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

Good afternoon, everyone. So we'll move to the next session. Thank you, if you can take your seats. I'm Samik Chatterjee. I cover the hardware companies at JPMorgan. For the next session, we have Arista. And from Arista, we have Ita Brennan, CFO; as well as Liz Stine from Investor Relations. Thank you both for coming to the conference.

And I'll really just start off with more sort of a macro question and a few questions we've been asking most of our companies to comment on. Just the remainder of the year, obviously, the macro backdrop remains a bit more challenging. But as you think about macro risk, what are the biggest risks that you perceive to your business?

Yes. I mean, I think certainly, the macro -- overall environment is offering lots of uncertainties. You think about debt ceiling, you think about all the banks, think about interest rates. So there is a lot happening at a macro level. I think when you take that back to the business, we still haven't seen a tremendous kind of direct impact on the business. We talked about this on the earnings call a little bit where on the enterprise part of our business, where we're really targeting larger big bet type companies. We've seen that continued engagement in spite of kind of some of these backdrops. And obviously, we'd be foolish to ignore those. We're going to continue to monitor and look at that over time, but we haven't really seen that impact on the business yet. I think on the hyperscale side of the business, on the cloud part of the business, that's more of a cyclical question, right? And we've talked about this a little bit where we've come off some very accelerated growth with cloud, where cloud grew triple digits last year. So we do expect that there will be some moderation in that spending. Is that macro? Is that not macro? I mean, I think it's just more of a cyclical kind of product cycle-related trends, and we'll see where that goes as we move through the year and get some better visibility.

Okay. Just to follow up on that. I mean one of the things, as you mentioned, like you haven't seen maybe the macro impact as much, but we've seen with a lot of other hardware categories. And I think even during sessions I've hosted today, it's come up that there was a pull forward of capacity by the hyperscalers in terms of their own investments towards capacity. How much of the slowdown you're expecting is largely a function of maybe a pull forward of demand from their side to add capacity and when you sort of look forward, it's more sort of just waiting for them to digest that capacity and move forward? Or is there a change structurally in how they think about their growth rates on public cloud revenues and that's changing their outlook for how they spend?

Yes. I think we had come through a very interesting period from a supply chain perspective where you definitely saw a lot more visibility from these customers in terms of what they needed to deploy, and we were able to kind of lay out deployment plans in advance and over a longer period of time than we would normally see. I think we've tried to be -- to have a good understanding of what was being deployed and when, and we're still working through that even now. I think the question is kind of once you get through those deployments, what does the trajectory of that investment look like as you head into 2024 and beyond, right? And I think that's something that there's been -- these companies are looking at their architectures, they're looking for efficiencies, et cetera. They've talked about wanting to do the things that they need to do for their business. I think their CapEx has been reasonably resilient even through this. So I think we're just going to have to wait a while to see exactly how that plays out. And we probably won't get visibility to kind of into that time frame until later in the year.

Obviously, one of the questions that I'm sure you answered all morning is the impact of AI on your business. Largely, I think, obviously, sort of the infrastructure spend is what everyone's focused on, but maybe flesh it out a bit more in terms of where you see the opportunities? Is there any sort of offset to think of in terms of where the business gets disrupted at the same time? Just help us think through that?

Yes. I mean I think for us, it's early. Jayshree talked on the earnings call about the fact that we're at the beginning of understanding what AI really means, and what the technology will look like. We have deployed some AI use cases, but it's relatively small. I think we look at it as -- it's a good underpinning of kind of future momentum and demand, particularly from some of the larger hyperscale customers. It's a reason why these technology cycles that are coming -- upcoming are going to be important, it's the reason why everybody is so focused on making sure that those cycles happen and happen in a timely manner. So I think that's all positive. It doesn't necessarily mean that tomorrow everything changed, and there isn't a lot of work to do, to get to kind of those new products and to get to those cycles, and there isn't a lot of technology challenges to sort out. But that's okay, too, right? That's really what we do. And -- so AI is a driver for innovation and for new technology changes, speed of deployment, all of that is positive for us. Exactly what it looks like kind of in this time slotted period, it's really too early to to start to fully understand that yet.

Okay. Since we're on the topic of AI, any way of sort of that you've quantified how big the opportunity for Ethernet switching is? How do you think about the risk from InfiniBand remaining a core part of that market or dominant in that market and Ethernet not being able to displace InfiniBand as a technology?

Yes. I mean I think if you look at the data that we have today, high-performance compute use cases, it's roughly 50-50 split between Ethernet and InfiniBand today. Obviously, that's almost pre-AI. But it is -- it does give you some indication or some place to start. Our belief is that if AI turns out to be so critical and embedded in all of these business processes, if that's really where this goes, then you're going to want to deploy this technology pervasively across these large cloud footprints. And I think that's where we believe that Ethernet has some advantages, right, in terms of being able to scale. It's standards based. It's already deployed in a lot of these footprints already. It's standards based. There's lots of expertise around it, et cetera. So there's a lot of advantages when you think about getting to scale that we believe kind of comes with Ethernet. It's not completely solved today. We need the new speeds. The team is working on, Anshul talked about near lossless fabric, but it doesn't need to be near lossless or lossless fabric. So there's work to do. But again, if you really think its scale as your end game, then we believe, Ethernet has a roll to play there, and that's where Ethernet kind of has some real advantages. Again, we're going to have to work through this. But I think that's -- if it's a scale play, we think Ethernet can have play a role there. It doesn't mean that InfiniBand is not going to have a role. It's just kind of the scale will, in our minds, favor Ethernet.

You mentioned moving to sort of next-generation speeds or higher speeds and being critical in that. And obviously, there's sort of latency, et cetera, that comes into the equation here. Does it change how fast these upgrade cycles are? Because you've been in the midst of a 400-gig upgrade cycle, does that 400-gig sort of naturally then bleed into the next upgrade cycle, muting some of the cyclicality that you've seen with your customers?

Yes. I think it's too early to really know that, honestly. We always want these cycles to go faster. And it seems like every time we've done one, there's some really urgent reason for why they have to go faster. I'm not sure how much that actually makes them go faster, right? So I think there's definitely going to be energy around trying to have this happen more quickly. But I think experience has taught us that it's still hard to -- there's real work that has to get done. There's an ecosystem that has to get kind of formed around these deployments, particularly at scale. So I don't know how much faster it really goes in the end. I think it's great that there is energy, and there's reasons why we want to make this happen faster. But I think it's -- yes, it may still take some time.

Okay. You mentioned on the last earnings call that you're already qualifying or sort of sampling some products with your customers. Maybe help us think through what that time line looks like? Where you are today? How much sort of -- how much of a time line does it look like when you're at that spot to sort of get something commercially into the market? Just trying to think about lead time to revenue here.

I mean we do have some revenue today coming from AI use cases. It's hard for us to track it exactly because it's obviously the 7800 platform that's being deployed, and it gets deployed into DCI and other use cases as well, right? But there is some traction around AI with that platform, but again, it's small, right? So I think it's more -- as customers kind of [ vet ] the solution and then start to deploy, we'll start to see some more revenue. But again, I go back to Jayshree's comments on the call. I think it will take some time for it to become a meaningful revenue driver.

Okay. When you say you track the use cases that this might be going into and it's difficult. I think there has been some confusion on the investor side, just given some of the other companies, including semiconductor companies mentioning more sort of exact revenue numbers that they're doing in terms of AI chips. I mean, is it really that easy to discern at this point? What's going into AI use case for you, what's not? And how are you classifying it? Are you looking at like -- do you have visibility into what the customer is using it for, or is it more of like triangulate based on what chips are in those systems?

Yes. I mean it's not something that I can -- that we can count kind of from a financial system with a very nice kind of black and white delineation, right, which is obviously where we'd like it to be. Because it's a multiuse product set, right? It's not just an AI platform, and the 7800 is our workhorse product that we'll deploy into large spine use cases across the business. So it's hard to get a very specific number. It's much more qualitative when you try to put some numbers around that.

Okay. I think one of the things that's come up is, obviously, Arista has gained a lot of share over the years and use the 100-gig and then the 400-gig as a key driver of share gains. As we think about AI, does it create an opportunity for further market share shifts, either in favor, or does it open up more ecosystem to more suppliers, new suppliers to come in? How do you ensure you stay on top of sort of continuing that progress on market share?

I mean it all still comes back to execution, right? I mean I think we've gone through multiple cycles coming into the 400-gig cycle. I think there was a lot of concern at the investor level about could we maintain our position. I think we're very happy with kind of how we came through that cycle. But it's all about execution. So now you need to go do that again at 800-gig. But I think the more in times of change and customers having new demands, new needs, new use cases, that has favored Arista in the past, right? And I think that the team is kind of excited about getting involved in some of those and helping to solve some of those use cases. But again, you have to execute, you got to deliver good quality products early in the cycle and have them scale. I mean that's the key to this. And so we just need to stick to the playbook, if you like, and continue to execute on that.

Okay. Just going back to your last Analyst Day, you talked about a 5-year revenue CAGR of 20%. As we look to future periods, and I think this is again sort of asking you to think about what the AI sort of in relation to magnitude could be, but is it sort of natural to expect that when you think about longer-term revenue CAGR, AI is sort of additive to it overall?

Yes. I mean, again, I think it's hard to completely separate those at this point, right? I know some of the industry analysts have tried to put some numbers on kind of what AI high-speed switching, Ethernet switching number would look like. I think it's extremely difficult to do that, right? Because again, it comes back to just how fast can you kind of resolve some of the technology issues, how fast do you have the new products? And then how does it ramp? So I think for us, again, we come back to it more as it's underpinning healthy growth, hopefully from the hyperscale piece of the business, in particular, over a period of time and a good period of time. And I think that's very positive to come back to kind of again what the slope of the curve is like. I don't think we have a good answer to that yet.

Okay. Staying on the topic, I think you said there is some modest amount of revenues that you're seeing on the AI use cases. I think one of the questions that keeps coming up is how much of the -- like your key customers like Meta have already been doing some level of AI investments? How much have you been able to leverage of that already, particularly any insights on sort of which technologies each of your customers are trying to favor relative to Ethernet versus InfiniBand. And Meta did have a recent announcement in terms of just sort of setting their priorities in relation to how they want to invest in AI versus the non-AI sort of infrastructure, or how do you think about implications on your switching business from that?

Again, I think there's going to be multiple paths here in terms of how these customers are going to go address this, right? And it's -- we were talking earlier today about when there's lots of changes and the technology shifts, it's messy, right? So there will be -- we will see customers have multiple different strategies. We're going to see them have phase strategies. We're going to see them try to drive efficiencies and drive better cost outcomes as they move through this, right? And that's not unusual when you've got kind of new -- lots of new products kind of come into market around the same time. So I think we're going to have to work through that with each of these customers over the next while. But again, we have that relationship. So we do have kind of that engagement, but it's going to take some time for us to work through it.

Ita, you keep referring to that the ecosystem needs to build out further like on the Ethernet side, particularly to compete with InfiniBand. Traditionally or historically, you've sort of taken the approach of more partnering on that side to really help out. But as you think about where you stand today, is then the sort of a more active evaluation of whether you can sort of wait to partner with companies, or do you want to sort of take a more active approach in vertical integrating and helping the ecosystem build out quicker?

Yes. I mean I think the ecosystem, both of these technologies have to evolve if they're going to satisfy the need. It's not just that Ethernet has a challenge. You also have InfiniBand being limited in terms of how it scales, et cetera. So there's going to be -- need to be development on both sides of that equation, right? So I think it's -- on the -- on our side, it's more as we move through these speeds, obviously, there needs to be optics and other things that come with that. We still don't believe you have to control an individual entity to make that effective. You really want to be able to influence the ecosystem, right? And I think we do a good job, a very good job of kind of engaging with the different partners and helping to kind of drive a road map, drive an outcome and not having to choose or become an individual player there. You get a lot more leverage and ability to influence the ecosystem when you're somewhat neutral. And you can help drive that road map. So I think that's still the strategy. It has been effective for us, and I think we'll continue to do that over time.

Let me open it up and see if there any questions the audience has.

Just on the AI revenue opportunity, is there any sort of conceptual framework you'd be comfortable providing in terms of how to think about that specifically with respect to maybe as an attach rate to GPU sales. I mean the concept being for every $100 you spend on GPUs, maybe you need making up a number of $10 of networking equipment to prevent any bottlenecks developing in the data transfer network. I mean anything similar to that, that you could frame for us to help us quantify or think about that?

Yes. I mean I don't know that we're quite there yet, right? We haven't offered that we have enough kind of experience yet to be able to kind of build out that type of framework, right? We've seen some very early data about the network being a bottleneck and causing kind of underutilization of the GPUs in certain cases. But again, that was a very early kind of point-in-time scenario. So I think we're not quite there yet where we could kind of create a framework that will be predictive. I think we need to see some more experience before we're going to be able to do that.

Any question -- other questions?

You mentioned optics being one of the things that you guys are thinking about or working with. Can you just describe that ecosystem? When do you think that it could be really adopted in scale. I know that there's a lot happening in innovation. But if you could just elaborate a bit more on your optics comment.

Yes, I think as we work towards a new product set, new speed transition. One of the things that's important is to kind of work with everybody in the ecosystem to make sure that there's alignment on all of the other pieces that we don't necessarily control, but that you want to have kind of happen and be viable so that you're able to kind of move at the speed that the switch wants to move at basically, right? So we do spend quite a bit of time. The team spends quite a bit of time working with the different players in that optical space to make sure that there's kind of clarity around what are the optics needs, and how can they best be solutions as we go through these cycles, right? And Again, I'm not probably not the right person to try and get into the very specifics of those technologies, but that's kind of -- you'll hear Andy, for example, Andy Bechtolsheim, our founder, talk a lot about optics. The reason why is, obviously, because you're trying to make sure that there is a clean path for kind of the switching and those products coming to market.

When you think about your business today versus 3 years ago, would you say there's been a shift in the importance of owning the customer relationship versus having the best product?

I think we're fortunate that the best product is still super important in networking, right? And even -- and when you think about kind of these -- even these large customers, they tend to want to buy best of brief. They tend to want to focus on, and they're very capable of putting all the pieces together themselves, right? So they really want like technology drivers and technology visionaries, if you like, people who could help kind of plan ahead and solve problems for them. So I think that's that engineering to engineering connection is still fundamentally important and then executing against us, right? You can own -- you won't own a customer relationship for very long if you're not kind of bringing the products and the technology to market and delivering kind of on what you commit to.