm. To take away that sort of volume discounting that happens, is there potential for a higher ASP related to, you know, an AI spine or an AI leaf versus a more traditional spine or leaf or not really?
Well, I think, I think we're very at the early stages, so I think that there's a lot of data points to go get before I could answer that question specifically. Yeah, I think we're working with the same customers, so we wouldn't want to necessarily take advantage of a moment with our customers. So I think we would have a balanced view going into that. And a lot of the products are not dissimilar, right? So it's not that you can have this 7800s for a data center, data center cloud, and this one's for AI, so therefore, we're going to charge you. So I think you have to be cognizant of the product can be the same product delivering different functionalities, so you have to be, I think, cognizant of that.
Just the one thing I would add, though, you know, if we look at a 400 gig switch, it is driving the higher port speeds in terms of AI and the AI interconnect. So it is driving the higher capacity switches in our portfolio.
Mm-hmm. Mm-hmm. And to that point, I think Jayshree has talked about this next upgrade cycle, 800 gig is going to be much faster than the 400 gig one was. What, what's driving that? Is it really AI demand, or is there another component to that?
What we saw with the 400 gig cycle, it was really driven by the backbone upgrade. So if you take all the tiers of traffic and aggregate them, you needed higher capacity to upgrade the backbone. 800 gig cycle, if we'd waited for that cycle, probably would've been pushed out, but 800 gig to the GPU is very real, and that use case is really bringing along the investments in 800 gig.
Got it. Okay. And then maybe last question, I think we'll open it up to a few audience questions. Is really for you, Chantelle. I just wanted to get an update in terms of your philosophy around capital allocation. I mean, Arista has, you know, almost $5.5 billion of either short-term investments or cash on the balance sheet. What's your view on, you know, how that should be allocated, whether that's buybacks, whether that's potentially a dividend, whether that's M&A, that's maybe more meaningful than it has been historically? Just love to get your thoughts there.
Yeah, happy to. So I think that the first one I would start with ... We're still at a size of company relative to our competitors, where we need to punch above our weight class. So we're always going to make sure that we have sufficient funds on the balance sheet for things that have happened in our past, and you can't predict, but there's always going to be something that comes up. So I think the one perspective is making sure we have pretty robust on hand, you know, given the size we are, to make sure that we can handle the storm. The second one will be in the sense of organic investment, and I think this is where you're getting into, maybe not M&A, but organic investment.
You know, that first box that you saw, why 9,000 customers choose Arista, for—I'm not sure what happened there, for the innovation. So we'll make sure that we continue to fund innovation, and that can be sizable at times, given new product introduction, et cetera. So that would be the second one, or talent, or processes in the sense of the other two items I talked about in scaling the company. The third one would be around buybacks, and so you've seen some of that happen since I've joined, and we'll continue to look at the right opportunity for buybacks. M&A, I go back to, not necessarily looking to break what's working, and we have a, you know, we think we have a pretty good thing going, and we think there's a lot more we can do with that.
So M&A would have to be very thoughtful and aligned to the strategy, aligned to you know, the operating model, to the elegance of EOS, et cetera. So M&A would have to be super thoughtful, but it's always something that's on the list, but not the highest end of the capital allocation versus the buybacks or the organic investment. Dividends, you know, it's—dividends is an interesting conversation. I, I feel there's a time and place for dividends, and I don't think that's where Arista is at, at the moment. So that would probably not be at the top of the list for capital allocation at the moment, but- You know, there'll be a right time and place for that, I think.
Yeah. Okay. And then maybe one more question for me, just, you know, take advantage of the fact that you're here, John. Just around supply chain, I think one of the characteristics of, you know, AI hardware is that demand is ahead of supply, and that, you know, takes me back to 2021- 2022, when we had all these supply chain issues in networking. I mean, we've learned a lot since then. So maybe could you characterize, you know, what Arista is doing to help mitigate any sort of supply chain constraints it has currently or may have within the next year once a lot of these Ethernet systems are up and ready to ship?
Sure. Absolutely. So I think one of the things is lessons learned through the supply chain challenges that we had was every commodity matters, even the smallest commodity. So we have a lot more focus broadly across the supply chain in different elements. If you look back at the tail end of COVID, it was really the small chips and devices, power logic chips, that were used ubiquitously across multiple industries, especially consumer, that were a challenge to get. The tightness in the supply chain now is really around those large chips, which use an HBM technology that are used in GPUs and others.
So we're keeping a thoughtful eye on that in terms of our purchase commitments and working with our suppliers, as well as vigilance just around the global challenges in terms of risk and being able to mitigate those risks as we go forward.
Okay, great. So I think we have two minutes here, so maybe we'll take a few audience questions. You were first. And if you could repeat the question.
Sure.
Yeah.
One thing that surprises me, given your growth over the last 10-20 years, is that the International is still a smaller percentage than the Domestic. I wonder, is that an opportunity, or is that a problem?
Yeah. So if you didn't hear the question, I think if I can summarize, is: Hey, Arista, you've been around for a little bit now. Why are you still 80% Americas? So I think that there's a few things in there. So as we grow the denominator, hopefully, that becomes a smaller percentage. It is absolutely an opportunity to your question. So part of looking to scale the company is how do we do that rest of world. However, it's still a lot of the cloud titan, AI titan business is Americas centered, so that's what keeps- that kind of anchor weight to the Americas, but we're absolutely looking at rest of world growth, you know, as we work through the next few years. Does that answer your question?
Yeah. It's still only 20%. It's got to be an opportunity.
Well, I said it is an opportunity. That's what-
How do you capitalize?
Go-to-market growth through the Arista 2.0 processes, the organic investment that I mentioned in go-to-market. Channel partners.
Great. Maybe time for one more question. Jason, go ahead.
John, does the Ethernet and InfiniBand battle remind you of the last, you know, anything over the last 20 years? I know Ethernet sort of had various challengers, so-
Yes
The industries are here.
Jason's asking a triggering question here about my longevity in the industry and other battles with Ethernet and InfiniBand. So absolutely, it reminds me of a lot of challenges. Reminds me of voice. So voice was on proprietary technologies, PBXs, for a long time, and Ethernet rose to the challenge to build quality of service. And one thing I would tell people about Ethernet, it's not the technology, it's the co-opetition around it, competitors coming together to agree on certain standards to make things better, even though we compete. And that's really been the bellwether, and that's why I'm pretty confident InfiniBand is gonna have a role, but I think Ethernet's gonna have a bigger role. Thank you for that.
Great. Well, I think we're out of time, so we're gonna have our breakout upstairs in Adler. Thank you for coming.
Thank you.