Good afternoon, everyone. Welcome to day two of Citi Global Technology Conference. My name is Atif Malik. I cover U.S. semiconductor, semiconductor equipment, and communication equipment stocks here at Citi. It's my pleasure to welcome Ita Brennan, CFO, Arista Networks, and we also have Liz Stine, Director of IR, in the audience. Welcome, Ita.
Thank you so much. Thanks for having us.
I'm going to start with my questions first, and then I'll take a break and ask you guys if you have any questions. And if you have a question, please raise your hand, and we'll get the mic to you. Ita, we were talking about the A word before the fireside chat, so I'm just going to, you know, get straight to that. What is the AI opportunity for Arista, and how meaningful of a revenue contributor is it today versus two to three years ago from now?
Yeah, I think, look, there's been some industry analysts who've tried to size this market in terms of just the Ethernet switching piece. You know, you've had kind of a $2 billion to $3 billion number being tossed around, but in reality, I don't think anybody has a proper sizing of the market yet, and that's, that's just gonna take some time. You know, certainly in terms of the products that we sell and the products that we will sell, it's not going to be completely obvious, you know, what use cases they're being deployed in. So again, we'll have to track that somewhat off system to get some kind of sizing on that market. But today, we're selling those products into various different use cases.
You know, I think what's interesting for us around AI is, you know, obviously we've been selling, you know, products into kind of the front end of AI networks probably for some time without even knowing that, right? We're-- you know, we were transporting data from some of their back-end processes out across their networks. I think the more exciting thing now, obviously, is talking about kind of what does, you know, what does the back end of these networks look like, and what kind of, you know, opportunity can that drive? That's -- right now, we're in the throes of kind of that's driving lots of, you know, technology development, urgency around, you know, some technologies that were in the pipe anyway, you know, around moving 800Gb speeds, some software features, et cetera.
So, I think more than, you know, having a clear, discrete view of exactly what it's going to mean, in the near term, it's more exciting about the fact that it's driving, you know, urgency around the technology, and it's allowing us to, you know, continue to develop new features and work with customers to develop new features and, you know, expand that relevance over time. So I think today it's... You know, we probably have some AI dollars on the front end of the processes. I think we're in the very beginnings of, you know, trials, pilots, et cetera, around some of the back-end processing, and that's going to continue, as we go through kind of next year, and beyond.
It's a very interesting time for the engineering teams, the software teams, et cetera, to be able to engage, you know, with customers around some of these problems and find solutions.
Great. Let's talk about the Cloud CapEx. What is Arista's visibility with regards to Cloud Titans' investments in classic cloud computing?
Yeah, it's been, you know, it's been interesting, right? We're just coming off of this, you know, most bizarre supply period that we had, right? We're still, you know, working hard to fulfill some deployment schedules that we agreed with those customers, you know, some time back, right? So we're continuing to execute against that. And, you know, we're improving lead times. Hopefully, by the end of the year, we'll be back to, like, a six-month, plus or minus, lead time. So we're kind of getting in range now, where we're starting to see, you know, customers, you know, start to figure out kind of what does the next level of demand with those new lead times look like?
We still have work to do around that, but you, you can see that, you know, they're continuing to invest both in the kind of traditional, classic, networks and also, you know, supporting some of that front end AI, and then obviously lots of discussion around kind of what, what other use cases can there be around AI, into the future. So I think we've-- you know, we'll end up back... You know, it used to be a turns business and a quarter, in the old days, pre-supply chain. Hopefully, we'll end up something a little bit longer than that, especially given some of the lead times on some of the key components. So somewhere around six months is probably where we'll end up.
Great. Ita, when I look at the workloads in the cloud, the majority of them are still, like, analytical based, and it's not AI everywhere. And so at some point, you know, the demand has to kind of come back for the general purpose, you know, servers. And for you guys, the 400G side, are you seeing any kind of, you know, kind of green shoots that the focus is shifting away from 800G to 400G, or it's still quite early?
You mean from 400G to 800G?
Yeah.
Yeah. Yeah, I think, I think it's a little bit early, right? Because you're just not—you're not quite there yet. It takes time, and we have silicon now for those products, but it's still gonna take some time to actually, you know, ramp the systems themselves, the software, get them qualified, et cetera. So I think, you know, 400 G is gonna drive kind of revenues, you know, for certainly through, you know, the next period of time here, right? Which, which is, which is normal, right? I mean, normally, you see silicon, it'll take somewhere like a year plus to get to a system, a fully qualified kind of revenue shipment, from there. So it's. Yeah, it's, it's, it's not unusual.
I think 400G is gonna be the, you know, the speed of choice here for a while, and then even when we go to 800G, you'll see a whole refresh of kind of 400G, 100G, platforms leveraging the silicon as well, right?
Okay. And then, on the enterprise side, you guys saw strength in the last reported quarter. We can talk about Campus Switching, Routing. How meaningful are Arista's Campus Switching and Routing portfolios from a revenue and margin perspective?
Yes, the routing is hard to break out. I mean, a lot of our routing work was initially done for the cloud and for the hyperscale customers, and we've essentially rebuilt the routing software stack, you know, for those customers, and we've definitely earned, you know, a more prominent position in some of their WAN networks as a result, right? So we have products that are shipping into those customers, some being used inside the data center, some being used for routing, 'cause that's the beauty of the overall solution, right? It's the same box. On the campus side, you know, we've talked about, you know, we were shooting for a $400 million number last year. We were shy of that a little bit just because of supply, but the demand is there.
We have a target of $750 million for 2025, and we're on track for, you know, for that number. So it's, it's, it's becoming a more significant piece of the business. And not only is it driving dollars kind of for the, you know, for the campus piece of the business, but it's also opening up sometimes some opportunities to sell into the data center again. It's another avenue of entering an account. Entering a large enterprise account is to actually engage on the, on the campus side, and then win that, and then we can kind of work our way back to the data center in some cases. Obviously, it works the other way as well, just depending on the customer.
Good. And, Ita, when I talk to Jayshree, she sounds, you know, very excited about the incremental share gains in the enterprise side from the incumbent. And we saw through earnings results, like Juniper, you, Cisco, you got all you guys had, you know, good kind of upside in the enterprise side. So help us understand how much of the strength you're seeing are share gains versus kind of the organic market?
Yeah, I think, you know, particularly in campus, like, the overall market itself, I think over a period of time, hasn't been growing that aggressively. You saw a little bit of growth recently, but it's still not a, an aggressive grower. So it's really more about kind of where we're taking share in that market, right? And we're targeting, you know, customers, high tech customers who, you know, are looking for the benefits that we can bring with, with EOS, with CloudVision, all the way through kind of their, their, in some cases, data center network through to the edge or, or in their campus networks. So I think a lot of what we're seeing is really share gains.
I mean, we're still very small in, you know, what's a very large market and, you know, there's a real opportunity to be, you know, a new supplier to that large enterprise piece of that campus market, and that's kind of where our focus is. So I think, you know, the excitement, I think, that we see is, you know, we are seeing good. You know, we're being pulled into opportunities by those customers, by those large enterprise customers, and we're becoming more known kind of in that part of the market. And it's more diverse. It's across, you know, lots of mini, what we call mini verticals inside that enterprise.
But, you know, you're seeing customers in the manufacturing world or in the, you know, Jayshree's tried to give some color, just the types of customers that we're winning now, and it's way broader than, you know, what we used to do, prior. And I think that's, that's part of the excitement as well, is that we're actually- we actually now have, like, credible solutions and references for different verticals within kind of that campus footprint as well. Right.
Great. And then putting it all together, and Jayshree gave us a little bit of a teaser on the earnings call, on the double-digit growth next year. And I don't want to front run your investor day in November, but how should we think about the bridge from the cloud, you know, this year into next year, as well as the enterprise demand, perhaps sustaining? Maybe you could walk us through some pointers.
Yeah, I mean, I'd still go back to the, you know, the thesis we've been sharing for a while now, which is, you know, after two outsized growth cloud years, we do expect cloud to be muted next year. And, you know, AI is helpful in terms of the long-term picture, but we're not sure that it changes that dynamic, right? So I think, you know, we've put out some guidance at the last Analyst Day that had us growing, you know, single digits in that 2024. When you back solve the CAGR, Jayshree has had a desire and a goal that she's been driving the organization to say: Could we at least get to 10%, right? And I think, you know, we're becoming more comfortable with that over time, hence the teaser.
I wouldn't get too far ahead of us beyond that, because I think it does assume, you know, that that cloud is muted, and that we are relying on kind of the enterprise and consistent execution and enterprise, et cetera, which, that's certainly one way to get there. And we'll talk a little bit more at the Analyst Day about maybe some other drivers there that can kind of... You know, I like to have the multiple building blocks. We'll talk about that some more, but I think it is kind of a, you know, a lower double-digit number, assuming the cloud is muted, and then we'll see where we go from there.
Great. Let's talk about the margins. How does customer mix versus product mix rank in terms of driving margin expansion?
I mean, customer mix is still by far the biggest driver. I mean, we have some tidy up to still do on supply chain. You know, we have a big focus internally in kind of putting supply chain back in the box, just in terms of, you know, cleaning up purchase commitments, you know, getting some of the in-inventory metrics back to, to normal, and then, you know, earning back kind of some of the cost increases, et cetera, that we bore, because we expect customers to come and ask for that back on the, on the front end as well, over time. So that's kind of work that's in progress.
That's been helping us kind of improve, margins as we've gone through the year, and I think we can continue to do that through the end of the year, and then it probably stabilizes from there. Then we'll be back to really the biggest driver being customer mix, and where, you know, obviously cloud, when it's heavy cloud, it will have a lower gross margin, but then, you know, still significant flow through to the, to the operating line. And obviously, when enterprise is, is stronger, we've got a higher gross margin, but then we will continue to kind of add sales and marketing, investments as we go to support that business.
Are 400G switches higher margin than 100G?
... I mean, it mostly nets out, right? If you go back and plot kind of our, you know, our gross margin over time, right, it, it tends to net out. And some of that's intentional. I mean, obviously we're, you know, we're negotiating all the time based on the underlying, you know, cost structure. So you're best as you can, you're always trying to match kind of some of the cost pressures that you're getting with some of the price reductions that you're having to give, and we manage that across the products. So new products tend to have, you know, initially have a higher cost and then kind of work their way down, but we try to manage those in tandem. So you haven't seen that much of a change in gross margin, you know, across the different product cycles.
You know, margins. You've had tailwinds this year from the COVID disruption as things are starting to normalize in supply chain. Will that remain a tailwind through to the end of this year? And how far more headroom there is for margin expansion from supply chain normalization?
Yeah, I think we've talked about gross margin kind of stabilizing at the end of the year, with maybe there's probably hopefully another, another bit of improvement kind of through the end of the year. We've kind of been stepping, kind of improving as we've gone through Q3. And then I think it probably stabilizes, and then it'll come back to more just being about customer mix and what the customer mix looks like.
All right. This is a question on investors' mind in terms of the CFO transition. As you're gonna be stepping away, is there a step pursuing an internal or an external search, and what is the targeted timing of the transition?
Yeah, I mean, we obviously talked about it early, right? To you know to give us and to give everybody else a chance to kind of absorb it and give us some time to look for the right person. So we're engaged in that now. Yeah, I'm committed to being part of that process and helping with the transition and you know so however long it takes for us to get to a good partner for the team going forward, that's what we'll do, right? But it's sometime next year for sure. It'll just depend on when we kind of find the person that everybody's comfortable with.
Okay. Then on the OpEx side, how do you guys kind of allocate the R&D priorities across multiple areas in enterprise and energy and AI, and-
Yeah, I mean, a lot of the core kind of software work is, is leveraged across all of it, right? I mean, when you think about EOS and, you know, just moving... Yeah, as we go through the speeds and we add features like routing and, and other capabilities, that's shared kind of across. That's part of the model. That's why the model works so well, right? Is that, that those investments, they are initially driven by cloud, but then they're shared across kind of other cloud service provider and through the enterprise as well, right? You know, there are certain things that we do that are for the enterprise business, right? And we've been working hard to become a little bit more disciplined about carving those out. And, you know, Jayshree has a big focus there in terms of making sure we're doing enough.
That's CloudVision, some of the visibility tools, automation, you know, some of the campus, you know, only kind of features that we want to fill in. But that's still a relatively small part of the overall lift, just in terms of where the resources are going. So there's good commonality across the different parts of the business, for a lot of the R&D.
Okay. And, it'll be helpful to know where we are in the 400G upgrade cycle. Are we, like, 20% through or 30%, or?
Yeah, it's been harder to kind of... You know, we used to think of it as like, well, 40G, then 40G goes away, and you get a 100G, and then a 100G goes away, and you get-- And then with this, this last set of products, what you've really seen is, you know, a 100G is continuing to be deployed broadly, right? You've got 200G in certain cases for certain customers, and then you have 400G. So I think even after we have an 800G offering, and it's been deployed in certain cases, you're still gonna have some customers who will be deploying 400G and 200G, right? So it's been, those, those high speeds are becoming...
You know, the silicon is changing underneath, and you're getting a whole new set of more efficient products, but customers are still deploying different configurations depending on, on the use cases. So I think, you know, we're, we're obviously well into the 400G deployment, but if you think about, you know, the business, say, last year, this year, I mean, we're still deploying a lot of 100G too, kind of in that, in that timeframe. New 100G, more efficient, power efficient, et cetera, but, but new 100G. So I think we'll see... You know, we're, we're obviously getting a lot of dollars from 400G right now, and we'll continue to, and probably continue to even after 800G shows up. Yeah.
Okay. Ita, when we talked to Cisco, I mean, they acknowledged that, you know, they, those guys, I mean, those guys have lost 400G to you guys. And on 800G, they're, it seems like they're putting a bigger focus. How should we think about competitive dynamics going into 800G?
Yeah, I mean, I think, you know, we've gone through multiple cycles now where, you know, it's 100G. 400G was supposed to be, you know, a cycle where there was a lot of noise around, you know, competitive products. We'd expect the same again at 800G. I'm not at all surprised. If I was in their shoes, I'd be putting wood behind 800G as well. But again, it all comes back to, you know, we have to execute well. We need to... You know, if we deliver the right products, if we work with these teams and these customers, we can earn our fair share of that business. And, you know, we'll let the competition take care of ourselves.
But we've been. You know, the team's been doing a really nice job of, you know, executing well against products with these customers. We saw it at, at 100G , we saw it at 400G . You know, again, there's a huge focus on 800G now inside the team. So, I mean, I wouldn't bet against them, you know, continuing to drive that execution, because it really is all about execution for these customers, right? Give them a product that's reliable, has a software stack that you can ramp straight away. That's what wins you that business, right? So we need to stay focused on that. And, I mean, I would expect them to compete hard for that, but I think the team, the team is good at this, we just need to deliver.
Great. You know, I mean, we were just, we were just coming back from the NVIDIA, you know, Fireside session, and they had this chart where InfiniBand is growing, like, 4x. Obviously they're seeing the success on InfiniBand, on the first wave of 800G. The big debate is, you know, like, what will be, you know, kind of the mid to long-term mix of this market? When we talk to fabless companies like Marvell and Spear, it sounds like Ethernet is starting to get adopted, as well, and are being evaluated right now into next year.
We'd love to know your thoughts or if you're seeing any kind of green shoots for Ethernet adoption for 800G.
Yeah, I think, you know, at the end of the day, what's gonna matter is kind of what gives you the best AI-enabled network, large network. I mean, that's what we're solving for, right? Is being able to have these large, you know, multi-tenanted large footprints with, you know, from a customer's perspective, with multi-vendor, you know, multi-providers and lots of, you know, lots of contributors to that architecture and that network. I mean, that in our minds is what's going to win. It's standards-based, right? And that is, that's where Ethernet's strength at the end of the day, right? If you look at the history of how Ethernet has evolved over time, right? It is a standards-based, broadly deployed architecture, and we believe it can solve any problems that need to be addressed around AI.
There's some work to do there, but we're continuing to do that, and it'll continue to evolve, right? But customers are gonna look for that environment, right? That's what allows them to, to scale to the levels that they're gonna want to scale this to, if AI is, is everything that we believe it's going to be, right? So I think we're betting on Ethernet because, you know, we've seen multiple different architectures kind of, try to compete against Ethernet. But in that environment where you want to have this kind of large, standards-based, open, architecture, Ethernet tends to, tends to win. Now, again, we have work to do. We need to do it, but I think, you know, from a customer perspective, that's a solution that they're using today. It's deployed pervasively today. They have the expertise there.
So that we believe that that's where it's going to end up. It won't be 100%, but there'll be, you know, those large-scale footprints will, we think will end up deploying Ethernet. But again, there's work to do. We need to, we need to just execute and, and do it. But customers are definitely interested in making that work.
All right. And then the threat of White Box, you know, makers in terms of competitive landscape, and that question never kind of goes away. When we look at the Dell'Oro data, it seems like, you know, that share has kind of stabilized for White Box makers. Can you just talk about what you see in the market for White Box makers?
Yeah, I mean, I don't know that there's been any fundamental shift in white box for a while now, really, right? I mean, it's been—you have kind of the two large players that do their own, then you have kind of the dual co-development arrangements that have been in place, but it's been pretty balanced, right? And, you know, the more complexity that there is, the more change that there is, that doesn't tend towards, you know, I think, more white box, right? If anything, it probably balances it, balances it out and maintains the status quo. So I don't feel that we've seen any shift kind of towards white box for a while now, but that question never goes away, right? We've been answering that since I joined Arista.
I think that was one of the first questions, big concerns everybody had back then, so-
All right.
Yeah, I don't think it's changing that much.
What has to change for you to change your share position at kind of the large cloud titan that you know I think investors expect at some point you'll have a shot at? Is it something on their side or something on your side that you have to kind of technology-wise get to meet their requirements?
I mean, I think on the technology side, there's good alignment, from a technology perspective, and, you know, Anshul's commented on that before. It really is, but it's, it's a huge decision for, for an organization, to make. So it's really in their hands to decide kind of, you know, what makes sense for them. I mean, we'll continue to, you know, to collaborate, share technology, et cetera, and, and it really is in their hands. But again, you know, from my perspective, I've said this many times, it's not something that should be in a near term, 'cause we can't control it, right? So again, we'll, we'll continue to, to build those relationships and, and share technology and, and we'll see where we end up.
Great. Let me pause here and see if there are any questions in the audience.
I was hoping we could go back and talk about margins a little bit. In the most recent 10-Q filing, there was a note that one of the contributors to the decline in margin was, provisions for excess inventory or, inventory obsolescence. So just hoping you could talk about that a little bit.
Yeah, I mean, obviously, our inventory has been growing, and particularly kind of on the raw material side of things, you know, the inventory levels have been growing. We always have a certain level of reserves that we carry to cover that inventory. As that inventory has been growing, we've been increasing those as well. I think it behooves us to, you know, make sure that we have kind of a consistent level of coverage on those inventory balances. You know, when we came into this whole supply situation, we always knew it wasn't completely risk-free, right? But we thought it was a manageable risk. I think we still feel that way, right? We're, you know, working through the purchase commitments and inventory and making good progress there.
But at the same time, as that inventory balance grows, we want to kind of, you know, we want to continue to maintain some coverage on that, not because of something that's going to happen today, but obviously, it's the tail of that inventory that you want to make sure you've got, you've got some adequate coverage for. So we are keeping an eye on that. But I think by the end of the day, as of relative to the inventory size, it won't be that different to what we normally do. It's just obviously as the inventory is growing, you're seeing those increases go through the, the COGS numbers.
[inaudible] All work in progress, the inventory, not finished goods?
I mean, the finished goods is still, I mean, if you look at it, our days of inventory on the finished goods side is still not nearly back to where it was pre.
Yeah, it was.
Yeah, yeah, that's the. I mean, it's because most of the inventory growth has been in the raw materials, that's kind of where we're, we're addressing that too, right? Just 'cause we're really kind of just working on the basis that the inventory balances are growing, right? And that we should kind of cover that off.
Thank you. Maybe if you can maybe double-click on the kind of challenges Ethernet has to solve now to gain more share on InfiniBand, and maybe kind of from the get-go, what InfiniBand has that Ethernet didn't. What are the different obstacles that Ethernet has already surpassed to kind of gain more share on InfiniBand? Just double-clicking on the challenges Ethernet already surpassed, and what's remaining.
Okay, I'm gonna try, but remember, I'm the CFO, so it may not be very satisfying. Yeah, I think, look, historically, InfiniBand probably had higher speeds, right? Ethernet was kind of was needed to catch up on speeds. I think we're pretty much catching up on speeds now, right? And then the other piece was just kind of the, you know, the management of the traffic flows, right? And there's this whole RoCE discussion, right, where Ethernet now also has basically learning from InfiniBand over time, now has some of that, that has the ability to kind of manage those flows as well in Ethernet. So a lot of those things have been, you know, so a good number of those things have been solved.
I think Jayshree put it well on the conference call, which is, like, neither Ethernet nor InfiniBand today is perfect for AI, because that's a whole new kind of use case and challenge. So I think both technologies will need to evolve and will want to evolve, right? And a lot of it is around you just managing traffic, making sure that you get to that lossless or, you know, Anshul likes to say, almost lossless, but really lossless, delivery, 'cause that's what you need for AI, right? So there's some more software management, you know, management of those flows that you can do to make sure, right, that you solve for that.
So that's an area for sure that we'll continue to work on between buffers, and then making sure that you have that predictability, and that you deliver every packet gets delivered, right, is a focus for Ethernet. On the InfiniBand side, I know there are other things that need to be solved there too, but a lot of it is just to do with the fact that it doesn't scale to these multi-tenanted, and that it's not open, and that it's not kind of... You know, you don't have that same standards-based supply base, that you do on the Ethernet side.
Questions? Ita, are there other adjacent markets that Arista is considering entering? And also, if you can talk about the software opportunity. Because when I look at the three, I mean, now the three networkers, Juniper, Cisco, you guys, you all talk about software and other unique drivers to your software attach rates.
Yeah. So I think, you know, we're a little different in that we haven't tried to take the core operating system, EOS, and bundle that separately, right? I mean, for our kind of that larger enterprise customer base, that's not something that they're interested in doing. It doesn't resonate well with them. So it behooves us to get paid in terms of, you know, the actual pricing on the port and the hardware, but we don't have a separate software stream per se for that, right? But obviously, we continue to do a ton of work there, and that's what earns us kind of the market share that we have, right?
So then the other piece of the business is more around the subscription standalone products, where customers really do have a choice, and they get to buy those, those software products if they think they add value, right? And that's CloudVision, that's some of the visibility tools. It's, you know, it's some of the Ava security tools. Right, so those we sell separately, and we sell as a software subscription. I mean, in terms of, you know, do we—what else do we do?
I think, you know, we kind of balance all the time between, you know, if we only did two things well, if we, you know, executed well in cloud and grew with that business, and if we, you know, if we continued to execute in the enterprise and campus and take our fair share of that market, like, that drives a very healthy business in and of itself, right?
So you need to stay. There's an element of focus that you should have in executing against that, and I think what you're seeing us do is kind of fill in where we have, you know, gaps in that, in the campus offering in particular, and where, you know, we've been partnering or needed to partner, and where it makes more sense for us to actually do something ourselves, whether that's the, you know, the WAN product that we announced, whether it's kind of the AGNI, the NAC product that we announced, right? So stuff like that, where, you know, it really truly does make sense from a customer perspective, for us to kind of solution that as opposed to, you know, introducing something else, into the middle of that.
That doesn't mean we won't partner and that there won't be things we'll partner with. There always will be, right? But there are certain areas where it's somewhat disruptive to the overall solution where we drop things in, and that's what you've been seeing us do over the last while, and we'll continue to do that.
All right, great. We can wrap it up here if there are no questions. We still have a few minutes if you want to talk to Ita, but, Ita, good luck in future-
Okay, thank you very much.
Thank you for coming to the Citi conference.
Yep, thank you for having us. Thank you.