Arista Networks, Inc. (ANET) Earnings Call Transcript & Summary

Good morning, everyone. My name is Tom Blakey, I'm the Infrastructure Technology and Software Analyst here at KeyBanc. We're very grateful to have the CFO, Chantelle Breithaupt. I think I pronounced that correct. She's still relatively new, so we can call her the new CFO. And we have -- we're lucky to have the VP of Systems Engineering and Platforms with us, Martin Hull. So we can ask him the technical questions, not that Chantelle can't answer them.

So let's just jump right in. The Street seemed to like what you reported last week. So we won't have to review everything here, but we'll start with cloud. Given the implications of your guidance, it seems like there could be maybe a pickup in cloud. Can you just maybe talk to us about what you're seeing with your leading M&M customers?

Yes. So thank you for having us. Pleasure to be here as always, especially in this beautiful location. I think that -- so if we talk about the cloud kind of segment in relation to the year, maybe to the guide, so as we entered 2024, we had expectations of moderate, what we call, classic cloud and then wait and see on the AI. Because AI was very much even just 3 months, 6 months ago, InfiniBand versus Ethernet kind of conversation and as we work through InfiniBand versus Ethernet. Now we're kind of settling, okay, Ethernet. So now we're talking about Ethernet. So we're seeing some AI momentum come in from pilots to trials to production. And so I think that seeing those things kind of gain momentum, both of them, classic cloud being moderate growth and seeing that, seeing the AI momentum. With that, we feel more confident in the year and hence, at least 14% revenue growth for the year versus the 10% to 12% we started with for the year for the guide. And so we're excited about that. We have to wait and see between those 2 parallel paths where they were going.

So you're saying that the AI portion of the cloud AI titans group is what is what...

It's both, it's both. It's classic cloud being moderate because there was risk that the classic kind of cloud spend would only divert to AI and not be both. We're seeing spending in both, refreshing and AI. We expect AI to be a little bit of a higher growth rate, but that's because it's off of small numbers. But the fact that they're both showing momentum is giving us confidence as we through the year.

So jumping to a deeper question on cloud that I was going to ask later. We're all tracking as investors. The CapEx rates of these companies keep going up, especially your key customers reported just last week as well that they're going to be spending a lot more money. And there's a correlation, whatever the R-squared is, I can't quote it right now, but there's a correlation between that CapEx and your cloud titan revenue. Talk to us about maybe wallet share and what's moving underneath the hood there. I mean I don't know if Martin you have an answer to that, but just how does Arista gain wallet share in terms of these bigger CapEx numbers? Because this can't grow forever, maybe.

I'm not the finance person, so I won't say it can't. So yes, you get into the big numbers on those CapEx, so eventually, it will moderate. If you think about what a lot of the CapEx is going into and we saw that from those large customers that reported last week, a lot of the CapEx investment is in real estate land, buildings, power stations, power capacity. And then eventually, they start to fill these buildings up with power, cooling, chillers, and then they start to put network infrastructure in and compute and storage and everything that goes along with it. So there's definitely a time lag between some of the early upfront investments in land why you can't correlate increases and decreases. Our wallet share, if the networking spend is a fixed percentage of that, then we like to be market share winners, market share gainers within the networking spend. That is the traditional front end, the data center interconnect, the peering points, all of that front-end network, which the size of that market has grown over the last few years, and our market share within that is staying fairly constant. And then this emerging back-end AI network, as the networks themselves get larger, they tend to be more balanced towards an Ethernet-centric world. And then when you're talking about an Ethernet-centric world, we're the #1 market share owners within that front end. So it's realistic to think that we will be significant market share gainers in the back end, if all things play out evenly. So how do we get more wallet share? As those AI networks get bigger, I think that it become more Ethernet and less InfiniBand. As Chantelle was saying, that was the debate earlier in the year. As they get bigger and bigger clusters, you're going to want the best networking vendor that's out there, which is the Arista Networks.

Yes. And I think the only thing I would add to that before you go to your next question is the continuous focus on being pure-play networking with the innovation that you expect from Arista. And so you've seen some of our new product introductions this year. And I think we have the position to gain share if we continue on that innovation. So that's a very big part of the R&D kind of investment side about the equation.

I'm sure we'll come back to some other questions on AI. But just I also noted on the call that there seemed to be an uptick in terms of bullishness on -- in my read anyway, tell me if I'm mischaracterizing that, enterprise. For the audience or for the folks that don't know, Arista is obviously the dominant market share leader in high-end data center just started ramping even just the last couple of years in terms of the end-to-end from campus to routing to the data center, broader product portfolio. Talk to some of the trends that you're seeing there to maybe have an increase in confidence in that business.

Sure. I can start, and Martin can jump in. So I think that -- so if you look at enterprise, we're talking about what you mentioned, data center and campus. And campus is basically a pretty nascent market for us versus where we are from a market share perspective. And so we have a thought or theory about land and expand in enterprise. So there are a few scenarios. One is that we're in on the data center and the customers are satisfied and excited by what they're seeing in the performance on the data center. So it's a natural conversation as our campus portfolio is now large enough to serve the needs of our customers, to talk about the campus or which is even more exciting, sometimes we're coming in first on the campus for some new logos and starting with campus conversations, which for us is a milestone to be recognized for our campus portfolio. And then once you're in either data center or campus, you can go horizontal. So you can talk about horizontal, all the things you mentioned, routing, security win. And that's only in the current verticals we serve. So some of the public sector, health care, financial, but there are many other verticals. So there's momentum there. And then the other segment, the other part of it is the fact that there is some anecdotal conversations with our customers on confusion of competitor road maps, competitor M&A activity. So we do see customers coming to us saying, hey, we'd like to talk to you now. You seem to be just pretty much focused on networking. Can we have a conversation because your road map seems a little clearer? I don't know if there's anything you'll add to that...

I think you covered it all off. I'm sure there are more questions.

Well, maybe more as an adjunct to that, what -- if we take -- Arista has been, in my opinion, doing a great job the last few years of adding a little bit more granularity to their results. If we back out some of the success that you're having, and I'd love to hear if there's anything, a factor or 2 that's kind of making this -- maybe it's just time, it was just as simple, but what's going on in the underlying data center where you've already clearly achieved a dominant status? Again, kind of the same type of questions about wallet share or any dynamics that could maybe lead that business. Because if you take away some of this new growth that you're having in campus, there is a bit of a moderating -- moderation of the growth in the core data center in the enterprise.

Yes. I mean, we're still effectively on the back end of all the COVID work that had happened 3 years ago, now 4 years ago. So there's still a lot of customers out there that are still getting to grips with remediating their existing data centers that were stuck in time for 2 years. They couldn't get to those facilities. If you look at the underlying data center market and you just characterize that as maybe the 100 gig sector. 100-gig Ethernet is not shrinking. There's a lot of customers out there today that have still got older 40-gig technologies, 10-gig technologies, which are maybe 5, 6 years old. They're coming to the end of a life cycle. They might be based on Arista, they might be based on the earlier vendors. And that then becomes an opportunity for a refresh. So there's a big refresh cycle in that enterprise data center. Some of that's driven by an AI solution, an AI on-prem or it's just driven by the fact that their existing compute is coming to the end of its useful life, 5 years, 6 years. So that refresh will constantly go on. And from a market share perspective, we've always talked about our market share within the high-end network, which is 100 gig, 200 gig, 400 gig and now increasing to 800 gig, where we have the #1 market share for a while. If you back out from that and say, just the data center network, 10 gig speeds and higher, we've now gained a bit more market share position there. And quite frankly, the cloud isn't buying 10 gig and 40 gig networking. So where is that growth coming from? It's coming from not cloud, which is the enterprise, the broadening out. And we're seeing that happen through reputation. You have an IT guy or a networking guy who's been at a particular organization. He will change jobs. He will take his best-of-breed vendors with him. So that happens over time as people move across organizations. Also just word of mouth, reputation we're gaining. So that's something where we can continue to take market share, is that broad enterprise market.

Yes. And the only other thing I would add to Martin's comments on timing, because you're spot on with your timing, is there's a nice segmentation of the enterprise market that we call early adopters, fast followers and more risk [ adverse ], if that's the right category. Yes. And the if I focus on the risk [ adverse ], so an example of that, working through Martin's example that over time, we hope to get there is -- or if you take teams that are highly tenured, about to retire, they're not necessarily going to be -- want to be leaning into risk to take out the incumbent. But we do have conversations, hey, when you get past that refresh in that environment, we'd love to talk to you. So there are examples of time that still give us market share to go get.

No, it's a huge opportunity. Shifting back to AI, there's some -- always some temporal news before we talk on these things. But there's some delay -- there were some announcements about some delays at some key GPU manufacturers on the planet. Does that impact any of your -- Jayshree obviously seems a little bit more bullish on having visibility into a $750 million target in terms of back-end AI-related Ethernet revenues for '25. Maybe a more timely question about potential delays, how would that impact Arista?

Yes.

Me?

You go first.

I'm not sure I have the numbers. So I think we were looking at that as a total 2025, and so a quarter or so shift, I don't think, fundamentally changes our view on that one. We've been conservative on those numbers, and we were getting some questions about why so conservative. And now, well, that wasn't the worst idea, was it? Decision-making in these customers is still the same process, it's just going to be a timing choice. They're still going to want to go forward with the AI compute.

Yes. I think the pragmatism that Martin is referring to is exactly for -- there's a lot that has to come together in the ecosystem when we're talking about AI specifically. The $750 million campus number, that's fairly in our collective control, but the AI one is a very broad ecosystem. So to Martin's point, I don't think there's risk on the current temporal news on the $750 million. I think it would take a bigger risk. But the reason we talked $750 million and not some of the bigger numbers that people have wanted us to say is because we recognize there are these bumps along the way.

We'll maybe just turn to the audience if there's any questions. No? Okay. Maybe just double clicking on that, that was a great jumping off point, Chantelle. Maybe, what are these key factors? Again, I think the market understood that there was an uptick in confidence from Jayshree with regard to the $750 million. You even talked about a number of new customers, right? There was the 4 to 5 and now that we're talking about 10s to quote Ms. Jayshree. So would you -- what are the factors that could -- up and down, what are the factors that could make you maybe go over or go under?

Yes. I think I'll start, and I'm sure Martin will have some thoughts on it. So if you think about the confidence in the $750 million AI target, which is only the back-end clusters, we're super clear because there are many definitions of AI revenue across the industry. For this year, we're encouraged because the trials to pilots are going well, the tens of thousands and some customers wanting to go to 100,000, those are great conversations, and it's working well. That bolsters the $750 million and could be -- if it happens sooner, could maybe make the $750 million a different number upward next year. But the things that need to come together for that are many. The suppliers, road map is on time. The facilities are ready, the cooling, the power. I think there's even people, cabling have to be available. And so there are many things, but hey, but if all those things fall into place and the momentum in the use cases prove out, there could be upside there. The downside is all those things take 3 quarters longer than they're supposed to. So it's a quarter-by-quarter wait and see. And on the -- and just the first part of your question. So when we talked about the 4 out 5 of being cloud and not just cloud customers, when Jayshree was referring to the 10s, we're talking about seeing some green shoots of enterprise AI discussions. And that's a beginning of the conversation. Martin, maybe you can give some of the use cases that we're seeing in enterprise.

Yes. And so we're hearing the same thing you are, and that is in the cloud, we're getting into this show-me phase, right? Is there an ROI? I think on the enterprise side, you prove the ROI upfront. So when they're going forward with the decisions to deploy an AI environment, they've got a couple of choices, AI on-prem in your own facilities. So you're going to have a conversation, you're going to stand up an AI cluster and you're going to get it working. The other one is AI as a service. And there are many companies out there today, software companies who are AI-enabling their solutions. So the enterprises will buy into that because they can absolutely see that it's good advantages for CRM, maybe logistics and supply chain and planning. Maybe the airlines were scheduling after a major outage or two. I think some people aren't here because Delta's still got problems, right? But there are definitely vertical markets where you can get a quick benefit from an AI application rather than saying, I'll build it and hope they'll turn up. So the enterprises are looking at this from an ROI from the other way around as like, I can prove the ROI, now can I have the funds to go and invest? We're seeing it in health care. We're seeing it in pharma. We're seeing it in logistics.

And these are enterprise customers looking to create those AI as a service -- for internal?

No, they -- well, it's as a service that they're building it internally. But I'm saying that if you've got companies that are out there, a Salesforce or SAP, they're going to offer -- like the Microsoft CoPilot, they're going to offer an AI solution on top of their existing portfolios. A financial services company might now start saying, well, we've got an AI-generated customer services or there are some large banks out there that have come out with their own AI services.

I think there was -- during the call, and this might be old news, but just maybe to clarify that there was 4 to 5 wins, maybe just double-clicking on the loss, with regard to maybe this particular customer taking the embedded solution from a maybe possible new competitor. Why would that -- why did the customer choose the other solutions?

You're talking about 100. So why? Some of it is timing and some of the supply chain questions that Chantelle was referring to, right? And if you go one particular direction, maybe a date plan is aligned and other ones, maybe that decision gets revisited in the future, no decisions forever. And a lot of these AI clusters that are getting built aren't one and done. There's a journey, right? The current generation of GPUs, the next generation of GPUs, what we're seeing from an Ethernet network perspective is predominantly 400 gig today, 800 gig is still new. As we get into the 800 gig rollout probably prime time 2025, then maybe the next set of questions get a difference that is..

So typical risk, you're still in contact with this customer, so to speak. And Jayshree likes to call it the boomerang effect. So there…

Yes. We never stop talking to our customers. We have -- these customers, we have deep technical relationships with and commercial. We're talking about what's happening this year, next year and the year after. We're moving off road maps in lockstep with them.

Just maybe look to the audience one more time, if they have any questions.

What's your view on [indiscernible]?

Well, you're going to need to step back, and you said it, right? Multivendor, multi-sourced Ethernet. Single vendor, single-sourced InfiniBand. If you're a large enterprise organization, nobody likes to be single-sourced. Whether it's your compute, your storage, your firewalls, your desktops, single sourcing is a bad plan. We saw that as we got into COVID. You need diversity of supply. So that is maybe the first trial, sure, InfiniBand, who cares? Maybe roll out a big one, who cares? But ultimately, you're going to have to come up with answers to these questions about what would happen if. So that's the kind of networking side of it. You'll say, well, is there only one company out there making a GPU? It feels like it some days. But there isn't only one company. But there's only one GPU vendor with an InfiniBand solution. All the other GPU vendors don't have InfiniBand. So as AMD and Intel and others come up and roll out their technology, there isn't going to be an InfiniBand choice. So if I want to roll out multi-vendor GPUs, InfiniBand isn't an option. And then you can get into UEC and preferences and technology curves. But if you just step far enough back and go, there will still be InfiniBand. We were talking earlier, right? InfiniBand was deployed as the first technology inside the low-latency cluster solutions that were deployed in the early 2000s. It was my first coming together with InfiniBand. And then Ethernet caught up and took over. I'm not saying it's going to fully displace InfiniBand in AI, but Ethernet is going to get the volumes.

So I'm not sure if the question included embedded Ethernet, but is it...

[indiscernible]

We've had RDMA workloads in our switches for over 10 years. That's not new news. High-performance compute clusters, HPC, has been doing the RDMA workloads. And Ethernet and InfiniBand have been going backwards and forwards on this for a decade. So really, it's a case, looking at the UEC and looking at simplifying the stack and maybe getting rid of some of the overheads that are unnecessary so you can streamline it. And then also getting new capabilities for load balancing, traffic management, spraying, packets, but that has to happen on an end-to-end basis, which is why it's multi-vendor. Silicon systems companies and the end customers themselves are all part of that UEC consortium. And the benefit of that there is you're not locked into a single vendor.

Thank you, Martin. If there's no more questions, I want to end on a model question with Chantelle. The implied guidance, we're in that kind of second half of the year where you guide to 3Q and I can do seventh-grade Algebra. But the 4Q implied gross margin seems to be a step down, like a multiyear low step down on the product side. And that would imply strength in cloud. That kind of goes back to my first question while we were sitting here. Maybe just talk me through about what the assumptions are for why product gross margins would do that in the context of potentially an uptick in cloud spending into the back half of this year.

Yes, happy to, and I'll speak more to gross margin for the company versus just product because I think it's an easier conversation. So the at-least 14% at this point in time in the year is because Jayshree and I are looking through different optionalities on how the year finishes. And that's all related to the conversations we've been having on timing and availability and CapEx spending. And so those all play into different scenarios as we see this year. So we'll come back in Q3 obviously and kind of finish out the year in the sense of what that means. But there is a mix element, there is a mix element in the current guide to say, hey, the moderate cloud and AI spending is why we raised to at least 14% so that it will have an impact on the second half margin. The other second half margin impact is that -- just trying to temper, if that's the right word. Because in the last couple of quarters, we've seen benefit that's not mix related but more related to the fact that we've had favorability of inventory-related costs, inventory-related reserves. And so that's kind of carried the year to be a bit higher than the guide was even at the beginning of the year. So those aren't necessarily repeating every quarter going forward. So I think the mix of looking at the inventory-related reserves, the variable cost productivity and the mix, that's why we stayed with the 63% to 64% in Q3, and then we'll see as we come, what the optionality is on the top line, what that means for the year.

And just to clarify that comment, you said moderating? Moderating cloud in the second...

Well, we said right from the beginning, we expected moderate cloud growth this year. Because coming off of a high cycle last year and coming into this year, Jayshree wanted to be sure that we saw momentum before we called the year to be different. And this at least 14% shows we are seeing moderate cloud. So the traditional cloud spend -- because part of the theory was the traditional cloud was going to be diverted to just AI. We're seeing there is some refresh, some cloud and some AI. So those 2 things coming together are why we're saying at least 14%. And then we'll see what the end point is as we come into Q3, what the year is going to look like.

And the opposite side of that is if it comes out the way you're thinking. So with enterprise stepping up, is the construct of mix to enterprise is still higher gross margin?

Well, it could be. Like we're working through depending how some of the timing of the customer engagements land. But that's the rationale for now, why the guide is where it is.

Yes, sir?

On the inventory-related reserve, you said they don't necessarily repeat [indiscernible]?

Well, I think as you work through doing -- if you think about coming out of COVID, I think a lot of companies went through, what's the right mix and the minimum order quantities and having to buy a lot of one. And I think we've kind of gone through -- Arista now speaking, we've gone through like 3 quarters at least of coming off of that. So maybe there's a couple of quarters left and then that kind of chunk is done and done with. So that's what I'm referring to.

Excellent. Well, thank you very much for your time, Chantelle, Martin. Thank you so much.

Thank you for your questions.