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

Welcome, everybody. I'm Meta Marshall, Head of Networking here at Morgan Stanley. For important research disclosures, please see the research disclosure website at morganstanley.com/researchdisclosures. If you have any questions, please reach out to your Morgan Stanley sales representative. We're delighted to have Arista. Very topical with all of the AI talk that's been going on over the last 4 days. We have Ita Brennan, CFO, who most of you know; and Martin Hull, VP of Platform Strategy (sic) [ Product Management ]. I think I got that right.

Product management, but close enough.

All right. Perfect. So let's just jump into kind of the AI opportunity. Investors have been very interested in the opportunity for Arista and where we are in that investment. How do you view the AI opportunity as incremental to kind of the normal Cloud Titans business you guys have had?

Yes. I think probably the best way to think about this is if you look at the total Ethernet switching market and the industry growth rates that are out there, they're pretty healthy, right? They're in the mid- to high teens depending on who you look at. I think about AI is underpinning that, right? When you start to think about, well, why would we need to see these continued investments in networking and expansion of networking, AI is one of those drivers in my mind. And the more successful and pervasive that becomes, obviously, the more important it is. I think for Arista, it's exciting because it pushes the boundaries again from a technology perspective and from an innovation perspective. It allows us to work with customers to solve new problems. And that's always something that the team rises to and helps us to bring better technology. So I think it's kind of the change and the need for kind of acceleration is important. And then I think it's probably baked in kind of those Ethernet switching numbers that are out there already.

Got it. You noted that we're in very early days or Jayshree noted on the earnings call that we are in very early days of AI with very little revenue coming from it today. However, there are many generative model builds, and players like NVIDIA have already seen a good amount of revenue. So just what informs where you think we are versus where we're going in AI?

I think it's definitely early, but that's not to say that we're not -- we ship [ per ] -- 7800 product set, which is where we've talked about AI and some of the dynamics of that product could the AI use cases very well. It's hard for us to tell exactly what's going into an AI use case and what's going into a broader cloud networking use cases. But there's definitely some of that, that has been deployed into AI. But I think it is early, right? You've seen some of the industry guys try to estimate that, but I think it's hard to know exactly how much it is. But you're starting to see early use cases. And then as they prove it, I think that can drive further expansion of that. So I think relative to the overall kind of cycle, if you believe that AI is going to have a big impact across different industries, I think we're certainly early.

Got it. And does it matter for Arista, whether it's inference or training or where we are with AI workloads or it's traffic-traffic?

I think it all drives more networking traffic, and that's the key for us, right? So I think it's -- we're somewhat agnostic in terms of exactly what the use case is, or what the app is as long as it's driving that increased networking bandwidth.

Got it. And obviously, people know today or they should know that you guys have 2 major kind of Cloud Titan customers in Meta and in Microsoft. But where are we in the kind of taking -- making inroads into other clouds or taking share from an InfiniBand maybe with the AI opportunity more specifically?

Do you want to take a shot at that one, Martin?

Yes. So our business with Microsoft and Meta, Meta and Microsoft, alphabetical and...

The M&M.

M&M'S, yes, for the green ones. Where we are with them is, yes, we're part of their cloud business, but we're also part of other parts of their business. That's why they kind of [ stand up ] so big. When you look outside them at the other clouds, we've got solid engagements with them, and they are -- there was revenue coming from those. It just doesn't -- it's not quite as big. They're dwarfed by M and M. So we have revenue coming from at least 2 of the other hyperscalers. And we continue to engage with them, understanding what their short-term, medium- and long-term needs are. And we're comfortable with where we are in terms of being able to grow that business year-over-year. And I'd love to be able to sit here and talk about how much they are, but until they cross that threshold, we won't be doing that. But we are engaged with them on multiyear cycles, multiyear investments, and whether it bears fruit in 1 year, 3 year, 5 years, we'll wait to see.

Got it. Maybe as we think about kind of you're currently in the midst of a 200-gig, 400-gig upgrade cycle with your biggest customers, just you had a phenomenal year in 2022, growing 49%. How do investors -- how should we then -- who are not too intimate with all these things, kind of judge where we are in these upgrades?

Yes. Look, I think the first real revenue we saw from kind of this 400-gig product set was the back end of 2021, right? So we're relatively early in what is typically a 3-year plus kind of cycle even at the -- at its fastest, right? We saw good acceleration from these customers in 2022. Some of that was probably catch up from prior years where really the spend have been more -- much more muted. So we did see an attempt to catch up kind of in 2022. And obviously, we think there'll be good contributors to growth in 2023 as well, right? And then beyond that, I think we just have to continue to work with them and understand their needs, what are they planning to do? We are fortunate to have some good visibility into the business in the near term, and then we'll continue to roll that with them. But the indications are that they do have critical [ extremities ] that they want to continue to invest in. But again, we'll have to kind of roll that out as we go through time here.

Got it. I mean one of the biggest investor questions we get is just you noted on the earnings call, you haven't seen changes to ordering. And -- but your customers have talked about order cuts, other -- suppliers have seen order cuts. So just how much is supply chain a reason for kind of that dislocation we've seen? Or is it early days of upgrades? Like how do we explain some of that discrepancy?

Look, I think the supply chain kind of extended visibility, but it didn't necessarily make their deployments go faster, right? So I think there is a -- like we talked about, there is some catch-up from an investment perspective. There are new products so there was fundamental health in the demand. And then supply chain kind of extended the visibility, if you like, but it doesn't necessarily change kind of the needs that they had and that they face. We need to continue, again, to work with them and see what use cases do they need to deploy or they continue to deploy. But I think mostly, they've been pretty protective of kind of the CapEx where they consider to be essential. Now again, we don't have perfect knowledge of what's going to happen into the future, but I think we have to continue to work with them within those boundaries.

Got it. I mean this will speak more to -- or a question kind of asking more about the partnerships you have with your customers. But the share loss question comes up every cycle, and we're obviously early days in 400-gig and you've largely extinguished those questions. But when we get to 800-gig in a couple of years or insertion opportunities with other customers, what is -- what do investors miss about that relationship that Arista has with those customers?

Yes. So as a financial community, you're worried about this quarter or last quarter. As an engineering organization, we're worrying about a little bit about this year, but more about next year. So our relationships with these customers is multiyear and generational. So we have visibility. They have visibility into what the technology can do for them in 2 years' time. So we start to get aligned on, well, what form factor, what shape, how do you want to consume that. So we're making these internal investments in the products and technologies that allow them to take advantage of more than 2 years' time. If we don't have that conversation with them, then we'll probably get it wrong. And so they know they have to talk to us about what their needs are going to be. So when we start to think about -- we're not through with the 400-gig transition by any means. But as we start to think about, yes, 800-gig will happen, is in the future. How does that look like? Is it a little bit of 4 and a little bit of 8? Or is it more 8 and a little 4? So we're having those conversations so that we can try and intersect where they want to be 24 months from now [indiscernible] engineering plans are. And we have a constant cadence for that. It doesn't end. We're always talking about what's next and what's next, and we just keep moving our time line out. That's how we anticipate and how we have a good expectation of how we're going to have the right product. And then hopefully, the customers will select us and we get the right business outcome as well.

I mean in that kind of co-development process, Microsoft has had SONiC initiatives, Facebook has had different initiatives over time. What does that relationship like that you've continued to kind of be able to maintain share while I think investors sometimes think, oh well, they're trying to develop all of these different alternatives.

So all of the large-scale web providers want to be multi-sourced. Multi-sourced from a supply chain perspective, multi-sourced from an operating system perspective, multi-sourced from every tier of their network. They don't want to be dependent on a single supplier for multiple reasons. But in that multi sourcing, we're okay with them having a SONiC choice or an FBOSS choice or a white box choice and an Arista choice. And our EOS, the operating system that underpins all the products that we build, I think is a differentiator. If you go to the next generation of the leading edge, we hope we have a track record of being there early and ready for deployment and maybe some of the second source isn't quite ready.

Got it. Maybe one more question just on that. Given your intimacy, like, with those customers, just like, what is the problem that they're trying to tackle most today? Is it latency? Is it just accommodating speed? Like what is the biggest hurdle that they're trying to work with you on or -- in a non -- all of the above, I guess...

I mean, their business outcomes are [ derived ] on -- are based on the same as way -- that was on -- making their end customers happy. How do they do that? They leverage the next generation of x86-based architectures in compute that brings more traffic into their sites when they have to take advantage of the latest generations of networking to be able to accommodate that. If you double the capacity of the data center and keep the network the same as it is, you need twice as many network systems. That is a straight line cost. If they can go from 100-gig to 200-gig to 400-gig to 800-gig, then you flatten that curve. So that's really why they want to take advantage of the next generation. As soon as it's realistic, obviously, without going too fast or too slow. So that's why with Moore's Law, hopefully, continues for the next few years, right? Process node shrinks, silicon bandwidth increases, network products themselves [indiscernible]. So we're riding that performance curve with them because they've got an increase in their business needs as well.

Okay. Got it. Maybe when looking towards the Tier 2 Clouds, just how are you seeing -- there's clearly been important customers to you over time. Just how are you seeing their decision to build versus buy when we're entering a more macro-sensitive time versus kind of during COVID when maybe they needed the cloud more for burst investments.

Yes. I mean I think that part of the business, the decision criteria are very similar, right? I mean, especially for the larger guys, right? They're, again, driven by technology and performance. And maybe through COVID, they were chasing supply and chasing the same dynamics, right? In fact, we probably -- I think we've talked about this where for enterprise and for the other cloud pieces of the business, et cetera, we do want to improve supply and shipments as we head into this year, just to show kind of improvements versus where we've been. So I think from a revenue basis, we should expect to see an increase in business towards them, right? Their decision criteria is very much the same. They care about all the same things that the hyperscalers care about the efficiency, the automation, et cetera, and they make the same technology decisions over time. So it's a very good kind of extension of kind of what we're doing for the hyperscalers. And like I said, I think we should see some improvement in supply for that piece of the business as well.

Got it. I mean what vertical -- maybe moving on to the enterprise piece of the business. Where do you see kind of the greatest opportunity here? What verticals are kind of provide the richest opportunities?

I mean the nice thing now is that with that end-to-end portfolio and being able to kind of go sell Campus, go sell switching data center where that's relevant routing, it's becoming broader all the time, right? And Jayshree has -- in the earnings call, have been trying to show kind of just the breadth of the, what I call, mini verticals that we're addressing now in the enterprise. We've definitely seen that expand. It's still kind of a larger direct sale enterprise model, but the breadth at which we can go now and the number of opportunities for engagement that we have because of the Campus and some of the other use cases, even the NDR and some of the other applications. We're able to insert with these companies more frequently. And once we get that starting point, then we can cross-sell from there. So I think it's definitely broadened out kind of we're starting to see retail, you're starting to see commercial and industrial, those types of businesses, where traditionally, we would not have had a big presence. We might have had 1 or 2 customers, but now it's really broadening out. And the Campus just gives you that opportunity to have engagements more frequently than when you were just sitting there waiting for a data center opportunity.

I mean do you think AI can help there? I mean, I guess, I was at Mobile World Congress last week, and people were saying, yes, AI is largely going to be in the cloud, but a pharmaceutical company is not going to want to hand that over to the cloud. They're going to want to do that proprietarily. And so just are you seeing that discussion even start happening with enterprise?

Yes. I mean I think pharma is a very good example of that, right, where there'll be -- there's so many reasons why they'll want to keep that kind of on-premise and within their control. I think there'll definitely be other large enterprise-type models that will mirror that. So you'll have maybe not dissimilar to how things work today where you have certain large enterprises who become comfortable that they need to do this stuff themselves, and then you'll obviously have a broader, more generic AI applications in the cloud.

Okay. Got it. So the Campus opportunity, you've outlined the $750 million opportunity by 2025, progress today hampered by supply chain. You just spoke about having an end-to-end portfolio and that helping with insertion of new customers. But as you think about that $750 million, is that primarily from existing customers? Or how do you think about expanding that business?

No, I think one of the good things about how the Campus business has evolved has been that we are seeing a good proportion of it being Campus first, right, and where we're winning new accounts with the Campus products, in addition to obviously cross-selling the -- into the data center and the existing set of customers, right? So it's been more balanced, I think, in that regard than certainly we thought kind of coming into this because we did think we'd lean heavily on the existing customer set for a long time, but it is actually bringing new customers to the table as well, still kind of large enterprise, direct sales type business, but definitely more opportunities for Campus first than we would have thought.

And what are those channel investments? You mentioned some of those are new, but they're largely coming direct. How do you think about channel as expanding that business?

Yes, yes. I think channel is something that we have to build, right? And it's going to take time to do that. We're adding some partners every year. We're having success with some partners every year. But we will have to -- it's -- we have a more technical approach to sales. We need to -- we'll need to find kind of partners that aligned with how we sell the products and how we think the product can best be kind of deployed, et cetera. So it will take time. It's not like we can just overnight replicate kind of what other players have done. But we have some time, right? So I think about it as we need to be kind of making incremental -- and this is how we do a lot of things, right, making incremental progress over time, but we do have kind of the direct sales model to drive the near term.

I got it. So I mean, we just discussed Cloud Titans, we discussed AI, Tier 2, enterprise. As we think about the 25% that you guys have laid out for 2023, what customer type is kind of critical to achieving that target?

Yes. I mean I think it will be reasonably balanced with kind of an opportunity, hopefully, to show some more progress for kind of the enterprise and the Cloud given kind of the fact that hyperscale was first and kind of benefited in 2022. But again, you want to have multiple different ways to get there to feel good. I know things will not play out perfectly to my singular model. So we will have multiple different ways that we think we can achieve that 25%.

Got it. Supply chain has obviously been impactful not only to your ability to deliver to customers, but also gross margins. Just -- I think a big question investors have had is just the pacing across networking is just the pacing of this kind of gross margin recovery and just how to think about some of those supply chain conditions within.

Yes. I mean I'd like to think that we're kind of -- we see light at the end of the tunnel that we're going to make consistent progress as we go through this year. We have -- you need some proof points because it's been this whole de-commit problem, means that you don't really know for sure until you've had some but hope -- I think we're starting to see some of the real improvements there. So if we can exit the broker market completely or mostly, which is kind of our goal for kind of the first half of the year. Then that certainly helps, right? That was a very discrete kind of burden on the P&L and we start to come out from under that. Once you do that and you get -- you'll get other efficiencies as well, right? Because we've been -- this whole kind of stop-start, de-commit type dynamic is not helpful to scaling manufacturing and to some other things, we've been tying up resources and some redesigns and other areas as well. So you do get some efficiencies kind of on top of that. But that will come kind of once you're through kind of any impact, any de-commits you'll start to get a chance to drive that some more. So I do think we talked about this on the earnings call. I think we can drive some incremental improvements on gross margin as we go through the year. And hopefully, the kind of the 60% outlook that we had for Q1 is kind of a floor to that.

Okay. Got it. In order to achieve the 50% or nearly 50% growth you guys had last year, you had to make a lot of inventory and purchase commitments to make that happen. How do you see working down some of those inventory and clearly getting out of the broker market helps you on gross margins, but just working down a lot of the inventory to kind of help gross margins as well?

Yes. I think the way to think about the purchase commitments is there's almost 2 pieces to it. There's kind of key components which we've always managed ourselves and that we've always carried raw material. You see raw material inventory and the balance sheet, et cetera. And then there's kind of a supply chain kind of where we stepped into the supply chain, something would normally have been flowing through contract manufacturers, et cetera. The first one, that's driving some of the growth in inventory. Lead times haven't necessarily changed that much there yet. So we probably will continue to build into those buffers for the next couple of quarters. But those are key components and things that we're very close to their strategic parts, right? Then on the other side, there's kind of the supply chain stuff, which we will work very hard now to manage those back to lead times. There's various lead times. Everything is kind of moving, but lead times are coming in across that slew of parts. So we are working with the contract manufacturers to make sure they [ conventionally come ]. So that should come down, and we want that to come down in terms of those purchase commitments as quickly as possible. We saw some reduction in the quarter just gone, and we'll continue to drive that and make sure that we're kind of leveraging and taking advantage of lead times as they improve.

Got it. I have a bunch of questions, but I want to leave time for questions from the audience. Any questions from the audience? All right. Perfect. As we move into the Campus market, there's a lot more products, whether it's SD-WAN or security or just how do you think about skill sets that you need to add into that market? Do you need the partner to have them? Do you need to acquire to kind of get them? Just -- it's a broader skill set than traditionally on the data center side.

Martin, do you want to take that one? We were talking about that the other day.

Yes. So you're right. There's a wider breadth of products you find in the data center. It is, in some ways, more complex. And our approach isn't to own every single piece of that puzzle. So yes, it will involve more partnerships, whether it's with companies that make CCTV cameras or IoT devices or badge readers or security. There's a whole plethora of other technologies in there. And so yes, it will involve a lot more partnerships and also involve working with the sets of key verticals you identified, what's most important to them. If you're trying going to health care and you're trying to go into the medical environment, you're talking about nurse call systems and those kind of environments. We're not going to start to get the medical devices, but we need to be able to certify with them, given the accreditation that they know they can be deployed over our infrastructure. That's something that we're doing and will take time. But as we get those partnerships in place, then the end customers can say, yes, I'm allowed to use Arista in this environment. It's something we've done previously in the media and video broadcast space, right? There was transitioning from legacy SDI to IP. We worked with a lot of the media companies that are making outside broadcast vehicles or the mix of desks to make sure that we are certified with them and they're certified with us. So it's something we've done in the past to say media, and we'll continue to do it as we go into this Campus space as well.

Maybe one of the markets that Jayshree has outlined is the InfiniBand market as kind of an opportunity for you guys to kind of take share from that market. I mean you talked about it a little bit earlier, but I guess I'm just -- is there a different go-to-market there? Is there a different kind of customer you have to attack to kind of say, hey, you can actually address this with Ethernet versus InfiniBand? Or just what is that process like?

So InfiniBand is an existing technology, been out there for, I don't know, best part of 20 years. And InfiniBand tends to get deployed in that data center. They tend to be in islands. So people try and grow these islands of InfiniBand, you naturally end up with an Ethernet border boundary. If you're trying to take that environment into your general purpose networking, there's only 1 technology there. It's Ethernet. So I think as InfiniBand attempts to get into general purpose, then it transitions to Ethernet. So are we taking share from it? Or is InfiniBand just not growing? I think we'll find out. Certainly, for these large-scale clusters, I think Ethernet becomes the natural solution to that. But InfiniBand probably won't disappear overnight. It's been around for 20 years. It'll probably be around for a few more years.

Okay. Got it. So take incremental growth from it but not necessarily even to...

And as some of that scales, I think it's all -- Ethernet is going to be all about scale, right? So the more it scales, the more the need for it to scale, more likely it is that it moves towards the Ethernet because that's just the standard in all of these big footprints.

Got it. And then maybe just last question for you, Ita. Just how do you best assess kind of capital allocation opportunities?

Yes. I mean, I think we've talked previously about investing in the business as clearly the number 1. And that's -- it's been kind of -- right now, it's kind of supply chain. It was making sure that we had the means to hold our position in supply chain, even though we're still smaller than a lot of the competitors. So I think that's been very important. So we'll probably always want to have kind of a what might appear to be an outsized kind of position to be able to do that, right? That's probably [ before ] us. I think secondly, we will continue to look at M&A. It's -- so far, it's been tended to be technology and people based. And I think that will probably continue, just given the platform nature of EOS and what we do with EOS and how we want to leverage that is likely to be smaller pieces of technology that we can integrate around the network and all related around the network. And then obviously, there's a return of cash, which we've been doing through buybacks, and we'll continue to do that at some level. It's nice. The cash is now earning, a decent return again, right? We're seeing some good interest income come off of that. So that's good. But we will continue to kind of have a return program as well.

Got it. That takes us to time. So Ita and Martin, thanks so much for being here today.

Thank you.

Okay. Thank you for having us.