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

Great. Good morning, everybody. Welcome to the Arista fireside chat at the Goldman Sachs Communacopia and Technology Conference. My name is Mike Ng, and I cover Arista and Comtech here at the firm. It's my privilege to welcome Arista's President and CTO, Ken Duda, alongside Chantelle Breithaupt, who's the CFO. We have about 35 minutes for today's presentation. But first and foremost, Ken, Chantelle, thank you so much for joining us today. It's an absolute pleasure.

Yes. Thank you for having us. Great to be here after a long weekend.

Yes. Exactly.

Well, to kick things off, Arista crossed a historic milestone last quarter with its first $3 billion revenue quarter. Networking is an absolutely critical part of client campus, data center and AI environments. So maybe start and talk about how you're balancing the strategic investments between scaling the core AI and data center switching fabric product portfolio versus what you're doing in software, campus and routing.

Yes. Thanks, Michael. We are so excited about what's happening in networking as the network is absolutely the lifeblood of so many different kinds of operations. And it's to connect everything together. When the network isn't working, nothing working is super reliability critical, and we feel like our opportunity to continue to grow and expand in that space is just really exciting to me. And with regard to your question about how do we invest, one of the great things about our business is we don't have to think about it in those terms, right? It's not sort of like, well, we do invest here or do we invest there? No. What we do is we work very closely with the world's most sophisticated hyperscaler operators, customized solutions for them, bring in the newest technologies, you can see that now with XPO, liquid cooled optics and that sort of thing, get all that working at scale. And then that -- those investments that we make in the hardware and the software, the technology naturally flow across the ecosystem into the broader markets, the specialty operators, ultimately into the enterprises. And so we're investing along the entire chain there. It's not a sort of A or B decision.

Yes. The only thing I would add to Ken's comments, which absolutely resonate with me in the sense of how we approach the company. It's a benefit and it's not confusing for our capital allocation strategy or investment strategy when you're just pure-play networking, right? So we love all those children equally because they're all important in networking portfolio. And if you think about our guide for this year being $12.6 billion, roughly 40% growth, that's a lot of absolute dollars added if you maintain 8% to 10% of R&D to revenue for Ken and his team. So we're super excited, and we have room for all of those things to be invested in.

Great. I mean that's a fantastic segue. Maybe you can talk a little bit about that revenue guidance for this year. What specific customer demand signals or supply chain improvements and availability inform that upward revision? And where do you feel like you have the most visibility today and potentially opportunities for upside optionality?

Yes, sure. So Jayshree and I were super excited to raise the guidance by over $1 billion in the last call. And I would say 2 things were true to make that a comfortable position for us. The first thing was if I go back to the Q1 earnings call back to the scenario that no one thought was popular, but was definitely needed talking about supply chain constraints across the industry. We got more comfortable with supply chain. We leaned into purchase commitments. We had some of our vendors rally that we needed to rally. And we had to make sure the team inside the company was on their A game to make sure this happen. So supply chain was something we're comfortable with, Michael. And the second thing is by the time you get to that kind of August time frame, we have 2 quarters of PO transactional visibility, so we could see through to the end of the year. So supply chain improvements with PO visibility timing, those 2 things came together to give us confidence. I think if you were to look at what could give us even more upside to that guide raise, it's that the supply chain eases up even more. I think that we're not out of the woods. I don't think anyone in the industry is out of the woods for supply chain constraints. But if that gets a little bit more opened up, there could be some potential for upside.

Great. Are there any particular areas of the supply chain or component availability that would bring true in terms of things getting better, leading to potential upside?

I don't think it's because it's different customers have different things, but it's everything from some of the peripherals to some of the main things, main components that come into it. So we have agreements on chips and memory, but there's PCBs and there's -- sometimes there's power cables and things like that, that just kind of whack-a-mole through the year. So we'll just keep our eye on it.

Great. I wanted to ask about EOS and competitive differentiation. It's an important competitive moat for Arista. I think it's a competitive advantage when you put Arista against any white box vendor. So maybe you can talk a little bit about EOS, its differentiation and has that changed over the years as white box operating systems potentially get better?

Yes. The situation with open source NOS is that it may be open source, it is not free. It takes a deep technical expertise to actually get all the components together, assemble them into an image and get that to actually work on the hardware platforms and use cases of interest. So I'm not foreseeing any deep penetration there outside of a handful of hyperscaler operators who I don't believe are even saving a single dollar here, honestly. What they're gaining is multi-sourcing of their software environment. They cannot accept the operational risk of being beholden to a single vendor. And that's the core driver, I believe, for what's happening in that part of the market. The competitive differentiation of EOS is very strong in the hyperscaler use cases, we're seeing still good traction with EOS even as they attempt an open source strategy, especially in high-demand routing use cases where you need fast reconvergence, you need to deal with large routing tables. You need rapid convergence when something changes in the network, getting the hardware tables updated quickly. And under the memory constraints of the platform, we see significant advantage there. But for the broader market, I don't even think the open source operating systems are a realistic option. And there, the competitive differentiation of EOS lies in the quality. The fact that you can count on the software really working and customer after customer tells me how much they appreciate having the switch software stack that they can really count on.

And Ken, maybe just to follow up on that. Are there certain products or areas of the network stack where EOS is critically important, whether that be in the spine or the leaf or certain AI use cases, whether that be like scale out versus scale across? How would you just think about where EOS' competitive differentiation matters the most and shines the most?

Well, again, I think it's really important to think about the hyperscaler market separately from really any of the other market segments. The new cloud operators are in a different category. Certainly, the enterprises are in a different category. And part of the strength of Arista is applying the same software. We have the same operating system, not just the same brand. It's the same code running across that entire spectrum. And the advantages that we bring vary across that spectrum. So there's not a simple answer to your question. I would say that what we see in the hyperscalers where we have the greatest differentiation is, again, in routing. We have a very advanced routing stack, all the policy controls, all the different tunnel encapsulation types and MPLS stack that's simply not available in the open source world. Whereas with the Neocloud operators and with the enterprises, it's the quality of EOS, the manageability through CloudVision, having a single software that runs across all of their use cases, whether it's a campus use case, a WAN use case, a cloud use case or, of course, in the data center, having that consistency of operation is a really important differentiator for us as well.

Great. I wanted to ask about merchant silicon in your high-end Ethernet switches. You guys obviously have a really great relationship with Broadcom, I guess, most recently for the Tomahawk 6. What are the advantages of using merchant silicon versus developing your own from a supply availability perspective, has working with merchant silicon and the allocation risk associated with that been a concern at all?

Well, the supply chain, of course, is a great concern. We have supply, we have demand. You need them both. And it's an industry-wide problem. There's nothing unique to Arista about this. In the semiconductor side, in particular, the merchant silicon versus in-house silicon isn't actually the big issue because the bottleneck isn't the chip designer, the bottleneck is the fab, right? So that doesn't really trying to matter very much. From a supply allocation point of view, obviously, we've been managing that through being aggressive with our supply agreements and our purchase commitments, and we feel good about our position there. Arista has been a merchant silicon company from day 1. I think our focus has been to add value through the system and through the software, better software, better integrated into the system. We have many examples where our software unlocks capabilities of the silicon, which is not unlocked through other vendors because of limitations in the software stack. So I think we've proven in the market that, that merchant silicon approach has been successful.

Great. That's very clear. Chantelle, maybe on gross margins. Arista has 2026 gross margin guidance of 62% to 64%, and you maintained that last quarter despite cost inflation, customer mix shifts. What are some of the puts and takes for gross margin this year?

Yes. I think that if we go back to our Analyst Day we had, I think, in October '25, we stated the 2026 margin guide to be -- gross margin guide to be 62% to 64%. And to your point, with all the different things that have come in, we still have that guide. So I'm very proud of the team to allow us to do that. So what are the main components? The main components usually for us is end customer mix from that perspective. And then the other one, too, that I would say is coming into this year a little bit similar to our peers is there's tariffs this year, they have tariff refunds. And so if you look at that 62% to 64% guide, maybe 30 basis points, I would say, on the year is going to be tariff onetime coming back into the P&L, just to be transparent about that. So if you think about the other pieces that are coming to mix. So if we talk about kind of the price component inflation side of it, our philosophy earlier this year was to measure 2 or 3 times cut once. So we tried to sweep in everything we thought could happen in the time frame, component inflation, maybe other different tariffs, et cetera. And we had one price increase, but it was just meant to offset the inflation, not meant to be accretive to our gross margin, but to hold it. And we only applied that to POs in the future to customers that had significant pieces of their bill of material. So it was pretty curated. It wasn't just everyone got -- it wasn't like, everyone gets a car, everyone gets a price increase. We're pretty segmented and pretty accurate on that one. And so those will start to come in those price increases as we work through the backlog into new purchase orders. The component inflation comes and goes depending on our supplier agreements. And so we've tried to make that neutral coming into this year, and we'll see where we get in 2027 based on the market. But a lot of great work by the team, both the commercial and the supply chain side, and we've been able to steer the ship within that guide so far.

Great. And could you just expand a little bit on some of the price increases implemented by Arista to recover that cost inflation, whether that be timing, magnitude, types of products and anything you could talk about as it relates to next year's gross margins, just given some of the puts and takes on the price increase tailwind and then the tariff refund kind of, I'll call it, a reversal.

Yes, sure. I think that from the perspective of the price increase, the largest component is the memory in the current kind of conversation. And so if the bill of material had a higher memory component to it, we would have a price increase in that situation and explain to the customer, trying to be as transparent as possible how that kind of flows through. If it didn't have -- not a lot of our products have high memory bill of materials, so those weren't necessarily impacted. And then you'll have some of the chips and some of the other factors that go in. So those were all averaged out to a particular product and particular conversation. I think going into next year, I'm not guiding next year at this point. But I think from the perspective of -- I don't know if we'll have to do another price increase. We'll have to see. But hopefully, the one we've done will carry into next year and keep us margin neutral within the same kind of range territory.

That's great. Yes. Great. Maybe we'll follow up on something that you guys talked about earlier, which is the multiyear purchase commitments. I think last quarter, they nearly tripled to $9.7 billion. What does this component commitment talk about or reflect in terms of your confidence in delivering and sustaining and fulfilling the consumer demand over the next couple of years? How much visibility on the component side do you actually have?

Yes. So we're -- as a company, we get -- we have many different frameworks. Some people call us conservative. I would call us prudent. And one of the ways we're prudent is the way we operate our balance sheet. And so when we do lean into things, it is with intention. And so our lean into the purchase commitments should be seen by this audience and those that will see this after as a demand signal, right? So that's a demand signal that you're seeing there. It's not that, that's all inflation on components, that's demand. So I think that's very exciting from our perspective regarding that. I think that if you look at the visibility side, so our biggest component are the chips, and that's 52-week lead time, right? So we're already leaning into fast forward a year from now. So how do I reconcile it 2 quarters of visibility, but I have 1-year lead time. The great thing about our portfolio is it's pretty fungible. And so it's not like we're ordering specific chips that we can't use or building specific products we can't offer a different customer. So again, going back to the flexibility of the portfolio, the fungibility of the components, we feel very comfortable we're not walking into a large obsolescence inventory risk environment. And so -- and some of the customers we have, the larger customers, Neoclouds, hyperscalers, we're having 18, 24-month technology conversations, right, Ken? So we have an idea of where they're going, and we're helping them design. So we directionally know without a PO in hand where they should be going. So we balance those 2 things.

Michael, if I could just comment. This is an example of how the alignment that Arista has between the hyperscaler side of our business and the enterprise side of our business is so useful because we've got competitors that have different business units going after these things with completely independent engineering engagements and then they're taking a lot more risk of component obsolescence than we are when we can utilize the same switch models. Which, of course, obviously contain the same components across multiple use cases, multiple types of customers, they're operating at different time frames. Some are a lot more aggressive than others, as you can imagine, in adopting the new technologies. And so we have a lot of sort of room there to find demand downstream.

Great. Yes. That's very clear. And maybe just on the other balance sheet item that investors focus on a lot, deferred revenue. How much visibility do you actually have in the deferred revenue recognition into actual revenue? And does things like data center readiness, the availability of power affect your ability to recognize revenue in terms of your products being shipped?

Yes. So just to remind, because sometimes it's not clear to all, deferred revenue means it's been shipped invoiced and usually cash collected within the payment term. So that's what's in deferred. Sometimes people confuse it for backlog. It's already shipped. There's no supply chain issue because it's already been deployed. So from that perspective, we have an idea. So deferred revenue is new use cases, new products, right? So right now, you think about what's left for Arista, it's scale out and scale across, eventually scale up. Those are new use cases, new product 1.6T. So those are the types of things, 800 gig that goes into deferred. And visibility is as good as the expected time frame it's going to take to get to the revenue recognition of the criteria. So of course, we have an idea, but we don't determine it in the sense of when it actually gets accepted. So we'll have a time frame, usually within a quarter or 2, and it's usually landed pretty well. But I can give you some anecdotes in the sense of what we're hearing, and I'm sure that you guys have many more than I do, given you cover many companies. But we had a large customer say to us, they actually had -- they said the 3- to 4-week delay that their data center stopped constructing because an employee was hurt, and they had to prove that the facility was okay for employees to work. So that's almost a month gone already just because of an employee injury, which you have to take very seriously. So add the power, the cooling, the facilities, the cabling, there are many reasons why it can take 18 to 24 months for some of these things to be completed. And so that's the average time frame each is bespoke. But every quarter, things are coming off being accepted and things are being added. And so I think that healthy kind of turnover, I'm comfortable with in the way that it's acting.

Yes. And just to remind us, the nature of the deferral, is it because you're promising a certain level of uptime and performance and it has to be in production for that.

Well, different things. It can be -- the thing I would say is the customers really enjoy it because we have skin in the game until they've gotten to where they need to get to. We don't ship it, leave it and say, good luck to you. Not to be flippant, but that's the opposite in that.

The customer has to accept the product as working in their case.

That's right. So it can be site by site. It can be performance. There's many different things for different customers, but it keeps us engaged in our SEs and our technical team engaged until they've gotten to because you think about some of these clusters are some of the largest in the world, right?

Great. I wanted to ask about Neoclouds. It constantly comes up as something when I do my channel checks and expert calls as like something that Arista is doing really well with in terms of a customer cohort, but maybe a little bit underappreciated. So talk a little bit about the Neocloud segment. How is competitive intensity in the segment evolved? Are these all like net new logos? Or are they expansions of existing relationships? How important is Neocloud in your AI outlook?