Arista Networks, Inc. Aktie

Welcome to the second day of our conference. Hopefully, you had a great breakfast. Today, I'm very pleased to host -- and I'm not saying it's just to say, but I'm -- really mean it, I'm -- the most favorite CEO in my universe. And she -- we sometimes argue, but it's always coming with a lot of respect to one another. So thank you, Jayshree.

So Jayshree Ullal, CEO of Arista; and Todd Nightingale, COO of Arista. And I've known Todd for many years from his days at Cisco and then Fastly that I took public and now at Arista.

So we're going to have a discussion for about 45 minutes. I'll try to keep it tight for about 40 minutes or 35 minutes and let you ask questions if you have any questions. And the discussion, it's -- as you know me, I'm not a quarter guy. It's not about the quarter. The discussion is really the value of Arista, the sustainability of trends. We do have a spending cycle in the industry with hyperscalers. And through the spending cycle, there are changes to the architecture. There are changes to the network. Customers are deploying certain technologies, and we want to go into it. So the purpose of this session is really to understand the core value of Arista.

And with that, so, last quarter, you raised the full year guidance from 25% to 28%. And I remember when we looked at your guidance, I'm talking about like a year ago, when you look at your guidance, and I told investors, these numbers don't make sense. They're too low. The industry is growing faster. The company, if you take out this and you look at this, it's supposed to grow faster.

So the question is what drives the growth? And I know this is kind of a soft question to start our discussion. What drives the growth right now? And how sustainable is what we are seeing in the past few quarters?

Well, first of all, Tal, you're my favorite too. There's not too many investors who we can say I know for decades. So that kind of times both of us a little bit. And if I start talking about mainframes, it will really time me a little.

But coming back to the growth vectors for Arista, as you know, Arista really made networking sexy again. And part of that was focusing on those growth vectors. We pioneered the leaf-spine architecture, and we really redefined or reimagined networking for the cloud, right? And as you know, we power some of the world's largest cloud titans, as we call it, with this leaf-spine architecture that was different than the standard blocking loss-oriented enterprise architecture.

So today, one of our growth vectors continues to be the cloud customers. But a lot of our growth in the last decade came from what I call the front-end network, the more visible network where we were connecting to compute, storage, WAN, building the active-active leaf-spine, et cetera. In the last 3 years, we really had to study the back end of the network that was historically built through proprietary technologies like InfiniBand. And the first thing we set to work on was really defining Ethernet for those use cases. We now call it AI Fabric, scale up, scale out, scale across, scale up within a rack, scale out between racks and scale across between data centers themselves. And that's really going to drive a huge part of our growth.

Now with me is, of course, Todd Nightingale, our President and COO. And you know he reimagined the campus and enterprise in some of his prior assignments. But I think there's another one coming there as well. So that will be a huge vector of growth for us, where post-COVID, the campus looks very different from the carpeted buildings and you really need a uniform experience client-to-branch everywhere. So I think the AI and campus are going to be our fastest-growing segments. That doesn't mean the others don't grow, but they grow slower.

Yes. And when we look at hyperscaler spending, it's up giant this year, was up giant last year. When you talk about -- when you talk to them about their build-out plans and how sustainable is the cycle? What needs to happen for the cycle not to be just 1 or 2 years or 3 years, but rather a long-term cycle?

Yes. This is a really good question because even I get scared, I look back at our '99, 2000, that was a short-lived cycle. As we all know, it started out the great burst of Y2K and then collapsed. I think the biggest difference here is there is a very sustainable advantage for AI, both networking for AI and AI for networking. So it's going to traverse not just the hyperscalers but go into the Neocloud and enterprises as well. That's one.

And we haven't yet seen all the killer agentic AI applications yet, that's just beginning. So I think this is a multiyear, in fact, a multi-decade cycle. Now having said that, some of this CapEx is just stunning to me. We never thought of CapEx in the trillions, let alone billions back when we started this. So it's important to understand, though, that the CapEx has many, many pieces to it. The biggest one being compute, power and obviously, all the real estate itself, the cooling, the chilling, the buildings, et cetera.

Networking is a small piece of it. But becoming a very strategic piece, meaning I can't let these compute nodes idle. And this is where we really come in because in the past, you could have a bit of underutilization, over utilization, over subscription, it didn't matter. Here, these idle CPU cycles is idle money, right? So the ability to -- for this CapEx to continue growing, I think is high just because -- it's not only the large hyperscalers. There's a force multiplier of others coming into it with the Neocloud and enterprises. I think it's sustainable as far as I can see, which is generally not more than 2, 3 years. But I think this is a transformation much like the industrial revolution or the Internet revolution.

Now, will it have its values and peaks and highs and lows? Of course, Newton's laws of gravity means nothing just goes up to the right, there will be some ups and downs. But overall, it will still go up and to the right.

When it comes to your position, the market is expanding, meaning we started with 4 hyperscalers then Oracle, then Neocloud, now a lot more second-tier cloud. Talk about your position, market share, but not the numbers, rather your position as the market expands to other areas outside of the big hyperscalers.

Yes. And so I think, first of all, if you look back at 2022, our AI position was outside looking in. We were seeing all these GPUs built by NVIDIA connected largely by InfiniBand. So it was -- our position was 0. But interesting thing even back then is customers were asking us for our thought leadership because they were building these proprietary back ends. So today, I would say 3 or 4 years later, our position with the top titans, hyperscalers or even some of the Tier 2 cloud providers is medium to high. There's still existing networks to deal with. But you've heard me often talk about 4 out of 4 or in the past, plus 4 out of 5. We now have production, we went from trials to pilots to production in all these 4 major ones.

Now having said that, there's a whole lot more to do there now because largely, we were looking at production networks that are scale out, connecting all of these scale-up technologies. Now we have the opportunity to not only expand scale out, particularly with the advent of OpenAI defining multi-plane technologies with MRC, but also to scale across because more and more power is a real issue. And I think this is going to be -- it used to be a niche DCI data center interconnect use case, but the power of routing, multi-tenancy, SRv6 and tunneling across this, encryption, working with coherent optics, is going to expand how we build those scale-out network -- scale across networks.

And then scale up is a new market for us, largely today deployed by PCI Express and NVLink switching. But with the Ethernet for scale-up networking ESUN coming out as a spec this year, I think you're going to see a lot more action on that within rack as well. Much more hardware intensive, much more latency, much more defined for high-speed training as well as inference, job completion time, time to first token is all important over there. So this market is multiplying to multiple use cases and is, frankly, redefining the data center as we knew it.

Right. When we think about -- by the way, Todd, I'll get -- I have some questions for you.

Oh, yes, he will get your day in the sun.

Take your time.

I want to first finish the hyperscalers and then we'll get to the important stuff.

Then we will go to [ the enterprise ] campus.

When we look at the hyperscale environment, over the last few years, there was white boxes penetrated to the market. So some companies started with it, Amazon and then Google and even existing customers migrated and expanded the use of white boxes. But as the market goes up to scale up and scale across, talk about your position in this market versus white boxes. And does it mean that your -- if the market -- again, I'm thinking about it conceptually, not about numbers. If the market shrank in the last few years because of white box competition, is the market going to expand again for you now with scale across?

I think the market never shrank by the way. Our total available market has gone from $50 billion to $105 billion is our stated number now. I think by 2030, it will be $150 billion. So there's no shrinking in our market. In the market, there's always segmentation. And anybody build their own PC these days, anybody build their own iPhone these days. Great. You are the candidates for a white box. If you're going to look to build something ultra cheap, don't need a lot of features, white -- we have always coexisted with white box. We will continue to because there's a sort of cost-effective do-it-yourself mindset that will require that.

Now with AI, it's actually scary because you don't want all your AI traffic to be going through across something you do yourself. So we're actually seeing less white box in AI installations and more branded vendors like Arista, like NVIDIA. But that doesn't mean it doesn't exist. It does exist when a hyperscaler customer is jointly trying to develop something at the scale-up level and design their own board and almost virtually give it to a CM who can build it like a white box. So it ends up being a joint engineering, not so much do-it-yourself, but manufacture it with the CM type of mindset. So that's going to continue to exist.

Low software, high hardware, highly defined, different margin profile, typically 20% to 30%, so low software content, we've always coexisted with that. It's been with us -- I mean, I can remember the IPO, when we were $2 billion, $3 billion in market cap 2014, where the first question would be, how are you going to compete against the white box? And as you know, we've coexisted and not only competed, but in many cases, they've chosen us over.

There is a new concept that's emerging for us called the blue box, which is hardware building very, very reliable, high signal integrity, hardware is a nontrivial effort. So what we are seeing now actually is the resurgence of a blue box where more and more customers are not only requiring Arista for the high-level software but for the low-level software, we're building a lot of diagnostics at the power supply level, counters, active components, passive components. This is what we've done for the last 10, 15 years because we have fantastic partners in Broadcom and other switch suppliers. They built great chips, but we build the software to twiddle their chips. And so we literally removed the software development kit of our chip vendors and put in our own network diagnostics and platform level software. And so we think a blue box, which is basically a white box, better designed for mission-critical applications will also factor in and is factoring in already.

Got it. I have many more questions for you, but I don't want him to be bored. Todd, you joined recently, what's your mission statement? What are you focusing on? I mean recently.

I am almost a year in.

I know. I know recently in Arista...

He doesn't like it when I call him a newbie.

When you look at the tenure of the length of position of the other managers, you are recent.

That's true, I am an infant on our team. I think, really, my mission comes in twofold. I mean, one is to ensure that the operations at Arista are smooth that we're always able to deliver to the customers that need it and supply everything worldwide. The second, and I think maybe what's most important when we talk about driving growth is fundamentally bringing the mission-critical differentiation that Arista has to every network. We spend a lot of time talking about the hyperscalers, and that's an incredibly important part of our business. But Arista has always stood for mission-critical networks, the most reliable, most available networks in the world.

And 5 years ago, even 10 years ago, when we spent a lot of time talking about it, there was a divide between people who needed mission-critical and the nice-to-have networks. I don't think that exists anymore. Everyone, every network, every enterprise network is mission-critical right now, and my goal is really to bring Arista to every enterprise.

And how do you make the campus business grow bigger? What are the challenges in campus that you don't see in enterprise data centers -- data center enterprise? What do you need to do in order to make it, as Jayshree said, like a giant business, a big business in the future?

Yes. So I think the first is that the breadth of the portfolio goes far beyond switching. It's why we've spent so much time investing in the rest of the network. 7 years ago, Arista invested heavily in the PoE switching and launching our campus network. That was fast-followed by WiFi. And then just recently, in fact, the day I joined, Arista acquired Velo. And that brought in kind of the missing piece with the SD-WAN technology which really completed the portfolio. Then we had enterprise routing, switching, both spine and leaf, all of the WiFi and now SD-WAN to really attack the branch. That's the first.

And bringing that portfolio into the Arista world and really delivering with differentiated quality the way we always do, that is the mission we're on right now. In fact, we've got the leadership here who's driving that.

I think the second piece is looking at the universe of customers. The data center enterprise data center and hyperscale data center, the amount of customers that make up the big chunk of that market are relatively small, maybe like the Global 2000, et cetera. But campus networks span a much larger group. And we are working very hard right now to evolve the go-to-market to be able to reach all of that TAM, not just in the U.S. but around the world. And that expansion to the whole market sort of go-to-market effort is key.

So elaborate on this, what do you need to do with the go-to-market in order to address the broader market?

Yes. I think one really interesting piece is the thought leadership. And we've seen this -- I've seen this even in my one year at Arista exploding the amount that we see Arista being sort of publicly part of the conversation. And in campus, especially, this has been a huge focus, including the exposure that we need from experts around the industry just a couple of weeks ago, the most recent Gartner Magic Quadrant came out and for the first time, Arista was in that leaders' quadrant, which has opened up already so many doors because people, of course, look at the leader quadrant there and look to...

I just want to take a moment to congratulate, it took us 7 years to get there. We went from being a niche vendor to a visionary and finally, the leadership. And by the way, we're the youngest kid there. Everybody else has been around 25 to 40 years. We entered the market over 5 years ago, right? So congratulations.

There's a great graph that shows the trajectory year-over-year.

So the question is what's your value proposition, meaning what do you bring? What do you -- how do you disrupt the market? What do you bring to the market that is different than what the others are offering?

Before he answers that can I tell a story about that?

Yes.

Okay. And then you'll know the value proposition answer. So about 7 years ago, we were in a room like this with 100 customers, most of them were data center customers. And I said, "How many of you think we should get into the campus?" And 30% of them raised their hand and said, "We are tired of all this other software, spaghetti OSs. We want the uniformity. We want what you brought to the data set into the enterprise." But I said, "What about the other 70% of you?" They go, oh, no, no, we don't want that because you'll get distracted and you won't give me a good experience in the data center or AI or whatever, right?

Fast forward 2 years later, they were like, "Where are your campus products? You're late." So there was a burning desire to bring the same stability, availability, reliability, analytics, automation, software-defined capabilities into the campus that we brought to the data center.

Over to you on the differentiators.

Yes. Of course...

You have a great CEO. She articulated her message so well.

We are in a world right now where the entire industry is focused on exactly how quickly they can execute security patches across their portfolio. And with every other vendor, they're looking at 5 or 6 or 8 different OSs, different packs of software that have to be patched and upgraded, et cetera. And Arista, it's one. And the...

Can I tell another story on that? So when the Glasswing project happened and there were 10 vendors invited in, little Arista wasn't in that 10 vendors. It was Microsoft, Anthropic, Google, JPMorgan, Bank of America. I'm sure, were in there. And everybody said, "Where is Arista?" Because they were expecting us because this vendor, again, has the least vulnerabilities and the most robust operating system to be there guiding them. So we didn't ask to be invited, we were pulled in literally to help solve an industry-wide problem. And that kind of tells our role in this, the silent network, I guess.

And it's a moment in time when every IT shop is staring down this enormous challenge with other vendors, 5, 10 different versions. At Arista, it's one. And that's enormously powerful, demonstrating the value of quality and reliability and not just the availability of their apps and their client connectivity, but how we got there through this diligence, year after year, decades after decades of sticking to this core strategy.

The second piece beyond just reliability is looking at sort of the future of how AI and agents are going to manage these networks. And EOS has incredibly, incredibly powerful data platform. In fact, it turns the whole network into a single observability fabric with complete state streaming, every single stage change across the entire network being fed into our cloud, which means that our AI agents are built on the best possible platform. And that is already being seen by users. And that differentiation heads, years and decades of future ahead of it. And it's really built on the shoulders of giants. It's the state streaming, it's the analytics that are used by hyperscalers that they are used by the biggest data centers in the world being brought into the campus space is awesome, and the response is already great.

So a question for both that was referenced on the last conference call, you spoke about decommits of supply chains and...

It's a word that haunts me now. I actually went and looked it up with the dictionary to make sure I got it right. And just for terminology, when Todd is running this worldwide manufacturing organization, and he puts in a request with silicon suppliers optic CPU memory, often 52 weeks ahead of schedule. This is why you see our purchase commitments now at $8.9 billion. There's usually a promise date they give us, just like we give a promise date to our customer. So the definition of a decommit or recommit is when that date moves. And so we've had a lot of dates move. And that's disconcerting because when their date moves, our date moves to our customers, right? And then it changes the lead time.

So nobody in the industry can tell you that they don't have this problem for if they do, they're an ostrich in the sand, and they're not admitting this a problem.

Second thing I would say is, this is a 2-year industry problem. I'm not saying it's a 2-year Arista problem. I'm saying it's a -- at least 2-year industry problem. And those that are stronger and can plan better and do things ahead of time are the winners. Obviously, we're going to not sit for 2 years and feel bad about the situation. We're actively acting on it. The memory situation was horrendous last November. We've gotten ahead of it. We are paying top dollar. It's going to affect some of our margins while we still stay in the band of 62% to 64% gross margin like we've advertised, but we dealt with it. So we will deal with it, and we will come out of it just like we did COVID, the lawsuit, or this one. But it's no fun for the industry and in my view, the strong will get stronger.

I'll just add, there's surging demand right now. And we see from our customer base upside and upside and upside on the forecast. And that means we have to push on the supply chain to be able to serve that, to be able to drive material needed. And this industry phenomenon. It's just how much upside can we deliver on? I mean it's putting pressure on it, for sure.

So 2 years ago, 1 year ago, we spoke about memory, memory supply constraints. What's the situation of supply constraints now? Did it expand to other areas? What other areas? Like talk about the environment and what is the gating factor for you to grow faster?

Let's just talk -- let's just get the time frame. That memory situation was like 4 months ago. Things are moving quick -- things are moving quickly. And it's going to be across the technology space. I mean we see these capital budgets coming down the pipe that are enormous. People are building out massive, massive data center capabilities around the world. It's not just memory. It's going to be in the wafers and silicon, not just in the high-end silicon. We see it in power ICs. We see compute and CPUs. We started to see surging demand on the CPU as well as the GPU side. A lot of these roads lead back to the same advanced-process fabs. In Taiwan, there's only really one vendor there. And so we are -- when roads lead back to TSMC, there's going to be capacity constraints.

And I think even there, the shortages are really not just at the wafer level, but at the substrate level. Because what's happening is the material shortages here is as you start to build more complex processes with chip-on-wafer then you got to build fab capacity for 2-nanometer, 3, 5, 7, but we still have all the processes that we need to address, right? So it's sort of an across-the-board material shortage that pops up like a game of whack-a-mole. First it's memory then it's -- like you said, optics, CPUs, PCBs, you name it. And holistically addressing this so that we can ship a product in time has been a challenge.

Having said that, I just want to be super clear on one thing, we're going to do everything to solve these problems in quarters, not in years. And any delays we have this year, doesn't mean our orders or our customers go away? It just means we have to address it next year.

Yes. Got it. One question, I want to understand it conceptually. Deferred revenues grew substantially in the last few years. I mean this is -- it's a reflection of the environment. What -- how does it evolve going forward? And I explained what I mean by that, I don't mean orders. I mean what happens when a big customer finishes to build a data center? What happens when -- because these are joint data centers. So what should we expect with deferred revenues to do like step-down and revenue jump when someone finishes a data center? Or is there a smooth kind of line to the deferred?

That's a really good question. And again, history is always an indicator of what might happen in the future. So I'll look at some history first. We had a cloud super cycle back in the 2016 through '18 where deferred wasn't as big, but neither was the CapEx, but it was easily 10% to 15% of our revenue because people were putting in Neocloud data centers, our products were new, they were going through 100-gig back then super cycle, sometimes it was a use case in the leaf, sometimes it was in spine, sometimes it is at the edge. So this went on for 3, 4 years.

And then right around 2018, '19, when COVID happened, all these use cases became more mature. And I mean, COVID just put a lot of things into standstill, but you actually saw a deferred not only go flat, but pretty close to 0 in that time frame. So I'm not saying we'll go close to 0 because this is a much larger number, and we -- there's more shock absorbers. But you can expect that deferred, which is in my view, all-time high right now, will start to slowly flatten as these use cases become mature as the products become mature, and there will be this gap between new AI centers and old AI centers. So I don't expect deferred to remain at elevated levels forever. Eventually, they'll come out as revenue. And as the cycle matures, it will be a new set of use cases, maybe in scale up or scale across. So I think we are at an all-time high now. And it could get higher or lower, but I think more likely, it will flatten.

Got it. Another question I had, which is very specific about a single customer, but you disclose it. You disclosed that Microsoft is your largest customer. Last year, they did not do Ethernet. They did -- I mean, large scale, right? They did mostly InfiniBand. But this year, they are. And they don't use white boxes. You are their vendor. The interesting part is when I look at your disclosure, they grew 26% without having you in the back end, in the front end, they grew 26%. So what happens this year, meaning, what happens when you start to see Ethernet deployment in the back end? And what happens to the front end? Does it -- is there a parallel growth in the front end as well?

Yes. This is a question you keep asking that I've never done full justice to answering, which is like, hey, if you're growing in AI, why is your core business not? And I think it's important to look at these things not across quarters or even a year, but across multiple years, right? So as you rightly pointed out, in that particular hyperscaler, but in general, a lot of upgrades were going on in the front end between '22 and '25 right? And -- but today, if the customer has a $1, they're not refreshing the front end as fast. Guess what they're doing, they're putting all the investments in the back end. Now that doesn't mean it will last forever because once they finish the back end, they're going to put pressure on the front, right? And you got to do another upgrade cycle there. So there's this sinusoidal wave where -- when the back end goes up, the front end doesn't so much. And when the front end goes up, the back end doesn't so much. In a particular use case. And you multiply that times the number of customers, you get the picture.

So when you see our front-end slowdown, it means they already built it or they're focused elsewhere. And by the way, there was a period of time where they were neither focused on back end or front end, they were just getting GPUs from NVIDIA, right? And that was the largest bill they had to pay for. So I think if you look at it over a period of time, it all evens out. But if you look at it at a point of time, I remember saying at one time that the front end today is 2:1 over the back. I have to correct that statement now because now it's looking more like 1:1 because our back end is growing so fast.

Got it. So the question is also there is a cycle, and I don't know how to refer to it, but there is also a cycle of technology at the semiconductor level. So when Broadcom releases a new chip, right, like Tomahawk 6, is it a tailwind or a headwind? Does it give opportunities for new players, white boxes? Or because you're so close to Broadcom, does it give you the tailwind to actually grow even faster? What happens upon release of new technology?

I think any new technology tends to drive a benefit to Arista specifically because Ethernet as a standard is becoming kind of de facto standard, especially scale-out, but in the rest of the market as well. And as Ethernet continues to demonstrate leadership in the space, then Arista becomes like the natural choice. We lead the standards. We drive sort of the state of the art around that. As we've seen this kind of rapid advance from 800G to 1.6T. We're starting to even look at 3.2T. Like you've got people looking at the capabilities that, that provides and really considering building out faster and faster data center. So it just opens up more and more TAM if anything, accelerates Arista growth.

And just to add to that, if you agree with Todd, I think the beauty of our business is, while we are going to enter this 1.6T refresh cycle, we still got an awful lot of 400G and 800G going on as well, right? It's not like they died and it's either or. So we're actually seeing the benefit of a double cycle where one speed is taking off at 400, 800 and then you're layering upon since I like desserts, a multi-layer cake, and the 1.6T comes on top of that. So it's definitely an accelerator to the overall speed refreshes we're going through.

I want to ask you about the part of the number that we don't talk about, and I'll explain it in a second. So we talk a lot about hyperscalers. We talk a lot about campus. If I remove these 2 from your numbers, there is a 40%, 50% in the middle. And that is enterprise data centers. Now I have the luxury of covering both software and networking. And in the last cycle that we have seen now, I'm seeing the early bird that enterprises are investing in AI cybersecurity. And that is the prerequisite for investing in the network because you have to invest first in cybersecurity. Palo Alto reported it yesterday, Fortinet before. You can see clearly, AI is a driver for cybersecurity.

The question I have is from your guidance, this part of the network doesn't grow that much, meaning less than historical, but that's part of your conservatism. So I will say, I will add to you in line with historical numbers, 5%, 7%. What happens in the future? What happens to enterprise data centers? What are the drivers? And you're greatly positioned there. Is there any other -- like is there a market share discussion even? Or it's not about the market share, it's about market growth?

Jayshree mentioned this, cyclical nature of the spend. So as people start to shift maybe big focus on to AI than the traditional data center might slow that year and then back. But there's enormous growth opportunity here in 2 ways. Number one is we continue to see ourselves picking up significant use cases, specific important franchises across existing customer base in the enterprise. No doubt about that. But there are parts of this market that we have yet to reach. Arista is the market leader here, but we still have an enormous amount of TAM that we can go after, and we're seeing growth internationally. We're starting to see big university wins. We're continuing to pursue the big kind of traditional financial wins, but the verticals that we are starting to penetrate in real ways are amazing. Manufacturing, retail, university, et cetera. And these are big data center build-outs that are important. And so this kind of pursuit of the enterprise data center continues. And I know it's easy for it to get overlooked, but there is enormous opportunity. We're still reaching the rest of the market.

We're gaining enterprise logos. We're definitely gaining enterprise market share. And as you already know, we're the #1 in high-performance switching. But I think we have to be patient about it. It doesn't happen as fast. We don't get the instant gratification we do in the cloud and AI cycle because think about it, an enterprise data center is building and then it's usually a depreciation or a usage of that data center for at least 7 years, 8 years, 9 years, even if they put the wrong technology and we got to wait, right? So we do best in greenfield situations. We do pretty well in brownfield situations. We don't do as well in legacy risk-off situations where the enterprise customers just going to sit there with their assets and spread it out for a while. And you see all 3. But I think over the long term, and as you know, Arista has been built with our software and hardware excellence brick by brick, this brick shall come, too. It's just not the first use case, it will come a little bit later.

Got it. I want to talk about competition. So we -- 10 years ago, we spoke about competition with Cisco. We don't talk about it anymore because there's nothing to say. But the question is competition with NVIDIA because we do see NVIDIA growing in networking. Can you articulate how you compete with NVIDIA?

Well, I think it's one of those classic co-opetition issues where we are very grateful to NVIDIA because our AI business is almost always tying in to, 80% of the time anyway, into NVIDIA GPUs. So there's no way they're a competitor to us in the compute space. The more compute capacity they deploy, the more we can do with networking. But obviously, this tendency to build a vertical stack with NVIDIA, and there have been many examples of customers where the customers want to go with a horizontal best-of-breed network from Arista, but end up going with NVIDIA because of the customer gets a better bundled strategy, better support, better availability of the GPUs, et cetera. So there's a natural part of the TAM that's just not available to us because it's an NVIDIA vertical TAM.

Now having said that, again, we got to be patient here because I think today, it's largely NVIDIA GPUs. But we're very excited about the other AI accelerators that are coming on, both in training and inference. It hasn't gone unnoticed that we love the TPUs and the Google architecture that started way back in 2017. And naturally, we have a lot of opportunities to connect directly or indirectly to it. AMD and the MI Series is, again, we have deployed a lot of successful networks already together and every customer is looking for an alternate accelerator strategy.

There's a whole range of inference accelerators that are coming out. You guys saw the Cerebras IPO, but many of them are being built in-house by the hyperscalers, which is, again, giving us a lot of rack opportunity. So we love NVIDIA when we connect to their GPUs, but we love all the accelerators and frontier models to give us that diverse one network for all the diverse XPUs, if you will.

The one thing I might add there is there's also a story of building standards-based networking. Networking time after time after time has been an area where open standards have won the day eventually. And we're seeing that across the board. NVIDIA has some Ethernet technology available as well, but Arista leads the way. We lead the way in the standards. We lead the way, pushing the limits of the technology. And in the fullness of time, I think that becomes a driver for us.

Right. If I look -- if I start to try to divide the market between GPUs of AMD or others and between GPUs of NVIDIA, do you have -- do you think you have higher market share when the GPUs are not NVIDIA, when it's something else. And the reason why I'm asking it is because that part of the market is actually growing very fast.

Yes. The short answer to that is yes, because there is no vertical bundling going on. There is a best-of-breed approach over there. Arista is naturally the first and often only choice. And there's so much risk in this deployment, the last thing they want to do is go explore some speculative things. So yes, it is the fastest-growing market for us as well.

It's also when we get off of the NVIDIA GPU then the scale up part of the market becomes available, right? And that is a whole piece of the AI space we don't play in.

Okay. My last question, I don't leave much time for Q&A, but I wanted to cover it: optical integration. Your competitor, Cisco, does have an optical capabilities in-house. What is your philosophy, first of all, about optical integration, owning assets versus partnering with assets. And take it from there.

Look, our philosophy is definitely to work with the best-of-breed optical ecosystem. We don't have to own all pieces of it. There's a lot of physics there, as you know, and there's a lot of technology with lasers and co-packaged optics and Andy and the team introduced our XPO, which is the latest and greatest for pluggable optics for 1.6, 3.2T. So we are big fans of optics because wherever we sit, we have to connect to them, right? And it wouldn't come as a surprise to you that we worked very closely with Acacia before Cisco acquired it and continue to. There are good coherent optics vendor for sure, we see them.

Having said that, I think Arista's position is not always to be the revenue provider or the product provider for optics, but integrate into the optics really well, which can be integrated into our boards in some cases or work with pluggable optics and provide the right MACsec and root cause analysis works for them, including managing, all of that capabilities. But because that is such a specialized best-of-breed type of companies, just like security, we don't pretend to be that, but we want to enable that at a system level.

Great. Sorry, I took most of your time. But is there any question from the audience? We have a microphone. No.

Well, you took all their questions, looks like. Thanks for coming for the first early morning session here. I appreciate it.

Thank you, Jayshree. And thanks, Todd.

Thank you.

Thank you.

Thank you.

My name is Sebastien Naji. I'm the research analyst here at William Blair, who covers Arista Networks. I am required to inform you that for a complete list of research disclosures or potential conflicts of interest, please visit our website at williamblair.com.

I'm very happy to have Chantelle, the CFO of Arista, as well as Rudy, VP of Investor Advocacy, with us here today. And I think Chantelle will begin with a quick presentation overview of the company, and then we'll jump into some questions.

Sure. Are we good to go? Great. Okay, so good afternoon. Thank you for spending your lunch half hour with us. We definitely appreciate it. It's very nice to see you. For those who haven't met, I'm Chantelle Breithaupt, CFO of Arista.

And so just a couple of slides to orient. Many of you know our story, but just in case you haven't seen us in a while or are new, I just wanted to give you an overview. So here, we're 12 years past IPO, and here we are guiding $11.5 billion this year, incredible journey from a technology perspective. Originally started with the high performance compute getting into hyperscalers, cloud, enterprise data center and now we're talking all things that are AI. An incredible journey being $200 billion market cap, 12 years post IPO, and we feel we're just getting started. So thanks for being here with us.

Very happy to show different sections in the sense of where we're recognized by Gartner. So you can see from left to right, some of the Gartner positioning in the top right quadrant. 2025 for data center switching, 2026, just happy to see this here that we're now there for enterprise wired and wireless. Very happy to see that just getting into the campus and enterprise wired and wireless LAN. And 2024 with the acquisition of VeloCloud also being in the Magic Quadrant for SD-WAN. So I feel like we're making ground in a lot of different aspects of networking and very proud to have these positions to show you.

Why do over 10,000 customers choose Arista? They can see the robustness of our quality and our product and our ability to serve the customer extensible performance and platform. We don't like vendor lock-in. We like best-of-breed choices and are happy to compete in that environment. And then just having this really great real-time telemetry and AIOps that we will get into a little bit through the Q&A, I think, with Sebastien. That really makes us a great total cost of ownership choice when you're thinking about getting the highest utilizations from your XPUs. And it's not just how we feel this is the customer representation. So I think you'll find very few in the industry who can say they have an 89% score on NPS. And you can read some of the comments on the bottom, how they interact with our company.

Very delighted in the response time, very delighted in the technical capability of the team they interact with and just the approach that we're there until the end. Until they get to a performance and outcome they are looking for. So very proud to have that and congratulations to Ashwin and his team for getting the company to that position.

We talked at Analyst Day in September of last year. Hopefully, some of you were there about the growth in our TAM. I started in January 2024. Just before that, we were at about a $60 billion TAM, then $70 billion at the New York event we had in June 2024 and then Analyst Day updated that from $70 billion to $105 billion, so some pretty great growth there. And you can see at the bottom the different categories. Of course, a lot of it is driven by AI and data center, but also some great aspects that we see now that we have our campus portfolio underway. We're working on the next revision of this. And so stay tuned as we look at even a larger TAM amount in the different categories and the nomenclature that we'll use going forward. We also like to position our incredible platform and portfolio over what we call centers of data.

So we have the AI center, campus center, data center and the WAN center -- SD-WAN. And it's just to show the breadth of optionality that we provide to our customers. It's really we have a portfolio of products in what use case are they being used for across the different segments of our customers. So very proud to have this to get to their outcomes and their cost of ownership that they're looking for. And now we have this new nomenclature. We're talking about scale up, scale out, scale across, and I know we'll talk about that, Sebastien. But just to say, we're ready for this. We've been planning for this. We'll be working on it and very well positioned, I think, with our customers on this kind of co-engineering journey. What are they looking for from the largest hyperscaler and AI customer down to the enterprise.

From a revenue growth perspective, I think it's important to look at the growth that we've seen. We finished Q1 results in the sense of 35% growth. And if you look at the combined growth in the P&L 35% plus what it goes into deferred, which means it's shipped and invoiced and cash collected. You could say, basically, it was 54% growth. And you can see the subsequent quarters beforehand in the sense of where we've seen the growth combined between P&L and deferred revenue growth. And just to remind you of our guidance for this year, we did call out our guidance again, looking at 28% growth, $11.5 billion, maintained gross margin, which we're very proud to do in this kind of environment where it's a little tougher from a supply chain perspective. World-class, I would say, operating margin for a company like ours of 46%. We have given ourselves accountability goals when it comes to AI and campus. And you can see the growth in AI and campus goals versus our last year performance, pretty healthy growth for both of those sectors.

If you think about $3.5 billion of AI at $11.5 billion, a high percentage of our revenue is coming from AI. And then we've raised our CAGR for '25 to '28 to be 20-plus percent growth before that was mid-teens. So hopefully, you'll see these kind of signs of the demand we're seeing and how we're showing them through our guidance. And I think now we can go to the questions, Sebastien. Okay. Thank you.

Great. Thank you, Chantelle. I mean, maybe my first question, I wanted to start off very broadly. Arista really made its name over the last 10 years, filling this need for high-performance networking at the hyperscalers. Today, as we enter the AI era, you really hear from all the networking vendors about how strong their businesses have become. Cisco, NVIDIA, they're all talking about huge amounts of growth. So as we enter this AI era, how would you talk about Arista's maybe right to win or how it really differentiates itself versus all these other networking providers that are also talking about really good momentum from AI.

Yes. I think -- so thank you for the question because I think it's important to look at apples-to-apples in this conversation. But my leading answer in this is that we're -- I think Arista is very well positioned for AI. We've been waiting our whole 12 years post IPO for this moment. So I think all of the innovation we've seen up until now and the way we interact with the market leads us to be very well suited. And I'll give you some examples and Rudy, obviously, if you have anything, you can chime in. So let me start first with networking for AI, and then I'll talk about AI for networking.

So networking for AI, if we look at our product portfolio, some of the things I just showed here in the presentation, we announced the 800-gig Etherlink portfolio in June 2025. And so we've had that for coming up to a year, and we were very happy to see that AI was a great use case for 800 gig, and we were ready. And so I think if you take that 800 gig kind of Etherlink portfolio, combined with EOS, which is -- which was an operating system and software, which allows great utilization of the XPUs and all the performance criteria that we're looking for. And you look at the sense and the fact of all the experience we have with some of the largest, most complicated deployments.

There's a reason that we're currently the #1 branded vendor for front-end and back-end AI networking. And I think as long as we continue to innovate in the ecosystem, so we've been leaders in open standards when it comes to UEC, the Ultra Ethernet Consortium. We've been working on. You saw Andy announce XPO as a technology and working on ESUN, so we're working with the ecosystem there. And I think probably most importantly, combined with those things is that we offer networking that allows flexibility.

We're agnostic when it comes to the NIC, we're agnostic when it comes to the XPU, we're agnostic when it comes to the optics choice. And so that allows -- because every customer wants to do it differently to reach their goal, and we're happy to do things differently with each customer. And I think that will continue to innovate. And if we keep that pace, we absolutely, I think, have a right to win if we execute with the customer. So that's networking for AI.

AI for networking. Now we're talking about all the things we're doing to help -- AI help networking, and we have this great suite of products called AIOps. It includes AVA, which is our virtual assistant. We have agents and we have insights that we provide that sit on top of NetDL. And so when you combine all of these things, I think we are the best positioned AI pure-play networking company when you're talking about the networking data center part of it.

All right. Great. That's helpful. And then I want to maybe ask about what I think a lot of investors are curious to talk about, which is in your most recent earnings call, you started to talk about more acute supply tightness, having to manage very longer lead times. I mean, could you maybe talk a little bit about how you are managing that supply and whether you see lack of supply potentially being a headwind to your growth, either near term or even over the next couple of years?

Yes. I think it's a good question. I think you're starting to see more companies at least speak a bit more openly about -- the supply chain has some constraints, but it has no constraint in the sense of us reaching our FY '26 guidance and we'll continue to take quarter-by-quarter what that could mean for future guidance calls as we go through this year and next year. I think from the perspective of looking at the different arrangements we have. So we deal with some of the biggest suppliers and have great arrangements and histories with -- history with them, along with working with our largest customers. And so I think we're well positioned in the sense of when it comes to availability, I consider it more about when, not if, and so if you think about a lot of our customer conversations, our 12 months, excuse me, in design, 12 to 18 months into design, the 52-week lead time we have on average, I think we'll serve that well. You've seen our purchase commitments go up pretty substantially to lean into the next 52 weeks coming into next year.

So I think we're managing it well. I think where we're talking about the constraint. The constraint could come in if someone's asking to have something not within the lead time that's material. How do we work within the ecosystem to have that kind of burst demand. I think that's the part we're just acknowledging it's not as easy as it was but we'll do everything we can with our arrangements to have that. We're seeing incredible demand. We just want to make sure people understand within the lead times, that's okay, but perhaps between a burst of demand. We just need to think a little differently how we can serve that.

And that supply tightness, I mean, is it really broad-based? Or is it more acute in chips and wafers and memory...

It's a bit of Whack-a-Mole. It's a bit of a Whack-a-Mole. I think that memory was the first one, and I think we started talking about November last year, and I think we've gotten to a good spot on memory. I think other components come and go. I think that from a perspective of -- it's not just one vendor or one supplier or one product. It really depends on the time and what's happening with a specific deployment. And so I think it's not bespoke, but it's not generalized. So I'd hate to generalize too much across one category. It really depends on timing and innovation and what our customers are looking to deploy. So from that perspective. Now on the other hand, we've actually reduced campus lead times. So our campus lead times are now down to 6 weeks, which I think absolutely gives us an advantage as we start to go after the campus market share.

Got it. And then I think the other area that investors have a lot of interest is in this product deferred balance that has really continued to grow over the last 12 to 18 months. And my estimate is that it's around $3.5 billion, $3.6 billion, which is way larger than it's historically been. Could you maybe unpack a few things first. Is that mostly driven by some of these AI products, these new SKUs. And then you've talked in the past about you have to hit an acceptance criteria for that to flow into revenue. What does that acceptance criteria entail? Is it the networking hardware is plugged in? Is it when the GPU rack has a certain utilization? Maybe just help us understand how we should think about that deferred product starting to flow through into the income statement?

Yes. I think that -- so absolutely, so the concept and construct of deferred has been with Arista since basically we started doing cloud deliveries. And so -- but what you're seeing now is it's the same construct, just much bigger market and deals, right? So during cloud -- and we have some information in our earnings deck, if you go to our Investor Relations website, but you see there, there was like a cloud cycle, and now we're in this AI cycle and the AI cycle definitely surpasses from just pure dollars, the cloud cycle during that time frame. So same construct, bigger dollars flowing through and more complicated. So it is new use cases. It is new products and sometimes new customers. And so from that perspective, things are arranged with the customer where we have to meet certain criteria, acceptance criteria.

So it means it's shipped and it's invoiced and cash is collected, but it is in deferred until those milestones are reached. Those milestones can take 12 to 18 months, sometimes a little bit longer for the most complicated largest deployments. And it could be a variety of things. It can be by site, it can be certain engineering things that we've promised to co-work with the customer. It could be that, from their perspective, a certain design element has been reached. And so there are different -- again, back to every customer does it differently, but the premise is the same until we're sure that it's up and running the way we have agreed to, that's when we take the revenue.

So it is mostly AI that's in deferred right now because that's the largest use case we're seeing. And it's a great example of demand, but customers also enjoy that we have skin in the game until they're seeing great performance outcomes.

And then I think you started to talk about -- historically, you've had 2 major cloud titan customers M&M as you guys like to refer to Microsoft and Meta. More recently, you talked about expanding that to 3 customers with 10% plus, maybe even 4 customers over time. What is driving that increased diversification and some of that traction you're seeing with hyperscalers that maybe historically haven't worked as much with Arista?

Yes. I think that we're absolutely -- as we get older as a company or further in our growth cycle as a company, we are looking to diversify. It just takes some time, right? So we have customers A and B now that we call them, we're using a letters to denote now that we expect more of them to come in. I think that -- so there's a few things that I can drive at. One is, AI is a great use case. But generally, anywhere that we're switching to Ethernet networking being an option, that's where we're going to start to see where we can have some more business that we didn't have before.

Now obviously, to get to this greater than 10% customer at the -- I'm sorry, I can't see you guys, the range we're talking about, they're not coming in with just one deal and becoming over 10% customers. So they have some existing business, and they're getting to over that 10%. And so it can be companies that are using Ethernet that didn't use it before. It could be new kinds of customers coming to market in some of these segments you've talked about, maybe the new cloud sovereign AI, these sorts of new customers that are coming in. So we'll wait and see where we report at the end of the year. But we're fairly confident that third one will come in. The fourth one, as we mentioned on the call with Jayshree, we'll have to see exactly where that lands.

Got it. Great. And I think I'd be remiss if I didn't ask you a question about white box.

Sure, yes.

So we'll ask you a question on what are you seeing in terms of the competitive dynamics there? I think over the last year, Arista has done a good job of kind of laying out your blue box strategy and your differentiation there. But given that a lot of these white box vendors are reporting really, really strong results. Are you seeing any increased traction at your customers for white box? Are there certain sockets where white box makes sense and others where you have to go with Arista?

Yes. I think -- I don't think anything has changed in the dynamic in the sense of which use case is a white box use case versus a branded vendor. I don't think that has changed at all. And so that remains the same, and that's a boring answer to that question. I think -- so that remains the same. I think what you're seeing is potentially when you're talking about these growth vectors Sebastien, it's okay. At a point in time, you ship, you drop it on a dock and you're done. Okay. Well, that's recognized right away. So that feels good in that moment because all that's coming through at that point in time. So there's probably a difference in the sense of the timing deployment and how that comes through because you have to also take into account our model, which has the deferred. But we're not seeing anything that changes in the use case. It's the same -- and we appreciate the dynamic.

And so I think that we understand there's a spot for white box in the deployments, there's a spot for branded vendor. And sometimes that choice for very select customers because you have to be able to have the team to support a NOS on a white box environment or a blue box environment. So you're not talking smaller companies that can usually do this because then you're talking a CapEx versus OpEx trade-off, it's usually not for free, right? And so from that perspective, we find it's very few customers that can even attempt to get into that scenario from a just a cost ROI perspective. So we don't think anything has changed. We don't think there's anything moving differently. If anything, we would probably double down that it's becoming more complex and the utilization of the XPUs is becoming more important and critical, and that would be in our favor.

And maybe staying on the theme of competition, NVIDIA has really emerged over the last couple of years as a real networking competitor. They talked, I think, on the last earnings call about a tripling of their Ethernet business. How do you go about competing with someone like NVIDIA, particularly given that they have an ability to potentially bundle their networking with their compute and all the other parts of the stack that they're selling. And where do you really see that you differentiate when you're talking to customers versus the NVIDIA spectrum portfolio?

Yes. So first, just so we are completely thankful to NVIDIA for basically opening up this GPU market, and so we're very thankful because we participate in connecting those GPUs. So it's not all just competitive relationship between us. There is a partnership relationship that way. But if you get down just to networking, the type of networking we do, we have to be careful we're looking at apples-to-apples, and then I'll get to the competitive differentiation.

So Ethernet growth can also have NICs, it can have NVLink, it can have other parts that aren't part of what Arista would compete in. So I would encourage everyone to look at the apples-to-apples comparison growth and see what that growth is. And I don't think it's the same growth number trajectory you were just mentioning. And I think that aside, so how can we compete because we can speak to what Arista can do. I think, again, with that phenomenal portfolio optionality, what we can do is compete on the options you have to have to build best-of-breed, the agnostic capability to bring in other XPUs, et cetera, and the best-of-breed usually is a good choice for our customers. They want to have the optionality.

So if they're open to that, we have a good chance of winning. I think that in the sense of if you look at the other factors we talked about, things like EOS and AIOps, those are also things that they're delighted to have. And sometimes we see customers get into a vertical stack and then come back later realizing they want to go best-of-breed. So maybe it's not the first time, but even the second time that we see them. So there's lots of options where we can see them and come back to them. But those are some of these scenarios where we absolutely can win for the networking part.

Yes, that's really helpful. And then I wanted to ask maybe about a longer-term trend, which is this shift to co-packaged optics that everyone is starting to talk more and more about. And then I think relatedly, Arista right before the optical conference in March, introduced this concept of an XPO pluggable cable. Could you maybe just talk about across both co-packaged and the XPO cable? What exactly is Arista's opportunity? Is there a way for you to monetize these trends? Is any of this a threat, particularly the CPO aspect.

Rudy, do you want to take this?

Yes, sure. So in general, optics are not a big part of our revenue, right? So we certify and kind of resell optics, but it's not a big part of our revenue. So really, our work in optics has really been about moving the industry forward. If you go back 10 years, we introduced OSFP, which is kind of getting to the end of its usefulness, if you will, right? And so we were really ready for a next generation. Now when you get to that next generation, you certainly have one option, which is going down the co-packaged route. Or can you extend pluggables. And so what XPO allows you to do is extend pluggables and comes with the benefits that when it's pluggable, it's a lot easier for customers to service them in the field. It also uses the same supply chain that OSFP has used for the last 10-plus years, right? So there's significant advantages there. And one thing that I think sometimes goes unnoticed with XPO is it shrinks the physical size of the networking stack, right, by reducing that front panel density or I guess, increasing the density and therefore, shrinking the size.

What that does is it opens up a whole bunch of other options for doing things like scale up, right, where you can do things like copper for longer, where you can do things like micro LEDs or RF. So frankly, solutions that are more efficient from a power perspective, from a dollar perspective, then going down a purely optical front. Now perhaps we get to a point where co-packaged optics are inevitable. We are absolutely agnostic to it, right? Like we don't believe that co-packaged optics is a negative for us, we -- if anything, the engineering involved with co-packaged optics and making it successful, it's going to be super important. But I think what we're hearing from customers is, a, delay co-packaged as long as possible. That's where I think XPO comes in. And when you do go down the road of co-packaged make it as open as possible because what customers don't want to do is be locked into a single supplier, a single vendor.

Because today, they're not buying their optics from their switching maker. They're not buying their optics from even the chip maker. They're going out to Asia and buying the optics directly from the supplier. And that's given them tremendous leverage. They don't want to be stuck in a place where they're having to deal with margin stacking and things of that nature. There's both tactical reasons and financial reasons why I think we're trying to delay as an industry CPO for as long as possible. And that's kind of what we hear from customers. But when CPO is ready and necessary, we feel very, very well equipped to be there.

Got it. And is it right to think that a CPO chip just becomes part of your bill of materials that you build around essentially?

Yes, pretty much, yes. With -- again, like I think the bar for failure in the CPO switch is far higher, right? Because if a switch fails with CPO, essentially the switch has to be taken down. Today, with a pluggable, if one optic fails, you take that one optic out of commission, which means whatever workload is plugged into that, not the entire switch. So if anything, I think the hardware design, the reliability that we've done so well with over the last 20 years is going to become even more paramount. So we feel very well equipped for that. But customers do want us to try and delay that inevitability, if you will as long as possible.

Right, right. And then maybe on the theme of maybe longer-term risks or threats, Google uses a lot of optical switching inside of their TPU racks. What -- are you concerned at all that you could see more and more hyperscalers move to optical switching? And if so, does that present a risk to the packet level switching that you guys are so good at?

I'll take that one as well. I mean, I think what we've seen is optical switching has a very specific niche use case, right? Google, as you pointed out, has been the most -- the biggest proponent of it. But if you've seen some of the recent announcements from Google, I think even they're admitting that once you go beyond a certain scale, really go beyond a scale-up domain where you're dealing with a fairly uniform kind of workloads, right? You have to go Ethernet because that per packet switching capability, the ability to decide, okay, where does this packet go as you look at each packet and look at the header is tremendously useful, right? With optical switching, what happens, you're literally pointing the mirror in a different direction, right, as you want to move stuff around.

So it doesn't really work well for any kind of real-time decision-making on, okay, what this package should go there. So as workloads get more heterogeneous, as workloads get more -- have higher entropy, really, over time, Ethernet has always won, right? So we don't feel like that's a threat. And like I said, the recent Virgo announcements are, if anything, a testament that while OCS has a space in the network, it's more confined to these narrower use cases. And once you want to go beyond a certain scale and beyond certain types of workloads, really, I think, Ethernet won that battle time and time again over the last 50 years.

Sort of simple optical path, simple connections...

Correct. Point-to-point where you're not expecting to change things dynamically on the fly.

Yes. Okay. And then maybe here in the last few minutes, I could turn the questions a little bit to the enterprise business.

Sure.

You guys closed this acquisition of VeloCloud, which is an SD-WAN technology about a year ago. Could you maybe just give us an update on where you are with that integration? And if you started to see any success in cross-selling either the VeloCloud customers to more Arista products or vice versa?

Yes, we are -- I am personally very happy on how VeloCloud has come along. I think that it was the largest acquisition we had done as a company and a lot of great work went in to making that happen and the integration has gone well. We are functionally integrated. So everyone's kind of gone and joined the teams that they recognize functionally. But most importantly, we've had great conversations with the customers. I think the customers are very excited that now it's part of the Arista family. It's part of -- in the sense of the care and attention we can give it, given we're 5,000 people who focus on the portfolio and networking. So I think the customers have received it well. They've given us great feedback in the things they love us to focus on, and we have a team dedicated to working through those. So I think it's gone as well as expected and very happy to have the team there, a very talented team, and they've stayed with us.

So I think that's a testament that they're enjoying their time and happy to be part of what we do as a company and our culture. For cross-selling, I think that given it's just coming up to a year, I think it's pretty good progress. We've gone through all the functional integration of everything from quote to cash and getting all the renewals and licenses sorted out. So that was a lot of just transactional work. Now the real customers of the cross-selling are happening. We're seeing it in a few of the enterprise segment, and it's both ways. It's VeloCloud into existing Arista customers and Arista into customers we did not have before. So it's good customer acquisition. But it's still early days, but definitely a pipeline is building. And I would say that's on track with what our expectations were from a deal perspective.

Got it, right.

Two things I would maybe add there as well, right? One is, it has opened up a new route to market through the service providers because they had a pretty robust service provider business as well. And then the other thing is from a campus RFP perspective, often SD-WAN is becoming a checkbox requirement, right? And so having that whole portfolio allows us to bid on campus. So even if it's not a direct upsell, cross-sell, it allows us to bid on a project that maybe a year back, we would have been disqualified because we didn't have an SD-WAN solution.

That's great. Okay. And then maybe just last question, again, on the enterprise space. I think last year, you had talked about the pending Juniper, HPE acquisition as potentially driving more leads, more interest from customers. That acquisition closed about a year ago. Frankly, it sounds like it's going quite smoothly for the company. I'm wondering, is that still an opportunity for Arista to take share from that customer base? Or has that maybe become less of an opportunity?

Yes. I think that from our position, we're starting at 3% market share in campus. So we're super excited to the upside. And we're also excited that there's a great refresh cycle coming up in the next few years and perhaps there's a pull in of the refresh cycle as campuses get ready for AI and their AI story. So there's a lot of goodness there and we have the portfolio to serve. I think at 3% market share and the overall market is growing maybe high single digits, we're -- our growth is going to be share-taking. And so I won't comment on specific who we're taking it from, but I can say that if -- our road map is clear, and we're winning campus first deals, and I think the more we can rinse and repeat that and have them as references, I think our share taking from whoever the incumbent is what we'll be looking to do.

Right. So we only have a few minutes -- seconds left. So maybe just if you can give us your updated view on capital allocation. I mean, that's one of the benefits of Arista as you guys have great margins and very high free cash flows and they're only going up. You've made some acquisitions, but not huge ones historically. Could you maybe just talk about how you think about M&A versus potentially dividend, potentially more share buybacks?

Yes, sure. So our capital allocation, just in a nutshell, remains the same. Organic investment, marketable securities because we're still getting very strong interest income in this environment. And then it would be share repurchases and then M&A. When it comes to M&A to answer your question, nothing obvious. We don't want to break the culture, and we don't want to go into something in new adjacency. But if we see something more maybe in the AI space, either talent or tech, that's probably where we'd be looking.

Okay. Great. Thank you very much.

Thank you for your time, thank you.

Thank you all for coming. We're going to have a breakout in -- Anna, where is it? [ Mar ]? In [ Mar ].

Hi. Good afternoon, everyone. Welcome to the fireside chat here with Arista Networks. I have the pleasure of hosting John McCool, who is the Senior Vice President of Platform Hardware at Arista; Tyson Lamoreaux, I hope that I got that right, SVP of Cloud and AI Networking; and then Rod Hall, who's probably not new to a lot of people, but Investor Relations and Finance at Arista.

Yes. Thanks, Samik. Yes, it's great to be back. Thank you for having us and organizing everything. It's really good to be here. And I'm going to kick off with a couple of points, and then I'm going to get -- we're going to get to your questions. But we -- two high-level themes that I want to flag to people and then I want to make a couple of points.

High level, there's 2 things you need to take away from this. One, Arista is extremely good at managing the supply chain. We -- if you go back to COVID and you don't have that -- maybe some of you don't have that history with the company, we went into COVID and ended up doing better than most others in our area in managing supply. And so that's why we've got John McCool here with us because he's the key guy that ran all of that, and we wanted to give you all a chance to hear from him and talk to him.

The second thing we want to leave you with is that our products are, from a value point of view, absolutely incredible. And we are doing extremely well from a competitive point of view, and that's why Tyson is here because he's Head of our AI and Cloud Networking Group and can, I think, articulate the value of those products really well. So those are the 2 really high-level things we want to say.

Low-level, we know there's going to be a lot of questions on supply. Samik has got a lot of questions for us as he's going to ask, but we are talking about an industry-wide supply problem. We don't believe this is specific to Arista. We've been dealing with supply issues for quite a long time, and we've called out long lead times for some of the components for several quarters now and that continues.

What we don't believe is this is necessarily changing anything from a revenue point of view for us. We already are in a situation where we can't ship as much product as we have demand for, and that's been the case for quite a while. But you will have seen in our recent earnings deck that we put out after earnings, we updated it any way, we're talking about a 20% CAGR for 3 years for our revenue, 20% plus. And that's meant to give you some idea of the confidence we've got in our ability to deliver revenue even in the midst of this difficult supply situation.

The second thing I want to say to you is that our margins remain stable. We have kept our margin guidance unchanged at 62% to 64% for the gross margins for this year. And we do that for a reason because we can demand value for these products. We've already put one pricing change through. We hope not to do any more. We want to be good partners for those customers that we work with, all of our customers. But at the same time, we're a company that has been able to demand value for its products. And I hope, as you hear from Tyson here, you'll get a better feeling for that. That's all I got to say. Over to you, Samik.

Thank you. Thanks. So that's a good overview. Now let's dive deeper, start with supply. We'll come to the cloud companies and the demand drivers in a bit. John, what are you dealing with in terms of biggest constraints on supply? And to Rod's point, how do you see the supply situation now, similar or different to what you had after COVID?

Yes. So different to COVID. So let me just start with supply. What we really see is a demand-supply imbalance, right? And Jayshree talked about on the call demand being greater than anything that she's seen in her tenure. So with that demand, capacity for advanced semiconductors have been put in place 2 years ago for this year that we're facing. As these folks are building out these big AI data centers, they really have 3 constraints. One is energy to fuel those data centers, the physical construction of the data centers and advanced components, whether that be 5-nanometer, 3-nanometer process nodes, advanced memory technology, CoWoS packaging and packaging test, DSPs for fiber optic components. So that's what we've been up against.

If you take a look at our shipments, if you take the revenue that we had in Q1, plus the deferred revenue, which actually represents physical shipments that we made from our factory, we were able to overcome those supply constraints and grew 54%. But we talked about on the call that we are living in a supply-constrained environment, and it is around those advanced processing nodes.

Any change in prioritization from your vendors in supply chain to Arista because that's been a big topic?

Absolutely. Yes. And just kind of back to the question about COVID and maybe what's different. In COVID, we started out really with a supply constraint. There were no people to build anything. I mean everything stopped. And that started to shift again to a demand profile later in. And if you think about it, that demand, part of it was IT. We were all working at home. There were a lot of computer and desktop upgrades, but we were also buying new cars, we were doing DIY projects and needed drills that had 10 FETs in them. So the things that were short were things that were broadly based across all electronics, whereas this is singularly focused around the components that need to go in AI.

In terms of prioritization, look, I think we're well represented by our suppliers who have an enormous seat at the table at TSMC as well as our customers. What we saw during COVID is our suppliers that needed to build out advanced networks to support their capability really represented us well. We haven't had to invoke that in terms of this current supply constraint, but I think it's really well recognized that our customers will need networks to connect their GPUs, their CPUs and the underlying pieces that put together their AI infrastructure.

Got it. Got it. Okay. Maybe moving down that path a bit more, let me phrase one more question, and I think Rod touched upon it a little bit, but you've talked about being a good partner to your customers, not doing multiple or aggressive price increases. There was some indication on the call that you could be looking to sort of help customers out and there could be some hit on gross margins on that account. Just maybe address those concerns while also talking to the purchase commitments that you've done and how do those help customers? And do we eventually get to a point in the tightness of the supply chain where some of those aggressive price increases or multiple price increases just become inevitable for you?

Sure. We already saw that with memory. And coming back to the comment about gross margins, what we were saying on the call is we're really going to lean into the business because of the demand. So in Q4, we began to see that sharp increase in memory prices. We went out and secured supply. We made the price adjustments in Q1. So there was some disconnect between the timing of our consumption action against the price increase. And we're going to be aggressive in terms of getting supply to meet this demand and there may be some lag between being able to pass that on to customers. But we absolutely have the value to really bring those price increases forward if there is an underlying cost increase.

Our customers are quite sophisticated. We've been with them for a very long time. So we don't believe in really kind of leveraging this moment to increase prices. They're also very sophisticated to understand the underlying mechanisms that would drive a cost increase. And no one's happy about price increases, but they're open to the conversation and have supported us in the past and are supporting us now with the memory increases that we've made.

Okay. Rod, for you. The decision to upgrade your long-term target to 20% following the earnings call, maybe just walk us through the thought process there and what drove that?

Yes. Before I say that, I also want to -- John made a point earlier in some of our meetings, we could have increased our margins quite a bit more during COVID and we chose not to do that because of these long-term relationships. So I thought that was a really good point that you made. It's not a commodity company like memory where we just double our margins when we can grab up that margin. We have these long-term relationships. We want to be good partners. So that's something to think about, even though we did keep good margins through that period.

In terms of the growth CAGR, I mean, I guess what we wanted to do, first of all, mathematically, that number, the mid-teens number we put out back in August didn't make any sense anymore in the light of our 28% growth. So that was one thing we just thought well, gosh, this doesn't make sense mathematically. But secondly, we wanted people to understand that we have confidence in our ability to deliver revenue even in the midst of this situation. So we wanted to just put some highlighter on that, make sure people understand that.

And higher number would have been more helpful, I can tell you that.

Yes.

Yes. The other thing that I think we addressed in that update was around deferred revenue. And as we were having callbacks and discussions, a lot of people conflated to discussions some people are having around order growth with deferred revenue and deferred revenue is actually things that we shipped, we've invoiced, we built and represents the growth of our shipments. So that also is explained a little bit better in that deck.

Okay. So finally, Tyson. You built one of the largest cloud networks in the world before coming to Arista. When you look at now the network architectures that your customers from this side of the lens now, the customers that are designing today versus even a year before or 2 years ago, what surprises you most in terms of how quickly things are evolving with your customers?

Well, I mean I think it's the rate and volume of change like -- and what I mean by that on 2 dimensions, the rate of change is just the quick turnover to new architecture. Some of our customers are super ambitious and their goals, objectives are unafraid of putting multiple types of competing architectures into production simultaneously and unafraid to walk away from ones that they determine don't work, don't scale, don't meet their operational criteria. We have some others who are a little bit more methodical and iterative, but their iteration cycle has shrunk. So that rate of change has increased dramatically. We've even seen it in the optics. We've seen it in silicon. It used to be accepted that you would see a new switch every 3 years. And now we're talking about 18-month cycles on accelerators, 18-month cycles on switches. Those go into new architectures. Because the architectures are advancing in parallel to the switches being built, there's sometimes some dissimilarity in terms of what is intended to be done and what can be done.

And so you see some of this iteration also apply to taking what they can use today, like maybe a Tomahawk 5 switch and deploying it at smaller scale in an architecture that's intended for something along the lines of Tomahawk 6 or maybe it's going to go out and air cooled and plan to convert it to liquid. But it's energy, it's link speed and transmission rate, it's reliability, it's the radix, the total bandwidth of the switch. You're seeing a lot of stuff evolving around protocols now at a much faster pace as well. The argument around scheduled versus unscheduled fabrics, the argument around switch-assisted computation, the argument around leveraging new protocols like the MRC protocol to be able to do better multipathing using the network silicon to assist in this as opposed to just purely unscheduled fabric, but having a protocol-native mechanism for it.

So this happens at kind of every customer, but across several customers, and then you look at across our product set and this is where kind of the volume kicks in. It's a lot of parallel change across the entire product stack kind of at the architecture level, at the system level and it doesn't really show any signs of slowing down. I mean we're staring at 1.6T coming, 3.2T is right behind it, optical technologies are evolving into XPO and CPO are coming as well. And then when we look at the clusters, john talked about the constraints in getting data centers online either due to power or real estate or construction or construction-related materials, you start to see things push around distributed training or novel ways of interconnecting locations together to drive inference or a combination of inference and training traffic.

So I think that this is similar to the cloud in several respects, but so much more on a much more compressed cycle. And hard to see when we're going to reach the peak of it. I mean it still looks like a lot of runway ahead of us to keep going.

Okay. So maybe turning -- going more specific here. A lot of attention recently around one of the large cloud providers publicly describing and detailing a new scale-out accelerator fabric, purpose built with higher radix commercial switching in a flat 2-layer topology, right? It looks on paper like a great fit for Arista?. Just talk to us about how you interpret the specifics there, and whether Arista would be well positioned with that cloud company or not in terms of their new scale-out fabric?

Yes. Well, I think your assessment of it looks great on paper, I would agree with. I mean I think that like anything that resembles our Tier 2 leaf spine architecture, higher radix Ethernet, anything that's Ethernet plays very well to our heritage, to the products we've built, to the ongoing investments in R&D that we're driving around the products that we're going to be building in the future. So ignoring the specific customer, generally, we think this is a nice validation for what we've been saying all along, which is open standards, ecosystem, interoperability, Ethernet tends to win out in the end. Customers want choice. They want to be able to choose the best-in-breed suppliers to work with. They need partners who can be an extension of their engineering capabilities and can help them achieve bigger things faster. And when we're talking about this accelerating and compressed innovation cycle, a lot of architectural level, a lot of industry-wide evolution happening.

Supply constraints. You need companies and partners that can move at the level of kind of dynamic that you have and can impedance match. So I think we're very well positioned for this type of architecture in general. I think we're building products that slot into this really well. And I think our company, in terms of its product, its operations, its capabilities also make for a wonderful fit. So we're going to go fight to try to win every one of these Ethernet deals that we possibly can, that's for sure. And we like to be part of the story there at this customer and the others.

What's exciting to me about that is if you think about 8 quarters ago, at a conference like this, we'd be talking about this Ethernet even have a role in AI, and that was all InfiniBand. We talked about the back-end network and the potential TAM. We tracked 4 customers. And if you look at our guide now for 2026, 30% of the business will be AI, right? So that's really exciting. And I think the continued validation of Ethernet and the scale-out architectures continue to drive that growth.

Let me ask you a follow-up on that, Tyson. As much as Ethernet adoption broadly is stable for Arista, what are you seeing in the competitive landscape for higher radix switches? I mean, to us, on the analyst side, hieratic switches just seems to be a great fit for Arista, but are you seeing more competition at that level of capability that you particularly do well in?

Yes. There certainly are some competitors who are aggressively pushing in for denser, higher radix switches. Some of this is kind of what I would call kind of a mixed modality, which is kind of a multi-RU or multi-OU system with several chips in it, but they kind of operate as disparate independent planes or nodes versus something more sophisticated like our modular chassis, which is kind of the largest that you can get, but it operates as a single unit and substitutes optics and cables for traces, for communication efficiency and in the case of high bandwidth and high link speeds cost savings as well. I mean, it's -- copper is always going to be cheaper than optics. And so we see competitors understanding that there is market for this and entering, but we really think we have great heritage in building here. We have a tremendous team that has been building these products for generations.

Now we've got really sophisticated customers that we've been working with in kind of somewhat of a joint development model, right? We're co-engineering together, looking down the range and saying, what are the constraints that are going to affect kind of our next-generation decision-making? Where do we need to be? Where do we need to place our systems? And that allows us to get out in front and build, in a lot of cases, some of these products that we ship that are pathfinding for our general purpose products, things that you wouldn't find in the Arista catalog on the website, but we're deploying at scale.

I think some of our competitors have recognized that, that level of sophistication exists and those opportunities exist. So they're incented to invest. Whether they can do it as well as we can, I mean, it remains to be seen. I think we're going to continue to focus on relentlessly executing with our customers. And then we generalize those things into architectures and systems that we can sell to the broader market, and that's where the enterprise follows and the Neoclouds consume and kind of slot into that quite well as well. So yes.

Maybe let's talk -- move to talking about scale across a bit. You indicated on the call, it's going to be about roughly 1/3 of your AI revenue this year. Why -- how -- like what drove -- what is driving the number to move so quickly? I mean this was practically nonexistent last year. It is going to be $1 billion almost in revenue, $1 billion or more, I guess, I should say, given your track record. But what's driving the number to move this quickly?

Yes. Well, like John said, I mean, the AI revenue even a couple of years ago was almost nonexistent. So that this one can grow independently is not super surprising. We do see some customers that are adopting architectures that heavily feature scale across inherently in their design. They're working around the data center space and power constraints proactively by saying we're going to adopt smaller tens, low hundreds of megawatt facilities, getting them together and construct them into a fashion that inherently requires on a scale across solution. Others are building massive campuses where they're layering in data centers over time on a singular campus and then the scale across works to interconnect all of those together as well.

We've also seen some novel architectures even on the inference and agentic side where customers are creating more of like an interconnect fabric for inference and access to different clouds and on-prem capabilities to power agentic. Some of it is redundancy, some of it is latency optimization, again, following principles that we've seen similar to kind of Edge computing and cloud really and its evolution into the Edge. And so -- but in terms of like is 30% durable, should we expect that every year, can it grow? When we think about the mix within the AI bucket, I think it's very dynamic still because there are so many architectures running in parallel. There's new systems and new silicon shipping in the next 12 months, another 12, 18 months behind that. These things are going to inevitably affect the architecture that folks are deploying.

I think we're a long way from seeing coalescence onto a common kind of set of design principles in AI. There's still going to continue to be a lot of experimentation. And as long as we're working around constraints, I think we're going to see mix shifts between whether it's a bigger singular facility with a lot of scale out versus more scale across. But I do think, as agentic continues to grow, as it gets commercialized, adopted, enterprises adopt, we're going to see more push to the Edge. That definitely serves -- or we have good products to serve that use case well in the scale across arena.

In the early days with the 4 customers 6 quarters ago, it was easy to track these ratios. It's getting really hard because they might build out different pieces of the network at different times. So it's difficult -- more difficult to see the alignment of the ratios...

Yes. Customer mix and then architecture choices within them for sure...

I mean scale across for inferencing has been coming up more and more. Are you able to delineate when you engage with the customer, whether it's a training, let's scale across or whether it's already planning for inferencing driven by that?

Yes and no. I mean, insofar as we have deep engagements with those customers, we have a sense of what we're doing, what we're competing for in the business and what their configuration and their architecture concentrates on and product selection. It's a little bit easier to differentiate. But again, some of these things are very temporal, right? There are decisions that are a moment in time. There are several customers who are building scale across to interconnect data centers, who, if they could gain access to a single gigawatt facility, would much prefer, right?

And so they're going to continue to work their own supply chains in parallel, data center side, compute side, storage network and converge it all. So sometimes we're going to be a lagging indicator of some of these decisions, sometimes our mix of our own products and shipments will just reflect some of that customer choice and how they solve for the constraints that they're facing. But broadly speaking, yes, I think the customers who concentrate on agentic, on inference, we can differentiate those pretty easily because there isn't the associated training workload running with it.

You brought up this earlier, Neoclouds or call it -- or maybe you can even take the segment that's just the Frontier AI labs working on Frontier models. Historically, when we look at the large hyperscaler space, your customers there, you've co-designed, innovated with them. What are you now seeing in terms of engagement model from the Frontier AI labs? Are you co-designing with them as well? What's the magnitude of the breadth of the customer vertical that you're looking at?

Yes. Well, you would have seen the press release this month of the MRC protocol spearheaded by OpenAI with a lot of partners, including ourselves. So we are deeply engaged with the Frontier folks. We treat them a lot like our hyperscale customers in terms of knowing that they're super strategic in the industry and the market. We have a lot of faith that they're going to continue to grow. We want to be able to grow with them. We want to be able to support and help power their growth as well. So those tend to be a lot of more deep relationship kind of close co-engineering, engineering extension kind of relationship. Some of them are earlier phase in terms of moving down the stack where they may have started in the cloud and they're taking on more and more of their infrastructure. We're uniquely suited to help come in and help them build to solve those problems. So that is part of our go-to-market motion when we're working with these folks is like how can we help fill in some of the gaps with them on staffing capabilities, architectural choices.

And so the motion, I think, and the engagement model looks pretty similar to hyperscalers in many respects with a lot of these Frontier folks. And certainly, even on the Neo side, as they continue to grow and become bigger entities, vice versa kind of -- it actually looks pretty similar there.

I think they recognize that token -- I've been doing some work on token economics. Token production costs are very sensitive to network performance metrics. So if you change latency or jitter just a little bit, you can really -- you can very quickly blow up your token production costs, and they're aware of that. So that's why they're engaging with a company like us.

Actually, on that front, Rod and Tyson, feel free to jump in. How are you guys thinking about networking content per rack if most of the infrastructure going forward is for inferencing, which again, I'm being simplistic here, but as you shift from training to inferencing, as more of the CapEx dollars are diverted to that, how do you think about networking content per rack?

Well, if I understand the question correctly, I mean, I think that when we look at our product set and we look at like the experience we have of high-density networking in general, it's kind of replicating itself now in just a different deployment model, right? So if you're talking about inference at the Edge, kind of complete package model, compute co-located with storage co-located with network and then the need to be able to pull from disparate data sources, provide security over the top, I mean, in many respects, it does mirror some of the common patterns of the kind of age of the Internet and the cloud in terms of what the demands are. It's just a lot more density, a lot more demanding, a lot less jitter tolerant, latency tolerant.

And so we've had a lot of conversations. And when I joined Arista, within the first couple of weeks, having conversations about the importance of like the rack as the product. And I think if you turn around and you look at the heritage on our largest chassis, the 7800s, there's a lot of technology that slots in very nicely for building these kind of co-engineered, highly integrated networking systems where the network is a piece of the story. And even with things like our CloudVision and APIs of our software, it lends itself very well toward automation and giving customers tools to be able to deploy these things, manage them, monitor the workloads, maintain them, security patch them, identify vulnerabilities to kind of help with partners and some of the other integrations that we have.

So I think the rack scale networking becomes a logical extension in the moving forward for us. And I think that's true in scale-up domain as well, not just inference at the Edge.

Let me just pause here and see if any questions in the audience. If you have a question, you can raise your hand and we'll get a mic over to you.

Okay. So maybe just continuing on then. One of the things that we've seen more recently is this thesis that's around agentic AI finally being a catalyst around enterprise customers, modernizing both their data centers and campus. Just, one, what -- how are you thinking about that overall sort of thesis? And do you -- are you seeing any evidence of it yet within your discussions with customers?

I think we're seeing a lot of vertical activity around AI and the enterprise for specific use cases, could be health care, financial, a lot of anticipation about agentic AI. I do think it's affecting people's forward-looking decisions, especially around campus and the Edge. Maybe they would have sweat the assets before. Maybe they're leaning in more to higher speeds and higher performance WiFi 7. But I think in terms of broad-based agentic AI, it's not really happening today. But I think it is people thinking about how it will in the future.

One piece that I think will be important when that day comes is you're going to lose a lot of visibility into what's happening with applications. And when we saw the transition to the client server, there was a lot of consternation about how you would monitor performance of applications. In this world, there will not be a lot of visibility. I think there's a unique opportunity in the network with network observability and the observability architecture we put in to EOS and CloudVision to really get a viewpoint into the topology and activity inside the network that should be an advantage.

I mean the other thing I'd say just as an anecdote, the silly anecdote maybe, but I've got an agent running at home and the thing is actively marketing Arista networking gear to me because I basically told it, "Look, I'd like you to be able to manage my network." It said, "Well, the stuff you've got not great. I said, "Well, what should I get?" It said, "Well, you really should get Arista, but unfortunately, it's too expensive because it's enterprise gear." And I said, "Well, I have a discount?" So that's a silly example, but when you get into enterprises and they realize that our API is so rich and broad, that is going, I think, to sort of start selling itself as they begin to use agents to manage their networks, and that's inevitable, I think.

And we have our own AVA technology we built on top of CloudVision and NetDB to help customers assist in the build out of their networks and visibility as well.

Before we run out of time, maybe just to hit a couple of topics that are of interest to everyone, XPO. You now have, I think, 100 vendor endorsements, [indiscernible] compared it to OSFP in terms of the run that it could have. How should we think about XPO? What it means for Arista economically in the sense that, yes, you enable a new standard in itself, but how does it play into the revenue opportunity for Arista?

We do sell optics that are Arista optics, Arista-branded optics we sell, but we do also enable our -- especially our cloud and hyperscale customers to procure their own optics. So there is some economic piece around the optic itself, but system readiness is key for us. We invested in the OSFP form factor and the MSA I think back in the 100-gig days anticipating 400-gig. We saw some adoption at 400-gig, but it really became mandatory and a better cooling and thermal dynamic in the 800-gig range. And I would say our market share gains that we made and our competitive differentiation would not have been possible without the OSFP. And I think we'll see XPO in sort of the 1.6 begin to be adopted, but to achieve the density, I don't think you'll be able to achieve it at 3.2 without it. I think you have some kind of use cases where it could be an advantage...

Yes, I think that's right. I mean I expect a similar kind of adoption uptake and 1.6T will be the kind of early adoption, but I think the modularity is really great here. You're going to get the density and power savings immediately, power efficiency out of it, but the ability to build a configuration on the fly or specify it to meet exactly what you deploy, if you've got a mix on a DCI use case where you've got ZR optics and you need to half load a system with ZR and you need the other half to be FR MDR internal-facing optics, you can do that very easily with an XPO kind of at deployment, at run time across a diverse supply chain ecosystem. And then you've got field replaceability for reliability and maintenance in the field. So there's going to be some use cases I think that will evolve pretty quickly where you need these mixed modalities of optics where XPO will play quite nicely. And then following on to it, obviously, we are heavily focused on XPO.

You asked about XPO, but we're also invested in CPO and we have a belief that a lot of the work we're doing around modularity, modularity to the optical engine, some of the manufacturing techniques that are being done and XPO can actually apply directly into CPO as well. It will be hard to ever achieve field replaceability because these optical engines are just super sensitive to contamination. They're super brittle. If you break something in the field, you are boing to break the whole switch. And so that's going to be difficult. But a model whereby you can replace the switch, have it serviced and then repaired and brought back into service later is much better than some of the early CPO stuff where everything is really manufactured, sautered down together at run time, it's a complex manufacturing process, yields are lower, that affects costs and then there's no reparability for it whatsoever.

So we think these 2 things will play very nicely in the ecosystem together, XPO and CPO. We just think some of the challenges that customers have around density and speeds and feeds are going to dictate that the XPO -- something get there a little sooner and the XPO will come.

XPO is an example of how we stay on the front edge of the learning curve as well because we're always -- it's such a geeky engineering-led company, we're always trying to solve problems for customers. So we're aware probably in advance of just about anybody else in networking what options are coming down the road because we're codeveloping things. So we don't make mistakes strategically thinking that maybe CPO is the only option. And there's others in the industry that bet the ranch on CPO not realizing there're other options maybe that will come along.

So we're able to distribute our risk across different technologies because we're on the front edge of the learning curve, and that's a really important thing to understand about Arista.

Okay. Maybe just while we wrap up, can you just dive into that a bit because I think there's a broader perception that Arista is delayed in terms of a CPO solution, right? Can you just dive a bit more into what are you seeing from your customers today? And if you were to put time lines around XPO versus CPO, where would you look to intersect the market with an XPO solution? What time line would you expect to intersect the CPO market as well? Or how are you thinking about the cadence of when customers start to adopt these?

I think the vision of CPO really recognized the law of physics where the connector and the chip they're closer together because you just couldn't span the distance and the opportunity to save power by putting -- not duplicating DSPs in your switching chip and in the optics. The problem has been mix and match capability of different optics as well as if an optic breaks, now you're replacing this very expensive switch part. So what we've been able to do working with the industry and the ecosystem is mitigate those downsides with the OSFP and XPO. And eventually, we believe there will have to be a CPO-like form factor, but the thing we're working is to make it really interoperable with multi-vendors and an ecosystem type of approach.

So as the power goes up, if you get past 3.2, there'll probably be some inevitability for the beginning of CPO in a more broad-based environment. I think you'll start to see some of that also in some deployments in the next couple of years as well, the beginning of it.

Yes. I think you'll see some uptake on CPO early. I mean, like Rod said, some people are betting the farm on it, betting the ranch on it. They're going to be very aggressive in trying to push it out and inevitably, there will be some adoption, but it's really well suited for certain use cases, a lot of high-density network traffic. So scale up, scale out to some extent plays pretty well, but I think where you're going to really see material adoption is when you start to get some of these other things that customers really want. They want the serviceability, they want the optionality and then the economics kind of fall behind it.

And like Rod said, some of that geekiness to us, like it's accretive, right? Our evolution of our thinking on CPO is because a lot of the work on the XPO is a pathfinder to say, "No, you can package these optical engines up into a meaningful way. You can use an interposer to interconnect those to the silicon. You can still use that silicon kind of instead of a DSP use the SerDes for driving all that signal." And now you can actually imagine a world that becomes, from a manufacturing perspective, easier to stack up as you kind of work up through that L10 assembly as well. So kind of what I would expect to see for us broadly in the industry is probably XPO really takes off '28, some early adoption may be late '27, and CPO behind it. There will be some earlier CPO deployments on that for sure. They'll kind of run neck and neck to some extent.

Great. I will wrap it up there, but thank you for coming to the conference. Thank you to the audience as well.

Thanks for having us.

Hello, and welcome to Needham's 21st Annual Technology, Media & Consumer Conference. I'm Ryan Koontz. I'm Really excited to host Arista Networks here today joined by CFO, Chantelle Breithaupt; and John McCool, SVP as well as Rudy Araujo, VP of IR. So welcome guys. How are you going?

Great. Thank you. Thanks for having us again.

Excellent. Super Well, Chantelle, let's start with your excellent quarter you guys had. Can you maybe walk us through how you felt about the results, obviously, beating expectations, and what's some of the kind of key demand drivers in your first quarter were?

Yes. Thank you. We were very excited as a company to report our results. I'm very proud of them as a team. We're very pleased with the 35% revenue growth. And if you take that, Ryan, with the change in deferred revenue, it's actually 54% of revenue forward-looking kind of growth, so we're very excited by that, right? We raised FY '26 again to the $11.5 billion, twice in the last two sessions with you guys raised the year, which is great. And if you guys go look, I encourage you all on the call to go look at our latest earnings deck where we actually also raised our 3-year CAGR outlook to be 20-plus percent. So a lot of great news, a lot of great momentum. So super excited. As Jayshree said, we've never seen more demand than we see at this time.

And we're really encouraged. We're encouraged by the progress we see in AI and the adoption of our 800-gig Etherlink portfolio. We're proud to be winning new logos across all the sectors, all our customer sectors. And the proof of concepts for Campus are really working well. We have a high win rate when we can get in there and look at the Campus where we're only 3% market share today. EPS growth, I think, was stellar. So sum it all up, we're super excited and feeling the momentum.

Excellent. And within the quarter, hyperscalers, your big customer is obviously a big contributor there. Some of your Neoclouds, I'm sure. Can maybe walk us through your thoughts about how your different customer segments contributed in Q1?

Yes. I think that we said it was a really great quarter across the board, but in the sense that we saw some really strong performance from the AI specialty provider segment, which, of course, is across the AI portfolio. So I think that's continuing to do well. And so from that perspective, I think all of those segments are showing really great signs for the year.

Super. Yes, you mentioned the nice step-up in deferred. I think it was up 100% year-over-year, it's just incredible. I mean, can you remind us again what's going into deferred, and what's driving that acceleration here for investors that they should be paying attention to?

Yes. I think it's really important for investors to understand and to go look at the earnings deck because we have some analysis and trending, which I think helpful to position where deferred is for Arista over time. But just to remind, what goes into deferred, so we talked about the $3.5 billion of AI revenue in the P&L this year. But most of what's sitting in deferred is AI use case and product. And so you have to look at both combined to look at the AI demand that we're seeing across the board. And it's not that we just put it in once and it's aged over 2 years, things come in and out of the deferred into the P&L. So it's rotating, it's refreshing.

And so I think from that perspective, it's important to look at the earnings deck pages we have. But it's a sign of growth and momentum in the sense of the AI demand that we're seeing.

It's clear, incredible demand environment right now for yourselves and a lot of your peers and your ecosystem partners, I'm sure.

And with regards to your full year, raising that to 28%, what were some of the puts and takes there? Coming out of the quarter, I think some of the investors were maybe a little disappointed that you guys didn't pull the second half up as much. And how would you characterize your constraints there on your ability to raise your full year a little more relative to your supply environment?

Yes. I don't think the year guide was in the sense of necessarily tying it to the supply environment for this year. I think -- and John will take us through probably a little bit during your questions in the sense of the supply environment. It's a when not if to us. I think you have to take our -- I think I would take it the inverse way, Arista raising twice, and we never assume 100% of everything is going to work in our guidance philosophy. So I would take it as, hey, Chantelle and Jayshree, raised twice in the last 2 quarters, and we're only in May. So let's keep watching the future quarters. And if everything can fall into place, we'll see where we can get to. I think that's how I would take it versus always trying to meet what people expect from us from a guidance perspective. That's not new. So we...

What do we get that? Trying to be conservative as you guys are, and beat and raise, we like that.

If I have to choose a philosophy, I would choose that one.

For sure. And John, how would you characterize the supply environment and some of the puts and takes, and what you guys have been successful in doing. You obviously tie up a lot of capital in terms of commitments with your strategic big partners like Broadcom, kind of walk us through your thoughts are on some of the things that are going well and some of the challenges you're feeling right now on supply.

Sure. Let me start with the going well. If you think about Chantelle's comments on deferred, the actual supply chain team had to deliver both the deferred revenue plus the revenue that was recognized. So 54% growth year after year. So we have a lot of busy folks and our suppliers are very eager to support us to fill this demand. That said, the entire supply chain environment with the AI demand is really pressing capacity on foundries. First, we saw that with memory and how that played through. There's a lot more fabs with memory on silicon, there's some concentration in those fab cycles. So that's what we're up against is an environment that's really booming across the board and really having to drive that and to turn it into finished goods and ship. So we're really comfortable with our execution so far and we'll continue to drive it.

In regards to your updated guidance, do you feel like you have everything you need in place in terms of supply chain to execute on the guide and hopefully, chase some upside there?

We're very comfortable with the guide on both margin and revenue, for sure.

Excellent. And can you maybe walk us through how you use purchase commitments to really lock in that supply, your purchase commitments were up pretty strong, is that around memory?

Yes. I mean that's -- it's across the board, but in the areas where there are constraints, specifically, as we look ahead into 2027, and what we're seeing from the demand environment, but it also helps us in short term, get very tighter with those suppliers in terms of our demand today.

Yes. And the only thing I would add, John, to your comments is that raised to $8.9 billion, it is across everything. The majority, I would say, is chip related, and it's a 52-week lead time, right, John? So we're leaning into your point into '27. So getting ahead of making sure that, that's sorted for what we need to deliver in '27.

TSMC guys are going to be pretty busy. I think...

I think they've been busy for a while.

Yes. Great. Well, maybe shifting to customers and what's happening in the AI domain. It's been a huge success story for you. I remember back a couple of years, a lot of questions, what was Arista's role in the AI back end. And you've certainly proving that out, and you're knocking down some great wins. I think in the quarter, you announced that your fourth AI customer had moved their back end from InfiniBand over to Ethernet. What brought that customer over? And how do you feel about that going forward, all of your AI accounts there?

It was the same trend we've seen on other ones. That particular customer was kind of early on had made some investments in InfiniBand and the full NVIDIA stack. But the compelling reason to move to Ethernet is multi-vendor supported, agnostic to the endpoint. So NVIDIA GPU, AMD, something you build on your own inference engine. Same capability as your front-end network, your back-end networks, so the operational simplicity around it.

And I would just add, you're seeing the same dynamic around NVLink and ESUN, starting to come up the industry rally around that and while that's not a near-term thing in terms of revenue, I think that also has bearing us in terms of the architecture going forward.

Super interesting. Wow. Maybe Chantelle talking about price a little bit. I think you announced some price changes to offset the memory costs. How is that flowing through the income statement here relative to price changes?

Yes. I think that -- back to John's point, I think the supply chain team has done great work over the last few quarters. And you've seen various companies talk about different quarters in the last 3 to 4 where memory has been in their conversation. Our philosophy or our approach and strategy was to understand the market, understand where we thought the pricing was going to go a cost to us. And the kind of the added measure twice, cut once. So we want to be sure we had a pretty good line of sight and visibility into what the cost was going to be for us for the next few quarters.

And so with that, Ryan, we did a price increase, which customers never love, but at least it was a well-known topic. Our goal was to be margin neutral over the time of the backlog converting to new orders. And so I would say that kept us in our 62% to 64% gross margin guide, which we've been talking about for, I think, 3 or 4 quarters now. So I'm super excited. We've been able to hold it, through it. We've earned the value and trust of our customers to have those commercial conversations. So we're pleased with the result, but never like to deliver a price increase to the customer. So I think margin neutral is landed. And that's allowed us to keep our guide.

Sure. I mean everyone in the hardware industry is feeling that one pretty hard. And are you guys exposed to both DDR4 and DDR5 across the portfolio?

Yes, not the same equal weighting, but yes, to both technically. Yes.

Great. As you think about your AI revenue bogey here of $3.5 billion, which is raised from $3.25 billion. That counts -- does that counts your 4 large customers there and not others that you've added -- new logos that you add?

Yes. I think that $3.5 billion, the way I like to look at it as you look at it as a percentage of the $11.5 billion, we're talking 30% of our revenue not too many years into AI totally is a pretty good win rate. I think that -- but it covers more than the 4 customers. I think that you heard a couple of quarters back, you could technically count about 100 customers across to Neocloud, some large enterprise, the 4 we talked about. The 4 initial pilots were just to take you along the journey with us as investors and the community, how it was going with the larger ones, but it covers AI, we think, is the predominant use case. So we're proud to work with some of the largest customers, with some of the smaller ones that are starting to get into their inference conversation.

I'd like to explore that, if we could. Just discussing some of the Neoclouds, what the Neocloud kind of procurement model, how that might differ from a hyperscaler type model? And then maybe talk about inference, could you guys dive into that?

Sure. Yes, I can probably jump in on that one, Ryan. So I think the Neoclouds have kind of evolved a little bit, right? Like I think when they started, A lot of it was looking for a rinse and repeat kind of reference architecture kind of model, frankly, they needed allocations of compute -- bound them on networking decisions, right?

I think what they realized is they were all starting to look the same, right? So how do you differentiate in the crowded Neocloud market in terms of what you bring to the table. And I think what they realized is the network is actually pretty critical to that differentiation, right, because the network can mean the difference between job completion times, taking longer or shorter, it can mean the difference between power utilization being higher or lower. It could be the difference between time to first token and when we start talking about inferencing, et cetera.

So what we're starting to find now is that they are actually realizing that they can't just be bound into these agreements based on what's good for compute because having the best-in-class compute is no use if you've got a substandard network, right? So that is opening up the market for sure. I think the other thing is they are definitely not buying at the same levels as the largest hyperscalers, right? So that's perhaps stating the obvious.

But they are also the folks that need a little bit more handholding, right? They don't have the experience necessarily of having built these large cloud networks, et cetera. So we're also seeing a higher attach with CloudVision, for instance, and with our Op capabilities that gives you AI Ops kind of visibility into it. So that's kind of a little bit of an interesting dynamic there relative to the hyperscalers.

Super interesting. And maybe to follow that up, you guys have talked about both scale across as well as scale out kind of that dynamic, can you maybe unpack that for us a little bit and where you've come from and how you see the market evolving in the back end there for you in terms of opportunity and what's driving your revenue?

Yes. I think we -- it's interesting to us because we saw a similar dynamic in the beginning of cloud. Once we were able to connect all the servers in the data center on a 2-tier spine leaf architecture the question is how do you expand because I don't have the physical capacity in this data center for more CPUs. Now it's GPUs. So scale across is becoming kind of the next generation of the leaf spine architecture and the Universal Spine to interconnect. So we see all different use cases, but specifically where you want one logical AI cluster across multiple physical instances.

And scale across sounds like it's in the pretty early stages of build-out. Is that right?

I would say relative to scale out. And I think in the sense of -- and I know Rudy, you're going to jump in. This is a very exciting topic for us. I think -- so the only thing I would say is that it's fairly new, but you can see how it might be required because a lot of this is when you need these kind of federated space land access to power cooling scenarios, and I think that's part of what's driving that need. But Rudy, I know you want to take that on.

Yes, I was going to say exactly that. And the other side of that point is, as you start talking about inferencing, Ryan, you brought this up. Inferencing is increasingly like how do I get as close to the edge as possible, right? Because if you're sitting in New York right now and you're asking -- you're trying -- you've got an agent running on your device and you're trying to interact with the model, you don't want to be waiting seconds [indiscernible] over time. And so that's the other thing that's going to drive, I think, scale across. So we're very excited about it, partly also because it's a very unique product set that it takes to be successful there, right?

And we've shown success there. We've got the products frankly, a lot of our competitors that we would run into and scale out don't have the product set to compete and scale across, right? So I think we feel very good about it, but it is relatively early to the point the Chantelle and John was making.

Excellent. Yes, that's great. And unless we forget about good old cloud, front-end cloud. And I think the back end has pretty much sucked all the thought out of the room and the attention. But maybe tell us what's happening in the front-end networks these days. That's been your core business, where you've been such a dominant leader for so long. How is the front-end changing to adapt to some of the new AI requirements? And what are you seeing happening in the boring old cloud business?

Yes. I mean it's funny, right? I mean, I forgot -- at least in the analyst world or in the world it is. I mean, for us, it's certainly a tremendously important piece of the puzzle, right? From an upgrade cycle perspective, if I can call it that, most customers in the front end are still in the 400-gig era, right? At least at the hyperscaler level. Some of them are maybe in the 100-, 200-gig era because frankly, the applications that we're using today are working just fine, right? I mean, we're on a web conference, we're all coming in nice and clear, going over those 200-, 400-gig networks.

It is triggering a cycle, especially as this agentic traffic pattern starts to become more commonplace, it is going to trigger that cycle. Probably next year is where you start to see the front-end networks kind of get upgraded. I mean we're certainly having conversations right now, it's in maybe more planning phases right now. But maybe next year is when you start to see the 800-gig kind of upgrade just as the AI clusters start to go into the 1.6 era, right? So there's probably a little generational gap between the front end and the back end kind of naturally if you would.

The other thing, by the way, is again, inferencing is having a huge impact there. And frankly, we didn't touch on this, but inferencing is also having an impact and the agentic is also having an impact on the enterprise side, and I think we can talk about that as well. I know it's not front end, but I just wanted to kind of get a...

I'd like to hear that on soon.

Yes.

That's great, Rudy. Yes. On the enterprise side, you're having these early discussions, I'm sure with big Fortune 50 types and financials. And I can imagine those are the leaders leaning in here. Like where are those discussions with you about architecture and build-out? Who's going to hold their hand to go do these sorts of things? I mean, do you think there are traditional guys like your competitors that are maybe more deeper entrenched in enterprise that are going to get their fair share there? How do you think about the enterprise kind of AI build-out to complement where they already have quite a bit of private cloud out there, I'm sure, where you guys are very strong.

Yes. I think one thing I'll add, or at least start with and the team can chime in. I think one thing that we've heard from some of the larger customers because part of that $3.5 billion is with enterprise customers, as we mentioned. So part of what they are starting to think about as they go through their refresh cycles as they think about new data center builds, Campus and data center, to be fair, is what is their AI landscape, what is their AI goal in the company. And one thing that we're finding is really great in the conversations when you can have one operating system like EOS across all of those components, it's super easy to have agentic AI sit on top of that because you're only having to hook into one operating system.

You can imagine having an agentic AI kind of portfolio where you're trying to go over different operating systems, not as easy to hook in, not as easy to be ubiquitous experience. I think that's a great -- like Ken thought about that maybe 20 years ago, not sure, but he designed something perfectly built for it. That's the things we hear from customers. We're hearing them pull in their campus refresh earlier to try to get into this get ready for at the edge agentic AI inference. Those are some of the things we're hearing. But Rudy or John, anything you want to add?

Yes. I think Chantelle, you hit the -- maybe the most important thing I hear from customers, right, they're realizing that AI within the enterprise is not just a Campus issue. It's not just a data center issue. It goes across the branch, the Campus. To Chantelle's point, that unified operating platform across all of those is really something that's exciting. It's frankly the other thing they're asking us and we've heavily invested in is what is our AI strategy for -- AI for networking, right? Not just we've talked a lot about AI, but AI for networking and AI for Ops. And we've continued to invest there in AVA, which is our Autonomous Virtual Assist, driving better outcomes, if you will, for customers and being able to do everything from root cause analysis, to helping them automate as much of the network operations as they like, right? Like this is not about, hey, you don't need a network operator anymore. It's about how can we augment what the network operator is doing and make their life more efficient, if you will.

Really great, Rudy. Let's shift gears to one of my favorite topics coming out of OFC, which was the XPO announcement. And I went in very excited to hear about it because I hadn't done a ton of work on it myself. I was probably a little behind. But wow, I walked away from OFC and saw just the broad industry endorsement and was so impressed. So can you maybe walk us through kind of how you got there, how you built such strong support and what it means to Arista for that to be adopted as an industry standard?

We have a very passionate founder, Andy Bechtolsheim, who...

Yes, you do.

Over many decades, has really led the definition of many MSAs working with these partners. It goes way back in terms of his development. The most recent before this was OSFP. And we anticipated that at some point, the thermal dissipation would be such that the conventional state-of-the-art would not be able to contain that. And that really was an enabler to the products you see today and our leadership in those deployments. And we're seeing the same thing happen as we go to 1.6T and beyond. The OSFP form factor was great, but won't be able to carry the industry on the next generation.

So he's been working with the team here at Arista and the optics team with those partners and building it out and we engage with them in testing and definition. So it's been a real great collaboration with the industry.

Yes. It's a super cool solution. Go ahead, Chantelle.

No, no. Okay. I don't want to stop you from saying it's super cool. What were you going to say?

No. Go ahead.

But I think that we're very proud because it's industry-leading, right? But as far as what it means for us, it's open to the industry and may all people play competitively in it. But I think for us, again, it's another thing where we've been kind of ready. We've got our product portfolio already thinking about it for the next generation of things. So I think it just place us well to be ready for something that the whole industry, hopefully, and our customers, more importantly, will benefit from for years to come, right? And so...

For sure. Yes.

The other aspect of this that sometimes I think doesn't get fully absorbed is the front panel density that this can add, right? So being able to shrink those racks opens up a whole bunch of other scale-up technologies from RF to micro LEDs, et cetera, that are far more power efficient, right? So it opens up, I think, an entirely somewhat tangential benefit but a super important benefit as these clusters start to get more and more dense. And the front panel density was becoming a limiting factor, right?

Not to forget, by the way, the sheer amount of sheet metal, less sheet metal that you need and the structural cost that, frankly, aren't paying the bills, right? Like they just deadweight, if you will.

Yes. I think the stat we would throw out there is 40-plus percent footprint reduction, just to give the audience an idea of what we're talking about.

Yes. And so really, you're saying not just for scale out, but there are other applications for that same technology that you can embed on that sled that XPO...

Yes. Scale up is probably the most interesting one because getting to the point where it was as easy definition, right? If you're within the rack and scale up, if you go outside the rack and scale up, well, we realized that we need more compute in that scale-up domain. And so now we're thinking of ways to go across multiple racks, but still stay within that coherent memory kind of scale-up the domain. Well, technologies like this are going to enable that, right, because it gives you a lower power way to interconnect these across a larger physical distance, where copper is really going to struggle, right?

So not only is there a direct benefit from an optical perspective, but I think it opens up other avenues. And like everything Andy always does, right, it's a completely open standard based, there's nothing proprietary, there's no like, hey, you have to have Arista to get this to work. But I think to Chantelle's point, it really continues to show the thought leadership that we bring to the table and how we move the industry forward with these open standards.

Super cool. So when do we start seeing products in customers' hands and supply chain start to ramp? What's the time frame that this starts to impact the industry?

So I think this is really something that will probably be most impactful, in the 3.2T era, 1.6T, you're still going to continue to see. And frankly, I think if Andy was here, he'd tell you, OSFP is not going away, right? There is a vast majority of optics out there for the foreseeable future will be OSFP. Especially when you count the enterprise and things like that. But I think for XPO specifically, you'll start to see the first products, I think, come out next year and really very large ramps in that 3.2 terabit era, where its value is much more impactful than to John's point, right, where OSFP would struggle.

Right. And the only other thing I would add to that is, I think about a quarter ago, so everyone thought 3.2T was when we'd have to go to CPO. And I think this shows that now we have a whole longevity even at 3.2, that's not required.

Exactly. Great. Maybe shifting to the Neocloud. I know we touched on them already. But you talked about a more consultative sale. Can you maybe unpack that a little bit for us in terms of what your role is in working with them, they're not probably as deep and sophisticated as some of these hyperscalers have been doing this for decades. So how is Arista's role evolving with the Neoclouds and AI specialists?

Yes, I can start and the group can chime in. So I think -- I personally think Neoclouds is a fascinating segment. Every one of them is different and a pleasure to work with. But they all -- they come to us differently. Generally, they come to us because they know we have the bigger cloud experience. So they would like more of that like what Rudy and John described. But there are different scenarios that we come into it. There's a scenario where the team has experience, and they know us and it's a best-of-breed open conversation, high win rate there because everything -- our portfolio of Etherlink suits that very well and all the things that Rudy mentioned.

Then there's the example like what Ken Duda mentioned in our earnings call, where the first try of it with the team, team capable, but in a different scenario. They went with a different solution. And now they've come to us because they realized the capabilities required of their incumbent technology and architecture didn't serve their scale-out needs. And so Ken Duda did a very good job, I think, articulating that scenario, we see those. And we see things that when people feel emboldened to go to different XPU vendors they come to us because they know we're agnostic to the XPU. So we see, they come at us from different angles, but we are known as being a great provider to them for their AI needs.

I don't know, Rudy and John, if there's anything you want to add, but...

Yes. I mean I think in terms of the selling process, the only I would add is, look what's done well for Arista is this deep engineering to engineering partnership, right? Like so having our really folks that are working on some of these largest networks be in that consultative selling process with the Neocloud is something that they appreciate because, frankly, they are breaking new ground. And they're trying to do this at kind of the speed of light, so to speak, right? What normally might have taken a year or 2 years to ramp, you don't have that luxury anymore.

The only other thing I'd add, and I'll tie this back to your previous question about optics, right? Like one of the things that they appreciate about working with us is we don't lock them into, well, if you want to do this, you have to go down the CPO route, right? Like we still give them that optionality. If they want to go down CPO, we're thinking about open CPO and things of that nature, XPO is obviously there. LPO is an option, traditional OSFP fully retimed, like whatever -- they don't feel as locked in, and I think that's something that they're really beginning to appreciate because as I was saying earlier, ultimately, they've got to bring dollars and cents in the door and from their end customers. And to be able to do that, they need to be able to differentiate from the Neocloud down the street, right? And how do they do that? They're all running the exact same architecture, if you will.

Sure. Makes sense. I want to go back and touch on agentic. It's going to come up more and more and more. I think it's really starting to impact the company's core business with all the different agents that are being rolled out. How is that affecting the architecture of your customers? We hear about more CPU content, locating CPUs closer to GPU clusters. I mean, how does this affect Arista's demand equation and the conversations you're having with customers in terms of agentic infrastructure?

Do you guys want to start? Or do you want to start?

Yes, I can start. So Chantelle kind of touched on this a little bit earlier, right? Like people are starting to think about, okay, what does this new bandwidth pattern on the -- look like, right? In our traditional world, maybe it's a little bit more bursty, right? You go to a ChatGPT, kind of more chatbot, kind of a world of LLM or generative AI. It is still bursty, but maybe the burst are slightly bigger, if I can use that term. You get to an agentic thing, and these things are talking to each other all the time, right? So how do you kind of manage through that process?

So as Chantelle had touched on that, right, WiFi 7 is becoming maybe a sooner cycle than you normally would have seen with these WiFi migrations. They're also thinking about things like what does that mean from a power Ethernet perspective, right? Because now you've got all these smart devices that are all across your environment that are also becoming agents in themselves. So how do you enable that? How do you enable them to be securely connected to your network, right? So Zero Trust Networking is another aspect to the conversation that they're having.

And then frankly, kind of the bread and butter stuff around things like routing and encryption on the wire and multi-tenancy, like stuff that, frankly, I think most people would think is boring until maybe 6 months a year ago, but it's becoming super important now. And again, I'll repeat what I said earlier, we're one of the few folks that has the products to play in that space, right? So we like our shot as we get into that. The competitive -- it's not that there's no competitors, but it is a different competitive environment than I think you see with maybe scale out, for instance.

And I think that -- sorry, go ahead, John.

Just one of those boring parts that Rudy has talked about that's baked into the architecture. Whenever we see like a mission-critical application, we see a thrust in those customers looking at observability, what's happening on my network. So now you can imagine all these agentic communications, this new type of workload. What kind of visibility can you provide, customers can see what's actually going on. That's going to be pretty important as well.

Is that where your telemetry kind of has a lot to offer?

Absolutely, not just the telemetry itself, but the architecture of the state-based architecture and be able to stream information is going to be critical.

Yes. That's a big differentiator for you guys. I've heard a lot about that. With regards to 1.6T coming, how are you thinking that the long-awaited Tomahawk 6, how is that going to impact your mix and demand? And is this just a normal generation? Or does this bring any kind of accelerators to the demand equation to keep up with all this bandwidth?

Well, I think we're not preannouncing anything, just to be clear in the conversation the way you asked the question, but I think you could count on us to be ready when the cycle is there to be ready with the great products we usually do. So take that as we'll be ready when the market is ready and everything comes together from an ecosystem perspective. I think -- I don't know if there's anything there, Rudy, or John, we talked about in the sense of accelerating at this time. Clearly, there's a demand in the portfolio in the customer set. But Rudy, anything you want to add?

Yes. I mean I guess just one interesting data point, right? Like I think we released 100 gig products. I mean John will probably keep honest here, in 2019, I think, give or take. The 800-gig products came out in 2024, right? So that what, 5 years? It's not going to take 5 years to get to. I know we're preannouncing product, but I can tell you it's not going to be 5 years to get to 1.6T, right? And people are already talking about 3.2T. So in that sense, the pace of innovation is absolutely rapid, right? I don't think it's anything we've seen in the world before. And frankly, that's why when Jayshree says, look, this demand environment is nothing like I've ever seen before. And by the way, she's seen a lot right in this industry. That's the say probably.

But in terms of how customers think about it, it is the next generation of networking. The uniqueness of course, is the cooling infrastructure is changing, right? Like I want to see a mix of air cool and liquid cool, right? So that's, I think, an interesting dynamic that customers are still kind of trying to get their hands around because this is breaking new ground for everyone, right? And it's part of the reason why Chantelle talked about new use cases, new products, et cetera, like this is some of that complexity that comes in, right? It's not just about can we ship the product out the door. It's can the customer also be able to absorb that and get value out of it, right? And that's where I think we continue to partner. It's a model that's worked really well for us, and we see no reason to change from that, right?

Yes, the shift to liquid cool data centers. I mean, I didn't see this coming. The mechanical engineering will be so cool again.

It's back.

It is. Let me just wrap it up here. But let's talk briefly about Campus and your progress in Campus and WiFi a little. Can we touch on that briefly?

Yes, I would love to. I think that if you look at -- I think last year was our first real external kind of accountability model, and we hit that revenue of $800 million. We raised the target this year to $1.25 billion. So 55% growth for ourselves with you guys to hold ourselves accountable. Super excited. You've heard others talk about, there's a great refresh cycle happening in the next couple of years. We're definitely participating. We're winning in campus, not even being in the data center yet, which validates our portfolio. Our proof of concepts are very well received. I think that -- so we're just -- we're super excited with campus. Our portfolio is there to meet their needs, including this agentic AI conversation.

Rudy, anything you wanted to add on the -- or John? Because I know, we're out of time.

That 55% in context, right? I think it's a market that's growing single digits. We are starting from a low base. I mean our share is probably in the 3%, 4%, 5% range today. But I mean, that is just tremendous growth. So I think just as much as we're excited about the AI opportunity, I think the Campus opportunity is incredibly exciting, especially because there's far more customers building Campuses than there are customers building AI, right? So it's a different selling motion. It's a little bit slower of a ramp, et cetera. But it is a long game, and we're in it for sure.

That is great. Well, super. I mean, I really appreciate you guys joining today. Any last comments you wanted to make Chantelle to investors just wrapping up where Arista is delivering?

We raised our year twice in the last 2 quarters. We had 55% across the P&L in deferred revenue, exceeded EPS expectations, and so we'll continue to deliver value to you guys. And hopefully, you're as excited as we are in the demand that we've pervaded to you today.

For sure. Well, thanks for joining. Really appreciate it.

Thank you. Have a good day.

Bye.

Bye.

Welcome to the First Quarter 2026 Arista Networks Financial Results Earnings Conference Call.

[Operator Instructions]

As a reminder, this conference is being recorded and will be available for replay from the Investor Relations section on the Arista website following this call. Mr. Rudolph Araujo, Arista's Head of Investor Advocacy, you may begin.

Thank you, Regina. Good afternoon, everyone, and thank you for joining us. With me on today's call are Jayshree Ullal, Arista Networks Chairperson and Chief Executive Officer; and Chantelle Breithaupt, Arista's Chief Financial Officer.

This afternoon, Arista Networks issued a press release announcing its fiscal first quarter results for the period ending March 31, 2026. If you want a copy of this release, you can find it on our website. During the course of this conference call, Arista Networks management will make forward-looking statements, including those relating to our financial outlook for the second quarter of the 2026 fiscal year, longer-term business model and financial outlook for 2026 and beyond, our total addressable market and strategy for addressing these market opportunities, including AI inventory management, lead times and product innovation, which are subject to the risks and uncertainties that we discuss in detail in our documents filed with the SEC, specifically in our most recent Form 10-Q and Form 10-K and which could cause actual results to differ materially from those anticipated by these statements.

These forward-looking statements apply as of today, and you should not rely on them as representing our views in the future. We undertake no obligation to update these statements after this call. This analysis of our Q1 results and our guidance for Q2 2026 is based on non-GAAP and excludes stock-based compensation expense, intangible asset amortization, gains, losses on strategic investments and the income tax effect of these non-GAAP exclusions, including the recognition of direct access tax benefits associated with stock-based awards. A full reconciliation of our selected GAAP to non-GAAP results is provided in our earnings release.

With that, I will turn the call over to Jayshree.

Thank you, Rudy, and welcome, everyone, to our first quarter 2020 earnings call. Arista has experienced significant velocity in all our sectors in Q1 and are now commanding the #1 market share in high-speed switching in the greater than 10 gigabit Ethernet category. With that, we have overtaken many incumbent vendors according to major market analysts for 2025. Our cloud and AI networking strategy for diverse AI accelerators continues to gain traction. Unlike typical workloads, AI workflow patterns can be long-lived elephant flows or short-lived and simply not predictable. This implies careful attention to performance where a flow can cause [indiscernible] for a long duration of milliseconds. The intensity of a flow can determine the line rate throughput, the shifting traffic patterns to massive flows synchronized to all in all, or all reduce or bursts of collective communication are all important for AI training and inference applications.

I would like to take a moment to review our 3 AI fabric use cases. In scale-up mode, we have familiar technologies such as NVLink and PCIe that have enabled vertical scaling of single compute nodes or REX. The advent of E.SUN Ethernet for scale-up networking specifications allows for increasing or decreasing computing power in a flexible manner with Ethernet to automatically adapt to workload demands. Scale-up will be a new entry for Arista in 2027 and beyond, where we will be working closely with our customers to build AI racks with very fast interconnects for co-packaged copper, CPC, or open co-packaged optics, CPO, as well as supporting collectives and memory acceleration.

Scale-out or horizontal scaling involves adding more machines to a [indiscernible] spine fabric, moving workloads across multiple servers or nodes or even connecting other elements like storage or CPUs. As you scale up [indiscernible] with massive data sets, bottlenecks can be resolved with collective and protocol acceleration at L2, L3, cluster load balancing, all at wire rate. The system must deliver consistent performance without degradation as more nodes participate. Arista is a shining example here with greater than 100 cumulative customers to date in 800 gigabit Ethernet deployments, and we expect the addition of 1.6 [ terabyte ] in 2027 at production scale.

Scale across drives across the cloud and AI as the AI accelerators in a location may need to be distributed to achieve the appropriate bandwidth capacity with the optimal power. As workloads become more complex and more distributed, the bi-sectional bandwidth must scale smoothly to avoid bottlenecks and preserve performance. This demands sophisticated traffic engineering, deep routing, encryption properties and integrated optics based on Arista EOS stack and using Arista's flagship 7800, R3 or R4 series. The 7800 has established itself in this category as the premier scale across choice.

You can see with Arista's accelerated networking strategy and these 3 types of AI fabrics. These are critical to deployment of diverse accelerators and frontier models. Traditional static network topologies with hotspot jitter that slows down job completion time or increases time to first token for inference are all not the way to go. Arista's Ethernet portfolio addresses both the synchronous flows for massive training and the low latency for concurrent swarms of real-time inference in this era of trillions of tokens, terabits of performance and terawatts of power. In 2024, you may recall, we discussed 4 Ethernet-based AI training deployments. And of course, since then, we've expanded and exploded to countless others. This fourth customer from the group has officially moved from InfiniBand to Ethernet at production scale over the last 2 years. The high-speed Ethernet AID spine with flexible air or liquid cooled infrastructure overcomes the physical constraints of power and space for AI workloads.

It results in a low latency distributed AI supercomputer fabric across global regions. What is clear to me and us is our networking progress with data, control and management and multiplanar orchestration is not only central to our AI searching performance, but also important for high-speed optics transmission. At the recent Optical Fiber Conference, Arista unveiled its extended pluggable optics, XPO form factor designed specifically for optics innovations at high speed. Now endorsed by greater than 100 vendors, salient features include record-breaking throughput, delivering 12.8 terabits per pluggable module, unprecedented rack density achieving 204.8 terabits per OCP rack unit, integrated cold plate capable of cooling up to 400 watts power per module and the universality and flexibility across a range of pluggable optics, copper as well as linear halftime our retimed interfaces. A special kudos to Andy Bechtolsheim, Arista's Chief Architect for driving from OSFP 10 years ago to this next-generation XPO, bringing structural improvements in power, footprint and cost reductions. Our enterprise business experienced strong results in Q1 2026, both in data center and campus. Our VeloCloud acquisition is also integrating well into our branch and campus strategy, bringing more distributed enterprise use cases and a new channel motion with managed service providers, MSPs. To share some recent wins, let us hear now from Todd Nightingale and Ken Duda, our Co-Presidents, to delineate our Arista 2.0 Centers of Data Strategy. Over to you.

Thanks, Jayshree. Arista is diversifying this business with new customer acquisitions covering a broad set of use cases, all unified by Arista's EOS stack and its ability to modernize enterprise infrastructure operating models. Our first highlighted win is a new cloud AI network. The customer was constrained by an incumbent white box architecture that simply could not keep pace with the massive scale-out requirements of AI. Arista was selected as a commercially proven and reliable scale-out architecture with unmatched stability of EOS and the ability to connect AMD MI Series X [indiscernible]. Arista's AI leaf and spine Etherlink products were deployed at 800 gigabits to provide the incredible performance, modern AI networks require. The AI fabric was tuned using Arista's cluster load balancing to scale out the thousands of XTUs minimizing hotspots and congestion. On the software side, the customer leveraged AVD, Arista's validated design framework, to automate network provisioning, which both reduces the total cost of ownership, but also provides an easy path to reliable network deployment at scale, where without AVD automation, a small mistake can cause precious days of debugging time. This was a strategic neo-cloud win with large potential for upside growth in an area where we are seeing enormous opportunity and velocity in both new cloud and soften cloud customers.

Our next win is in the service provider sector with a leading regional fiber-to-the-home provider serving hundreds of thousands of subscribers. As subscriber bandwidth demands have surged, this customer realized their legacy routing architecture was too rigid, too brittle and too costly to scale. They needed a solution, which would modernize their next-generation backbone and Internet puring edge. Arista won this upgrade by proving an automation-first approach with a modern operating model, driving operational savings and increased subscriber reliability. On the hardware side, we deployed [indiscernible] 7280 routing platforms using EOS's FLX capabilities, which unlock deep buffering, a rich control playing software stack and full Internet route scale. On the software side, Arista's AVD framework, again, automates router provisioning to reduce the time it takes to turn up services while also reducing errors. Here, we saw great results from our technology partnerships with Palo Alto Networks, ensuring the routing edge integrated securely and seamlessly with our overarching security architecture. And here, Arista's core value proposition of lower operating costs and greater reliability to grow the competitive win. Now I'll hand it off to Todd.

Thanks, Ken. Our third win is in the insurance services sector. Following a year of strategic collaboration, the customer wanted to modernize their infrastructure with a streamlined, automated foundation capable of delivering granular real-time insights to secure and monitor critical applications. Here, observability was truly the key. Arista secured this comprehensive win after executing a flawless proof of concept proving our architecture significantly exceeded operational standards. To achieve deep network observability the customer deployed our R3 series for filter and delivery roles on our monitoring fabric DMS. Additionally, they deploy campus switches to radically simplify out-of-band management. Leveraging rich telemetry capabilities of EOS, the customer unlocked advanced features like VXLAN header stripping and transition to a fully automated declarative operational model.

Our final win is within the manufacturing sector where we're seeing amazing momentum. Here, we have a customer operating more than 100 factory sites globally, servicing consumer, health care, aerospace, defense and AI infrastructure customers. This was a true mission-critical use case and their legacy campus network have become the bottleneck for achieving real 24/7 production. Shifting traffic patterns, manual provisioning and importantly, a lack of visibility and forensics into microbursts and drops for keeping them from achieving their goals. Arista won an extensive bake off against 2 established vendors, both of whom proposed campus design that could not match what Arista delivered, a universal leaf spine campus based on open standards running a single EOS binary across campus, data center and WAN. The Cognitive Campus solution leveraged 100-gig campus spine, high-powered POE leaves and Arista WiFi 7. CloudVision drove provisioning, configuration and life cycle end to end with consistent tooling across the network infrastructure. Here, it really was Arista's modern operating model that drove differentiation in the engagement, hit list production upgrades, latency analyzer for microbes visibility and true packet drop forensics.

The teams were able to significantly reduce production impacting maintenance windows and exposed events that had previously caused lying interruption. In all 4 of these examples, Arista's support team stood out to customers for its best-in-class service, well known for troubleshooting issues with customers long after Arista gear is no longer suspected to be at fault. Arista's modern operating model also played a key role, especially the AVD tooling that Ken mentioned, for architecture, validation and deployment. We're excited about the momentum across the entire enterprise business and especially the diversification that it brings to Arista Thanks, Jayshree.

Thank you, Todd. Thank you, Ken. It was so fantastic to hear of happy customer outcomes. We had another fitting example of that at our Innovate 2026 event here in the headquarter facility held in March. The energy and enthusiasm of our greater than 250 customers who attended was truly [ infectious ] and inspiring. I want to especially give a shout out to Ashwin Kohli and Divia Wagners teams who have already improved our outstanding Net Promoter Score from 87 to 89 ratings translating to a 94% customer approval. This really exemplifies the lowest security vulnerabilities in the tech industry. It enhances our ability to better cope with the many risks that AI is creating.

As I look ahead at the year, our Arista 2.0 momentum continues to march on and resonate. Our demand is actually the best I've ever seen in my Arista tenure. The supply, however, is a slightly different and opposite tail. We are experiencing industry-wide shortages across the board, be it wafers, silicon chips, CPUs, optics and, of course, memory that I referred to last quarter, coupled with elevated costs to procure these.

Clearly, our demand is outstripping our supply this year. While we hope the supply chain will the next year or 2, the Arista operations team has been diligently engaging with our vendors in strengthening supply agreements and engaging in multiyear purchase commitments. We anticipate gross margin pressure due to mix and trade-offs we are making to pay more to assure supply continuity to our customers. Nevertheless, it gives us confidence to increase our forecasted growth slightly to 27.7%, aiming now for $11.5 billion for 2026. We also increased our AI target now to $3.5 billion this year, thereby more than doubling our AI sales annually. And with that good news, over to you, Chantelle for the financial details.

Thank you, Jayshree. I continue to be impressed by our company's ability to deliver such a breadth and depth of networking innovation. It is a core tenet that underpins our strong financial return to shareholders. [indiscernible] want to detail our most recent financial outcomes. To start off, total revenues in Q1 were $2.71 billion, up 35.1% year-over-year and above our guidance of $2.6 billion. Growth was seen across the customer sectors led by our AI and specialty providers customers within the quarter. International revenues for the quarter came in at $418.9 million or 15.5% of total revenue, down from 21.2% last quarter. This quarter-over-quarter decrease was primarily influenced by Americas-based sales to our large global customers. The overall gross margin in Q1 was 62.4% within the guidance range of 62% to 63% and down from 63.4% in the prior quarter. This quarter-over-quarter decrease is due to the lower mix of sales to our enterprise customers in the quarter. Operating expenses for the quarter were $396.8 million or 14.6% of revenue, down slightly from last quarter at $397.1 million.

Our R&D spending came in strong at $271.5 million or 10% of revenue, despite a slight sequential decrease due to the timing of new product introduction costs, Arista continues to demonstrate its commitment and focus on networking innovation. Sales and marketing expense was $103.5 million or 3.8% of revenue, down from 4% last quarter, representative of the highly efficient Arista go-to-market methodology. Our G&A costs came in at $21.8 million or 0.8% of revenue, down from $26.3 million last quarter, reflecting our strong base cost productivity within a pure-play networking business model.

Our operating income for the quarter was $1.29 billion or 47.8% of revenue. Let me pause here to thank the greater Arista team for all of their efforts and resulting excellent execution in a dynamic environment. Other income and expense for the quarter was a favorable $110.8 million, and our effective tax rate was 21.1%. Overall, this resulted in net income for the quarter of $1.11 billion or 40.9% of revenue. Our diluted share count was 1.27 billion shares, resulting in a diluted earnings per share for the quarter of $0.87, up 31.8% from the prior year.

Now turning to the balance sheet. Cash, cash equivalents and marketable securities ended the quarter at approximately $12.35 billion. In the quarter, we did not repurchase our common stock. Of the $1.5 billion repurchase program approved in May 2025, $817.9 million remain available for repurchase in future quarters. The actual timing and amount of future repurchases will be dependent on market and business conditions, stock price and other factors.

Now turning to operating cash performance for the -- for the quarter, we generated approximately $1.69 billion of cash from operations in the period, the strongest in the history of Arista. This was driven by a robust earnings performance, coupled with an increase in deferred revenue. DSOs came in at 64 days, down from 70 days in Q4 due to the linearity of shipments within the quarter. Our inventory turns improved slightly, landing at 1.7% versus 1.5% in the prior quarter. We ended the quarter with $2.38 billion in inventory, up from $2.25 billion last quarter. This marginal increase is a calculated investment in the mix of raw materials to fulfill our growing demand.

Our purchase commitments at the end of the quarter were $8.9 billion, up from $6.8 billion at the end of Q4. As mentioned in prior quarters, this expected activity mostly represents purchases for chips related to new products and AI deployments. We will continue to have some variability in future quarters as a reflection of the combination of demand for our new products, component variability and the lead times from our key suppliers. This could also result in quarters of elevated inventory balances ahead of the deployments. Our total deferred revenue balance was $6.2 billion, up from $5.37 billion in the prior quarter.

The majority of the deferred revenue balance is product related. Our product deferred revenue increased approximately $643 million versus last quarter. We remain in a period of ramping our new products, winning new customers and expanding the use cases, including AI. These trends have resulted in increased customer-specific acceptance clauses and an increase in the volatility of our product deferred revenue balances. As mentioned in prior quarters, the deferred balance can move significantly on a quarterly basis, independent of underlying business drivers. Accounts payable days were 54 days, down from 66 days in Q4, reflecting the timing of inventory receipts and payments. Capital expenditures for the quarter were $54.5 million. We continue the construction work to build expanded facilities in Santa Clara. In Q1, we incurred approximately $40 million in CapEx related to this program and estimate it will reach $180 million in 2026. These Q1 results have provided a strong start to our fiscal year 2026. As Jayshree mentioned, we are now pleased to raise our 2026 fiscal year outlook to 27.7% revenue growth, delivering approximately $11.5 billion. We maintain our 2026 campus revenue goal of $1.25 billion and raise our AI fabric school from $3.25 billion to $3.5 billion.

I would like to take this opportunity to remind the audience that the timing and outcome of customer projects with acceptance terms can create quarterly and sequential dynamics that do not follow prior year trends. For gross margin, we reiterate the range for the fiscal year of 62% to 64%, inclusive of mix of anticipated supply chain cost increases for memory and silicon. Given this challenging supply backdrop, I am proud of our sourcing team's execution, which strongly contributes to the gross margin outlook holding in our guidance range. We feel confident that we can source the necessary supply to meet our customers' needs. Our operating margin outlook remains at approximately 46% for the fiscal year, with the tax rate expected at 21.5%. On the cash front, we will continue to work to optimize our working capital investments with some expected variability in inventory and cash flow from operations due to the timing of component receipts on purchase commitments. More specifically, now our guidance for the second quarter is as follows: now with the added quarterly metric of diluted earnings per share, revenues of approximately $2.8 billion, gross margin between 62% and 63%, operating margin between 46% and 47% and diluted earnings per share of approximately $0.88 with approximately 1.27 billion diluted shares. Our effective tax rate is expected to be approximately 21.5%. In closing, we are optimistic about the fiscal year ahead. The industry has many times demonstrated the pattern of landing on Ethernet is the winning technology, and that is where Arista shines best. We appreciate our customers' choice of working with us to achieve their business outcomes. Now Rudy, back to you for Q&A.

Thank you, Chantelle. [Operator Instructions]

Thank you for your understanding. Regina, please take it away.

[Operator Instructions]

Our first question will come from the line of Simon Leopold with Raymond James.

Great. I wanted to explore your commentary around the scale across opportunity in particular. And I guess what I'm trying to get a better sense of is how much revenue, if any, did that contribute last year? And how material is that to the $3.5 billion forecast you're giving this year? And how should that trend longer term?

Sure, Simon. I think last year, on scale across, we were just beginning. So I think they were small numbers. And majority of the numbers were really scale out. That's sort of our heritage and that's where we excel. If I were to anticipate how it would be this year, again, scale up is virtually 0 and nonexistent because it really only comes to play after the ESN spec. So consider that more a [ 27, 28 ] kind of number. So I think the number will be really shared between scale across and scale out. I don't know if I can say it's 50-50 or 70-30 or 60-40 but scale across will definitely contribute at least 1/3 of our AI number.

Our next question will come from the line of George Notter with Wolfe Research.

Maybe just continuing the discussion on scale up. We are starting to see rack design wins. One of your competitors in the ODM space, I think, has got a couple of designs that they've announced at least. And I know you're kind of pointing towards ESON as being kind of a key catalyst in generating business there. But can you talk a little bit about where you are in terms of designs with customers, progress. Anything you can tell us there would be great. And in fact, I think a few quarters ago, you said you had 5 to 7 scale up rack designs that you were at least working on. I'm just wondering if you can update that.

Yes, that's correct, George. I think there is no doubt in our mind that we will have a number of racks and number of scale-up use cases in 2027. Maybe some of them will be in early trials, but majority of them are looking at really starting with 1.6T and 1.6T chips will really happen in 2027. There may be a few, a handful of them that tried some experimental stuff at 800 gig. But we continue to see at least 5 to 7 rack opportunities. Some of them are multiple racks with the same customer. We're actively designing with them. There's a huge amount of liquid cooling designs with very dense cabling options, acceleration of collectives and memory features we have to work on for low latency. So I definitely feel we're an active engineering phase with Ken and Hughes teams this year.

But unlike the ODMs, I think we're held to a higher bar and we have to just make sure that this thing is production worthy and specification adhering to ESON. So I would say today's scale-up is mostly limited to NVLink from NVIDIA and maybe some PCI switching but majority of the Ethernet scale-up will only really happen in '27 and '28.

Our next question will come from the line of Antoine Chkaiban with New Street Research.

So with the supply outstripping demand, I'm wondering how much does your current supply allow you to grow this year and next. If you update as top line growth guide of 28% growth, a good reflection of how much supply you've secured for this year? And what could that number look like next year based on how much supply you think you can get as of today.

Antoine, I think the supply chain problem, and Todd, maybe you can add to this, is not a 1- or 2-quarter phenomena. We now think it's a 1- or 2-year phenomena. When you -- at first, we thought it was memory. Now it's all the wafer fabrication facilities. Every chip is challenged, and you can see how Chantelle has leaned in with the purchase commitments for multiple years. So while we will continue to improve it, this is a reflection of not just demand, but how much we can ship this year. And as we continue to ship this year, we can give you better visibility on next year. But I can just tell you, we see multiyear demand, and we are going to do everything, including hurt our gross margins to supply to that demand this year and next year because we believe that we certainly don't want to keep GPUs idle and AI infrastructures underutilized because Arista didn't supply the network.

So can the number get better this year? I think this reflects our best attempt at a good number. We started out -- we started at 20% -- 25% growth? Yes. So we started out to 20%, we are 25%, now at 27.7%. Could we improve to the tail end of the year? We'll see. But the amount of de-commits we're seeing doesn't feel good. So we think a lot of this will continue into next year and keep us constrained for the next couple of years.

Our next question will come from the line of Aaron Rakers with Wells Fargo.

Jayshree, last quarter, you had alluded to kind of engagements with other hyperscale cloud titan customers? I think you also pointed to maybe having 1 or 2 new 10% customers this year. I'm curious of where we stand today? Any updated thoughts on adding 1 or 2 new customers at 10% plus? And maybe qualitatively, just talk about your engagements you're having beyond your 2 big cloud titans across the hyperscale vertical.

Yes, absolutely. First of all, 2 big ones. We never take them for granted. Microsoft and Meta, there are all-time favors. They've been on 10% and greater customers for over a decade. And the partnership could never be stronger, and it continues to get better both in cloud and in AI. In terms of the new entrants, we still expect at least one, maybe two and maybe I should caveat this by saying, certainly, in demand, we see 1 or 2. We shall see Todd, how we do on shipments to see if we can achieve the greater than 10%. The 2 of them have very interesting characteristics. They exhibit what I would call the 3 use cases I just alluded to, scale up, scale out and scale across where we really have a fabric notion of creating -- so far, we've been working with them a lot on the front end, and now we get to complement that on the back end, definitely for scale out and scale across and maybe even a little bit of scale up in some of these use cases.

The other thing we're seeing with a lot of these use cases is the lack of power insights and the ability and demand to distribute and get a more multi-tenant scale across is very high in these 2 use cases. A third common thread, we're seeing across of them, much as we all talk about ODM and white boxes, they deeply appreciate the U.S. and the features and the reliability and the observability and the -- just the fact that we have a robust, highly scalable Layer 2, Layer 3 stack commands a lot of superior advantages. So I believe the diversity of these cloud titans is largely due to the fact that we have great hardware and software combined. Ken, you want to say a few words on that?

It's just been an incredible journey to live through this and see the level of infrastructure that all we're getting and how well positioned our hardware and software road maps are to address these ever evolving more [indiscernible] use cases. [indiscernible] to get work on the store.

That's always fun when your job is a blast. So [indiscernible], I still see 1, maybe 2 10% customers. And Todd, hopefully, we can ship it -- sorry, Aaron.

And our next question will come from the line of Ben Reitzes with Melius Research.

There you go, Jayshree, here I am. So yes, I wanted to ask around constraints. Are you able to say what the number was in the quarter and what it's taking away in terms of the $2.8 billion guide? Is it safe to say things would have been $100 million or $200 million higher for both? And then if you don't mind, just if you can touch on why the gross margin should go back up to 63. What is it that you guys are doing that at gives us confidence that it can actually expand a tad from here?

Yes, I think I'll just -- Chantelle, [indiscernible] I think that -- I don't think the commentary about the demand outstripping the supply as of Q1, Q2. I think we're talking about looking ahead Q3, Q4 into next year. So I don't think there's something outside of what we've guided what we've delivered in the first half. I think in the sense of the margin. So the margin is a mix of things, right? And I think that all the team members are executing in full force. I think the supply chain is doing everything they can on ensuring that we have the best supply at the best price. And so we've incorporated that I think that the mix of customers, the only chance for mix expansion or margin expansion would be due to mix. And so I think that's the opportunity as we look to see what we can deliver in the second half then. I think that would be the opportunity.

The teams are also doing everything they can to make sure we control our costs, especially on the manufacturing side, and that includes bringing on secondary providers, calling new components, et cetera. to make our supply chain more resilient and more cost effective in [indiscernible].

And one thing to clarify also on gross margins. So we view this as a partnership with our customers. So while we would consider and have raised prices a little bit, unlike our competitors, we haven't done 2 pricing ceases. We haven't done major price increases. And the price increases really come into play once our backlog starts to reduce, right? So you won't see the impact of that. So our gross margins are a strong factor of cost going up and are still eating a lot of the costs and giving our customers the benefit and promise of the pricing we said we would give to them.

Our next question will come from the line of Michael Ng with Goldman Sachs.

I was just wondering if you could talk about whether or not Arista is seeing networking attach opportunities for customers that are using TPU or TPU like architectures. And then -- anything you could comment about as it relates to growing Neo cloud traction. Is that something that you think may be a little bit underappreciated by the analyst community.

Yes, Michael, you're absolutely right. I'll take your second question first. It's easy to talk about the titans because the numbers are so ginormous, right? But the new clouds are a very important sector because they don't always have the staff to do everything they want to do, and they really lean on Arista's design expertise, U.S. expertise, network design configurations we can provide them a family of 22 products we have in AI. So yes, I would agree with you. It's an underappreciated and the neocloud was very strong this quarter if I recall, Chantelle, for us in the specialty and cloud providers. What was the other question? You had 1a, 1b?

The TPU.

Yes, the TPU. So in general, we are seeing diverse accelerators. Last time I spoke about the AMD accelerators. This time, I will definitely give a nod to the TPUs because in particularly scale across use cases, we're seeing multi-tenants connecting to different AI accelerators, including TPUs as well. So I think the diversity of accelerators is creating tremendous multi-accelerator opportunity and multiprotocol features that we can provide for them in our network.

Our next question will come from the line of Sean O'Loughlin with TD Cowen.

Great. Congrats on the results, and thanks for letting me joining on the fun here. Jayshree, I wanted to get your thoughts on -- we've been talking a lot about agentic AI and the demands that it's placing on maybe some of the more general purpose infrastructure that we has been maybe in the background over the last couple of years. You've talked in the past about a 2:1 pressure on front-end networking created by back end. First, I guess, is that still the correct way to think about it? And second, as a genic workflows become more common, is there any additional demand from your perspective, having a single-image EOS platform on the front and the back end? Or is the front and back end still pretty siloed.

Yes. Well, first of all, Sean, welcome to your first call. It will be fun. Join the fun. So agentic AI, it's kind of a buzzword, but let me sort of break it into how -- the biggest killer application we see in agentic AI right now is still training. And indeed, it's going to move to more distributed inference. And we'd also like to see agentic AI move into a lot of enterprise use cases, all of which we're seeing, by the way, but I would say large, medium, small, the largest killer agentic AI application is training, the medium is enterprise and the smallest -- medium is inference and the small is obviously enterprise. The -- in terms of back end versus front end, we are now seeing way more back-end activity, particularly with our large AI titans and cloud titans because there is just so much scale they need to prepare for the billions of parameters and tokens, and this is where a lot of -- so much so that I think the front end, they might come back and refresh, but they're almost ignoring right now in favor of the back end.

Having said that, though, by virtue of the back-end deployments, I don't know if we any more see a 2:1 to the front end, but we at least see a 1:1. And the 1:1 can be wide area, CPU and storage. Those are probably the 3 common use cases. Not all the customers are up and lifting everything and doing all 3, although we've had cases where some of them did an upgrade at the front end before they went into the back end. But usually, they will have to come back to that because the minute you put that kind of performance pressure and scale on the back end, you almost have to do something in the front end. But at the moment, I would say it's more one-to-one. And at the moment, I'd also say the scale across in the back end has become a bigger use case than we imagined this time last year.

The other thing I had to mention here is just how good it feels to be have the same set of products in the same common operating system management suite and operating model across the front end and back-end. This lowers cost for the customer, simplifies their design process to get that leverage, and we're one of the few vendors who can do that.

I think only.

I think so.

I think only. Yes, absolutely. Good point, Ken.

Our next question will come from the line of Meta Marshall with Morgan Stanley.

Appreciate the question. Maybe just a question on XPO monetization or just how it helps you kind of continue to gain share with customers or just mind share with customers by being so front footed with the technology.

Yes. Thank you, Meta. I think as you know, we're not a classic optics vendor. But almost always, whenever we are selling our switches, you have to connect to something. And usually it's some form of copper or optics. So and these innovations with OSFP, I remember this super well where everybody was saying, "Oh, no, no, we can just use QSFP, has proven to be not only a contribution for Arista, but really for the industry-wide." And that's still how we see it with XTO as well.

While the industry has been talking a lot about copackage optics, these are still science experiments, and they're very proprietary with individual vendors doing their own thing. We embrace open CTO a few years from now, but we think XPO has a 10-year run, especially at 1.6T and [ 3 point T ] where you need liquid cooling and you need that kind of capacity. So all the scale up racks we're talking about wouldn't be possible without XPO or CPC or any one of those technologies. So we see this as just as the last decade was greatly influenced by OSFP. The next decade will be greatly influenced by XPO. And remember, 99% of the optical market today that we connect to is all pluggable optics. So this is a very crucial invention and innovation, not just for Arista, but the industry at large.

I think it's a great example of how Arista enables an ecosystem and then we profit as that ecosystem grows. And with XPO and [indiscernible] is a standard interop [indiscernible] way to get to 4x of [indiscernible] density in liquid cooling, which is absolutely critical for these AI use cases. Without that, you have this huge bottleneck at the front panel, the amount of extra rack stasis required to get through OSFPs. It's -- so we're really enabling the future growth of our industry this way, which we benefit and other benefit as well.

Yes. It's stunning to me. I remember, when I first talked to Andy and Vijay, they said, "Oh, we think we'll get about 20 signatures", and then it was 40, and now it's north of 100. So it tells me the whole consortium is coming together for things like Ethernet, IP and standardization of optics.

Our next question will come from the line of Tal Liani with Bank of America.

I promised [indiscernible] to be nice today. So I have a good question for you.

I promised to be nice too.

Deferred revenues. Deferred revenues doubled in the last year. And it went up -- if I combine short term, long term, it went up $826 million. It went up significantly in the last 4 quarters. What needs to happen? What are the conditions for -- to recognize deferred revenues, meaning what needs to happen for deferred revenues to be recognized over the next few quarters? Is it about data center going live in traffic goes into data centers? Or what are the sources for the deferred revenue increase?

Right, right. So I really do like you. So I'm going to be nice to you not because I have to, because I like to. So I think if you remember 10 years ago, Tal, we had a similar phenomena where in the cloud, the whole Leaf spine design was brand new, nobody really knew how to build it or monetize it and we were building some of the world's largest networks for Azure, et cetera, right? And we had new products, they had new designs. They had done traditionally the access aggregation core and we're now moving to the flat pack topology. And we had some fairly lengthy qualification cycles. So I would say there's a customer aspect of it and a product aspect to it. The customer aspect to it is they need to have the space, they need to have the facilities. They need to have there. In this case, GPUs now back and then it used to be CPUs, they got to have their rack and stack. And in many cases, by the way, we're running into examples where they -- it's literally they need to manually install the cables, and that takes several months, right? Thousands of people have to do that.

So there's certainly a customer acceptance piece of it, which starts with being ready. There's also a new product. Many of these new products in the Arista ethylene family, particularly for the AI are brand new, brand-new chips, brand-new software, the familiarity with it, particularly in the back end was scale out and scale across is new to them. So there's a level of testing and level of making sure it works with the rest of their ecosystem, including the front end that is super important, and Arista bears a huge responsibility to that as well. So all just to tell you that the length of time to qualify this, which used to be 2 to 4 quarters has extended more like 6 to even 8 quarters. So it's gotten much longer. Chantelle, do you want to add something?

Yes. The other thing -- thank you, Jayshree, is that we do recognize some of it every quarter. So it's not like it's on balance, this is aging in [indiscernible]. We recognize things every quarter, things come in and things are recognized to the P&L. So I just want to make sure you understand that the...

It's not piling -- some things go in and some things come up. Yes. Does that make sense, that? Tal, you're on mute?

No, no.

[indiscernible].

We can go to the next question.

Our next question will come from the line of Amit Daryanani with Evercore.

I guess, Jayshree, you folks have kind of positioned XPO as the next OSFP. And I'd love to kind of understand that XPO ramps from the OFC demos to potentially deployments in '27, how do you see change in the optics architecture within AI clusters? And then maybe specifically Arista -- does that change the growth profile or your content per AI IRAC or cluster as we go forward?

Yes. Thank you, Amit. I think you should look at XPO as a partner to OSFP. So at 400 gig and 800 gig you'll be fine with OSFP. And as we go to higher speeds in '27, '28 or even beyond, OSFP will run out of steam, and this will be the new connector of choice. So the migration to higher speeds equals the migration to XPO, particularly for scale out and scale across. Within a rack and scale up, there's still a number of choices. I think within short distances of 2 to 3 meters, you're still going to see a lot of co-packaged copper and I think XPO in terms of density will be another alternative. But I don't rule out open CPO as well over there. They're really looking to maximize dedicity in a minimum amount of space. So I think XPO will be particularly prevalent in scale out and scale across and will be 1 of the choices in scale-up.

Our next question comes from the line of Ryan Koontz with Needham.

This is Jeff Hopson on for Ryan. I appreciate the question. On the scale cross, it seems like that would be a really good fit for all Arista's capabilities. And I know you mentioned it would maybe be around 1/3 of revenue this year. But is this something where scale across could even be larger than scale out over the next couple of years?

Ryan, or rather Jeff, I think the answer to that would lie on how well we do with both. And what form factors are used for both. So majority of the scale across today is a very premier valuables, heavy-duty routing platform, the 7800. So if we do lots of that, it could get well beyond the 30%. But some of them may do it with fixed boxes, too or fixed switches and choose to add a lot of cable in which case, it wouldn't go well above that. So we don't know what we don't know. But I would agree with you that scale across is by far the most significant and differentiated opportunity that really highlights Arista's prowess in both platforms and software.

Our next question comes from the line of Samik Chatterjee with JPMorgan.

Jayshree, maybe slightly related to the last question here. Just trying to think about, you said most of the cloud revenue near-term is going to be scale out and scale across as we wait for scale up to ramp. How are you thinking about your market share when it comes to scale out versus scale across in the early days of scale across, what are you seeing in terms of market share? And are you seeing customer decisions being led in scale across by sort of the incumbent and scale-out or -- is it a different decision altogether in terms of how they're designing vendors for in-force scale across.

Good question, Samik. You're making me think. So I would say If it's greenfield deployment, then they tend to think of it together because they're not only building the sites, but they're thinking of the interconnect across them. And therefore, market share is generally strong in both. In some cases, where Arista has not been a historical participant within the data center, we now have an opportunity to offer the scale across multi-tenant even in a non-greenfield situation and let's say, in a brownfield, where now they've got disparate data centers or AI clusters that we now have to bring in. And so once again, I think Arista's really fitting example to be in scale across both those use cases, but has the additional opportunity in a brand-new data center to be in all use cases, if that makes sense.

So it's giving us a chance to participate with different types of accelerators and different types of models because people aren't getting the power and they're having to distribute the data centers. And as a result of distribution, you need more traffic engineering, routing, multi-tenancy. So I would say scale across is the common denominator in all our use cases and scale up and scale out maybe nice options and brand-new greenfields.

Our next question comes from the line of Karl Ackerman with BNP Paribas.

Jayshree, you are doing more networking design today more than ever. Does that change your ability to monetize your services to capture more of the work of the other value that you're adding to these applications? And I guess as you address that, given the large mix of services revenue within deferred, could services revenue accelerate faster and represent perhaps 25% or 30% of sales going forward?

I don't think so, Karl. I think we're a product company, and majority of our revenue generation and interest in Arista as a company for all the designs we're doing comes from our product heritage. And it's not like we charge for services. In fact, we work closely with our partners also, we will recommend network designs. We will support services and certainly, things like we are the gold standard for worldwide support. But I don't expect services as a function of our revenue to go up. I continue to be -- see ourselves as a product-led company.

Our next question comes from the line of Matt Michman with Truist.

I just want to go back to gross margins. So I know we were sort of in that 62-ish range. They dipped about 170 bps year-on-year. And I wanted to dig into whether it was primarily mix related or maybe if you can quantify whether the -- how significant the memory and cost-related impacts were, if there's any color you can provide.

Yes, I think it's a great question. I would say the majority, if you look at -- even if you look at prior quarter or prior year, the majority of the difference is mix of the customers. And just to clarify, our larger customers have have a lower gross margin accretion. And so that mix is the primary driver. And then the secondary, although not as significant would be things depending on the quarter, depending how deferred moving tariffs or the memory cost or the silicon cost depending on the quarter. So secondary driver, but the primary drivers mix of the customer segments.

Our next question comes from the line of David Vogt with UBS.

This is Andrew for David. From a high level, with $2.4 billion almost of inventory and almost 2 years in COGS of purchase commitments, how should we think about the supply constraints and where that inventory and purchase commitments are not satisfactory to meet demand? Where are the holes in your inventory?

I wouldn't say we have holes in our inventory, but we have surging demand, especially on the newest platforms, which of course, is driving our need for the most modern silicon from our providers and it's driving need for an expanded amount of memory even more than we were expecting before the year began. So that's driving us to be a buyer in the market. Luckily, we've got pretty good spending power. We're a very reliable partner in these scenarios. And so we partner closely with these vendors. But there's no doubt that like the newest platforms that we're delivering, especially in the AI space is driving needs of ours in the high end of our portfolio.

Yes. And just to add to that, David, the real hole is lead times. We are experiencing such significant wafer fab shortages that we're not getting the chips in time. So more than a hole, I would just say our purchase commitments are multiyears because we're having to deal with forecasts that are out multiple years so that we get them in time because the lead time of these chips is so long. So I think that's the biggest hole lead times.

Yes, we are experiencing 52-week lead times pretty reliably with reservation needs beyond that, and our customers certainly do not want to wait that well.

Our next question comes from the line of James Fish with Piper Sandler.

Chantelle, maybe for you, the guide raise was primarily all on AI. Are you guys prioritizing these shipments or what's given the hesitancy around sort of the non-AI noncampus at this point and leaving that roughly flat still? And Jayshree, just for you, just as we think about the mix here on gross margin, -- what are you guys seeing in terms of Blue Box adoption now? And are you seeing any sort of net pull-in of demand just given you have a lot of smart customers here and they're very much aware of the supply chain constraints.

Yes. Thank you Thanks, James. I'll start with mine first in the sense of the order of your question. So I don't think we're saying because we're raising the revenue and a tribute to AI that we're not excited about all the other customer segments. I think you heard both Jayshree and I talk about -- we're very happy with how the year started, what we're seeing across all 3 customer segments. We're very happy what we're seeing in enterprise, which I wouldn't say is quite AI yet. So let's cover that as a non-AI bucket that you referred to. So wait and see, we're in Q1, reporting Q1. We'll see how the year goes. But we're very confident across all 3 that we're seeing strong demand. So I think I would leave it in the sense of let's see where we get to in our future quarter guidance. Jayshree?

And I would agree with that. Just to remind everybody, we've raised now from $10.5 billion or whatever we said last September to $11.5 billion. And yes, a high degree of that is AI, but we have aggressive commitments on the campus to go to a $1.25 billion quarter and continue to service and grow our data center and cloud as just as well. So all 3 are growing, but certainly, AI is taking the news headline. Regarding Blue Box adoption, one of the customer use cases you actually heard about was moved from -- that you heard from Ken moved from white box to Blue Box. And their goal right now is their desire to move to blue boxes, it works, number one. It scales too. It actually does the job for us with AMD Accelerators, number three. And down the road, they may use open operating systems, but they were very pleased with the diagnostics capability, the platform, SDK, where we literally rewrite every piece of software and bit twiddle all the Broadcom chip transistors very, very well and the EOS features. Down the road, they may use some open losses as well, but that would be a really good example of a blue box that has U.S. today and may go down to other losses.

And we continue to see that, particularly in the Neo clouds. We've always seen a bit of that in the cloud and AI titans because they know how to work with open noses. So we've had that hybrid strategy always, but we're certainly seeing more of that in the Neoclouds now.

Regina, we have time for one last question.

Our final question will come from the line of Ben Bollin with Cleveland Research.

Jayshree, you referenced inference a little bit earlier, so it's kind of a smaller use case right now. I'm interested on your thoughts on where you think enterprise is in terms of their ability to consume inference and create agents and then how that develops over time and where you think the network and edge networks are today and their ability to support those use cases. basically, just do we get the sustained investment period because what you're seeing now bleeds and becomes much more significant in enterprise? And how long lasting that might be?

Yes. No, then I tend to agree with your thesis that while today we are in a training fever, that a more distributed AI generic of AI paradigm with inferences, which means you don't always need the GPU. You're going to have high-end CPUs and you're going to have a smaller set of parameters and tokens to manage and you're going to have specific agentic AI use cases and applications. We're seeing very, very early trials and stages. Nothing super big yet. But we are seeing -- I mean, they're not in the hundreds of thousands of GPUs like you see on the AI titans. But we are frequently seeing our customers in certain high-tech sectors want to deploy clusters that are 1,000, few thousand, definitely not 10,000, but in hundreds of thousands. And they tend to be exactly, as you said, not training, but more inference based, more agentic AI edge influence space as well.

So I think we'll see more of that. This is -- this is the calm before the storm, if you will. And as we -- as the AI gets more distributed, I think it doesn't need GPUs alone, it's going to need more high-performance compute. And many of them seem to feel to us like high-performance compute, HPC use cases. that are sort of getting revived for AI. So I agree with your thesis, Ben. I think it's going to take a couple of years to fully happen.

This concludes Arista Networks First Quarter 2026 Earnings Call. We have a presentation posted that provides additional information on our results, which you can access on the Investors section of our website. Thank you for joining us today and for your interest in Arista.

Thank you for joining. Ladies and gentlemen, this concludes today's call. You may now disconnect.

Thank you for being here. I'll read some boring disclosures to kick off and allow you to open up your chips. So for any research disclosures, please see morganstanley.com/researchdisclosures and reach out to your sales representative with any questions. I am Meta Marshall. I cover the networking space here at Morgan Stanley. We are delighted to have Arista, Jayshree Ullal; and Ken Duda, a new special guest joining us on stage, who is President and CTO as well.

So Jayshree, welcome back. It's been a phenomenal couple of years for Arista since we last had you on stage. Just how do you think the core like value proposition of Arista has changed with kind of AI coming into the framework?

Meta, it's always good to be here. You must be my good luck charm. Every time I come, if we grow like that, I'll keep coming.

The value proposition, and Ken Duda, our Founder and President, started this hasn't fundamentally changed. But of course, the use cases to make it greater has changed. So we always started with the belief that mission-critical networking needs a foundational software, which started in the data center, as most of you know. And we're now the #1 market leader there. Great technology, great U.S., great merchant silicon and just a great product, but more importantly, a great system. We began with this belief that you had to build a great system with our leaf-spine architecture, but then the leaf got plentiful.

Our Universal Spine connects today to campus leaves, enterprise leaves, branch leaves with our Bella Cloud acquisition. And of course, I have to say AI. With the advent of AI, we have been able to build this very unique architecture that no one else in the market has, which is an all-Ethernet, AI spine and leaf that takes advantage of all the goodness that we brought to bear in the data center. So I call this the centers of data era, where whether it's a data center, campus center, WAN center, AI center. We are able to build that uniformity in our network and yet deal with the heterogenity that's coming in front of us with different frontier models and different AI accelerators.

Got it. We get a lot of questions about blue boxes, branded boxes, white boxes, just what do you think that, that conversation -- I'm sure you guys get many of the same questions, like what do you think that, that misses about the value proposition of what people get from Arista?

Yes. I'm going to tell a little story and Ken, if you don't mind, you'll continue the story. Although we won't censor your version versus mine here.

When I started with Arista and I said, oh, we're going to build another switch? I was frankly thinking, I've done enough of that in Cisco, why would I ever do another switch? But what Arista has done is not just build yet another switch, but really build purpose-built network mission-critical platforms that have different use cases.

And part of having those different use cases is we had to make sure we could take different merchant silicon and ignite it and make it better. Because silicon leaders need better software to ignite them.

A white box would be a cacophony of an existing silicon, their STK, open losses. And I'm not saying anything negative about that except to say, the use cases are more limited. A blue box -- and then, of course, a purpose-built EOS is what can really sort out to develop. Maybe you can add a few words on that.

Yes. Look, I think that what we've shown is that networking software is tremendously important. That without the right software running across the network, the network just does not have the reliability, stability, flexibility that it needs by having one operating system, EOS versus extensible operating system running across the whole fleet, whether it's campus, cloud, WAN, data center, all running the same software. It simplifies things so dramatically for our customers that they're able to qualify one release, have one platform to automate against.

And it's just -- and its reliability applies across the whole thing. There are use cases for white box, but that requires advanced engineering on part of the customer. The customer has to figure out how to integrate white box with software into their overall network management suite. It is not for the faint of heart, okay? And we're seeing that the vast majority of companies need a network that works that they can count on and one OS is the path, the simplest, most reliable infrastructure.

Got it. I mean, right now, there's a lot of new data builders or data center builders in the market, they all might have certain aspirations of what they think that they can do. You might have better insight into what you think that they can do. Just where do you find that kind of mesh of what is the best product for that customer?

Well, I think we're all struggling with one common problem. Power, right? So if these data center builders can actually find the location, the colos and have the power to ignite it more power to them, really, right, because that's become the biggest scarcity.

And today, we're not talking about building a megawatt data center power, right? Everything is being translated into hundreds of megawatts and often gigawatts of power. Because of not just the CPUs or compute as we know it, but these powerful accelerators, whether it's GPUs, TPUs or the AMD accelerators. So the pressure that's putting on the power is really making us work with these builders and models to come up with at least three use cases that we see.

One is scale up, which is how you scale all of the compute capacity, especially in the back end of AI accelerator network. And typically, a scale-up configuration is a more limited configuration within a rack, if you will, where you might be connecting 100 or 1,000 or a few thousand GPUs. Power is somewhat constrained there. But at the same time, the surface area requires you to pack dense amount of cables and optics in that scenario. And by the way, a lot of times, people will use proprietary technologies like NVLink or PCI switching. We're big believers of Ethernet, and we're going to see -- we believe we're going to see an Ethernet for scale-up, E.SUN standard coming out this year that will further that.

The other is the scale out, which is how do you connect all of these racks together. And this is where Arista has literally flourished. A lot of our AI business is coming from the GLA configurations, but that takes advantage of all of the rich protocols, the telemetry, the availability, the visualization the intense features we've been working on for the last 15 years, in particular, for AI over the last 3 to 5 years. And so again, that's a power hogger and every kilowatt matters there. Because it's not just a GPU that's contributing the most, but the compute, the storage, front end, back end, and obviously, the network as well.

And then what we're seeing more and more with these colos is they're not able to get the power. And therefore, they're building many more distributed AI centers because if you can only get a gigawatt, you take the gigawatt you build as large as the stadium can bill and then you move across to another one. This distributed scale out of XPUs is also a huge use case for us because now not only do you have to deal with the different types of accelerators but you have to deal with isolating them segmenting them, traffic engineering across them, again, routing across these data centers in a coherent fashion, short, long distances. So power is a culprit in all three. But I think in order to deal with the power, we're coming up with different ways to centralize or distribute the designs.

Got it. With this now massively growing TAM, there's a lot of, kind of, competitors in the market, whether from white box, branded, silicon, software. Just how are you thinking about maintaining and expanding your ability to kind of capture your share of this TAM?

Look, I'm going to point to Ken first and say, if you don't have a network that works in an innovative differentiated product, we can never maintain share. I think what the company has done throughout, not just now, white boxes have been with us since we started shipping product 15 years ago, is to always coexist with it, build our value chain across, like I said, we have 22 Etherlink products in AI that have nothing to do with white box. They all complement that with the use cases.

So I think a great product platform is our greatest differentiator. Having said that, the other very important part of how we'll be working is with our ecosystem of partners. We see a world where it won't be just one homogeneous AI accelerator but it will really be a heterogeneous world with TPUs, AI accelerators, some of them may be built by specific vendors like AMD, some may be built in-house by our Cloud Titan customers themselves. And so having a common homogeneous network infrastructure for all of that heterogeneity is going to be very important as well as the model builders.

I don't see a world where it's ChatGPT's, the winner or Cloud's the winner or Gemini's the winner. Certainly, those three are going to be prominent, but there's going to be many more. And again, just as we lived in a multi-protocol networking world, we're going to live in a multi-model AI world.

Got it. Another question that we often get, I'm sure you're tired of is just you've had leading margins for a long time. How do you keep margins against this backdrop? And just where do you think those additional areas are for value capture?

There's two aspects to our margin, and maybe even three. One is what is our hardware cost and how much will the customer pay for that? And you would be surprised to know that the significant value-add and differentiators in our hardware, we don't just throw a bunch of chips. The huge amount of signal integrity. Every power -- watt of power we save translates to millions of dollars. Every latency, megabit, nanosecond we're saving translates to millions of dollars.

So I think there's a growth underestimation on the total cost of ownership of providing better performance, better hardware, high ratings, et cetera. And Ken's team has been -- I think your hardware engineering team is now doubled or tripled for AI. There's another whole aspect of it now, which is liquid cooling. So don't underestimate the power of amazing hardware designed over the last 10, 15 years for the right performance, availability, power constraints. That therefore, the customer willing to pay value because they don't want -- these are not throwaway toys.

These have to be there with them for 5 to sometimes even been 10 years, right? And then there's a software, I'm going to lean on you to talk about some of the huge differentiators. The combination of which often a customer looks at it and says, yes, you're worth a premium? Not because I want to pay you more, but because you don't come down and you have higher quality and but just better products.

Yes.

The software is absolutely key to where the margin comes from in certain segments of our markets. We have many -- we have addressed many markets, as you well know, for the hyperscalers, the big AI companies, but also into the enterprise, and government, health care and so on. And especially on the enterprise side, the margin comes from the fact that we have better software.

It is easier to deploy and operate. It's consistent across the whole suite. It works and it's all managed through a single cloud vision management console, which manages the whole estate from the campus to the data center, to the WAN, into the Cloud. And when you're able to deliver that kind of a consistent experience, a network the customer can really count on. And it is so much easier to manage and automate against, you can command a higher margin.

Got it.

I just want to add one thing, though, to that. Sometimes we don't command a higher margin. I want to be clear, too, right? In very large Cloud Titan situations where they may be using an open NOS certain products are lower margins. So it's -- you guys get to see the mix of our high, low and medium margin. So it's not always that we have a perfectly high margin. But the combination of it looks good to you.

All right. Perfect. You just spent some time talking about scale up, scale out, the scale across opportunity has gotten a lot more attention of late as a lot of traffic becomes north-south versus east west and training gets done between data centers because of power constraints. Just how do you see that opportunity for scale up emerging for you guys -- sorry, scale across emerging for you guys?

Scale across? Yes. We have been pleasantly surprised by the adoption of scale across. And think of it as a 2-step or a one-step removal from connecting the XPUs, right? So even in customers where we were not connecting their XPUs, because they had their own optical switch or they already had a prior design. We're now getting a unique opportunity to connect to their accelerators, one hop away through these distributed data centers that they have to build because as I said, the name of the game right now is multi-tenancy.

It's not just one homogenous AI accelerator. But as they want to access these colos, they don't have a product that can do multi-protocol routing EVPN, segment routing, traffic engineering, simple things like security and encryption, and also MSS or multi-domain segmentation becomes important because I don't want the traffic from one set of accelerators to talk to another. And yet, I have to do all these AI build-outs. So the flagship product for this that Arista has been developing for some time is our AI spine, the 7800.

We launched it late last year and running at 800 gigabits and providing this kind of real-time throughput with a rich set of features is very unique to Arista and describes the value of Arista in a tremendous way.

I could just emphasize something Jayshree said is that combination of having the raw performance characteristics that AI workloads require in terms of a load balancing, the buffering latencies, reliability, delivery of packets. It's a sheer throughput. Having those sort of low level of traffic characteristics. At the same time, as you have all the advanced features that she was talking about, the routing, security, the isolation and segmentation quality of service. Having those together on one platform, all managed coherently is a good differentiator for us.

And it took us about 3 to 5 years to build that. So it's a nontrivial effort.

Okay. All right. I'm not going to make it out of this room alive unless I ask you about the cloud titans. So...

We want you alive.

Yes, exactly. You made waves on the earnings call, noting the potential for 1 to 2 more 10% customers this year while still seeing growth in your other two cloud titans. Just -- that was a big statement. Just how should investors think about these kind of two more emerging kind of customers?

Look, I'm going to keep working hard to increase the denominator so that we don't have 10% customers. But the reality is these are large purchases. These are long-time partners. I fully expect customer A and B to continue to be 10% customers, maybe the percentages will vary by year. But the reason I expect we will have they might be high single digit or a 10% customer is the spend is just tremendous, right?

I know you guys track CapEx more than I ever do. And while our CapEx is very small compared to everything else they buy, it is very clear to us that the combination of the front-end clouds. But now the AI as a huge Copilot assistant to it is having an effect, not just on our titans, but I would also say on our AI specialty providers and some of the Neoclouds. So I think our base has got bigger. Our opportunity and TAM doubled in the last few years from $60 billion to $105 billion. And so you would expect us to cross the $10 billion mark this year. We were 9 last year. I think we've signed up to north of 11 now. But that won't be enough. I know we have to go into the teams and beyond. So naturally, that kind of result expects a customer base, therefore, that we'll spend. And we're certainly very intimately involved in co-designing and making that possible.

Got it. Kind of a couple of follow-up questions that I've gotten since the quarter and since that announcement is just all four of these customers kind of be the same use case, so it highlights the breadth of Arista's portfolio. Is it the same as these kind of 100,000 clusters you guys have been talking about for the past couple of years?

No, not at all. I think if I look back a couple of years ago, the benchmark was how big and bad is your GPU? And we too got strung along by that. But today, I would say it's not how many thousand GPUs in my building and cluster. It's more what's the aggregate going to be? Because they may build small ones and then go scale across and therefore, have a million GPUs. So they may start out with small racks and scale up.

So we fully expect to see all use cases for AI shine in this number, scale up, scale out and scale across. But more importantly, we also expect to see the front end of the cloud connecting to them as well and get some refreshes. So it isn't just isolated AI. It's a combination of AI and cloud networking as well.

Okay. And then another question that we get is just -- is it always going to be the end customer that counts as the 10%? Are some of these model builders that might be using kind of different data center builder partners, kind of who is the customer and who's kind of mandating that networking decision?

I think we've taken a pretty pure approach to it. So we don't count the influencers, we count the end customer, right? Influencers could become a large end customer over time or decide to do something different themselves. But we're pretty clear that the channel isn't the customer, the end customer is the one.

Okay. And then there's a lot of -- we've talked a lot about the Neoclouds. There's a lot of data center builders popping up. But obviously, kind of require some level of prioritization from you guys just in terms of kind of how to judge some of those opportunities. So how are you doing that?

I think this is an excellent question. I'll kick it off and -- we do have a vast number of requirements, clearly from the AI sector, but don't underestimate our commitment and investment to the enterprise. We've got a very large set of customers now over there. We have over 10,000. Our specialty providers also include service providers. So we keep them all in balance.

And we look at this as not just a near-term priority, but who and what are we in 3 to 5 years so that we don't get seduced by, the hottest thing we can do now and forget them. So Ken's team works very closely with Todd Nightingale, our Co-President and myself and the product management teams to prioritize, prioritize, prioritize. Because no matter how many engineers you add, it's never enough, is it?

Never enough. The other thing I'd like to add here is, this is another example of what an advantage it is to have one operating system. Because when I'm building new platforms for AI use cases, when I'm building new software features, targeting a hyperscaler operator, I can then leverage those in other segments as well. So we're getting that sort of -- that alignment between our different efforts. And so we are able to do both at the same time, because we operate as one big software team on a common software platform.

Got it. We get a lot of consternation from investors around NVIDIA and Spectrum X, particularly at Meta. Just are there any meaningful changes to how you see your relationship with the Cloud Titans and maybe kind of how you see NVIDIA as a participant within the networking market?

Sure. Well, first of all, shout out to Jensen. He's just done a tremendous job of building the world's best accelerators. That being said, we are going to work with all accelerators and build the world's best networks, especially if they're Ethernet. So -- and I think the two can coexist. And I'm really grateful to see this whole bunch of accelerators crop up because it gives us greater opportunity to connect those networks. So NVIDIA is very much a partner when it comes to the accelerator side. NVIDIA is very much a competitor if you look at their Mellanox division and they're trying to push NVLink or InfiniBand or even their own version of Spectrum X, where they have a natural tendency to bundle everything together. It's only normal, right?

So I like to focus on the positive side of this because there's so many more accelerators to commit to than worry about the vending. More specific to Meta or any other partner, our position and our partnership has not changed. We continue to co-develop with Meta. Our partnership is as strong as ever. And we see occasional blips depending on their purchase behaviors when they skip a service cycle or, whatever, you've seen that in the past. But I think Meta's just being smart about working with multiple accelerator vendors and making sure that the network partner of choice continues to be one they can closely work with.

If I could expand on that just a little bit. I wanted to point out that in the context of NVIDIA. There's been -- they've been very successful at the back-end networks interconnect GPUs. But our Cloud Titan customers, certainly including Meta, have much more broad and diverse networking needs than that. So if you're talking about the backbone, talk about the data center spine. You're talking about the access networks. That's a place where I think Arista is well differentiated and it's very strong.

That's a really good point. Back end and front end have to come together and almost always Arista is always the chosen one in the front end.

Exactly.

Got it. I know you've kind of talked about this, Jayshree, that it's getting a little bit harder to kind of distinguish what is AI, what is traditional networking at some of these customers. Do you think that kind of people just get too obsessed with, okay, what is the AI number versus just kind of what is the cloud number in general?

I agree with you. At first, I tried to be very pure and say only the first connection to the AI accelerator is back end. And -- but then I found that people were counting optics as AI and so everything is getting whitewashed with AI, right? We're still trying to be very careful and look at AI as at least the first and second half. But I do think people need to pay attention to, if the AI spend is really high, chances of them doing a refresh on the front end or the cloud may get delayed, or they may do it ahead of time like one of our cloud titans did before they do the AI refresh. So not all of these happen together. They happen in waves. And so if we have a first wave with AI and we don't get much cloud spend, Remember, there's time for that later, right? And so they do go together, but they don't happen at the same time. That's an important thing.

Okay. Got it. Maybe talking about optics for a second. We've heard a lot about optical circuit switching and the opportunity for optical vendors, at least some of them talking about kind of partially being at the expense of spine switching. Just -- where do you see kind of optical circuit switching fitting into the market? And are they more symbiotic than people think about?

I think they're more symbiotic. There may be a couple of use cases, but we have seen optical switching mostly and predominantly in a couple of customers in scale up. We don't see any optical switching in the scale across, although you can't put coherent optics if that's what you call, but that's not quite optical switching. So where there are 7,800 spines, there almost always is a VR, ZR+ or some sort of coherent optics that we're connecting to.

But other than that, we've not seen any disruptive designs that make optical the only one choice. And there's a simple reason for that. People need not just the plumbing to connect, but the intelligence to get the performance, the package, the forwarding, the quality of service the security, so they wouldn't risk their expensive AI clusters to just do layer one switching. They want much more.

Yes. maybe moving that conversation to CPO. There's been a lot of talk about whether we're going to -- we've had little introductions kind of on the scale-out side. There's kind of talk about there being larger introductions on the scale upside. Just where do you see kind of CPO timelines? How do you -- Andy has obviously been very involved in this over time?

Yes, he's been very vocal. But I want to say on behalf of Andy, he's not an anti co-packaged optics. He's pro good co-packaged optics. And that's what we haven't seen for the last 10 years. It's largely been in the labs, right? So first, first and foremost, I think it's important to understand that the versatility of optics, particularly within a data center or across is highest with pluggable optics.

If you can build the right form factor like Arista pioneered with OSFP and continue that over 800 gig, 1.6T, 3.2T, I think you'll see that as the 80%, maybe even today, 90% of the user case. Because it's so reliable, it's pluggable, it's versatile. You can change your mind, you can mix, you can match, et cetera. Now there are places for co-packaged both copper and optics. I think co-package copper in a more limited radius within a scale-up, if you can be 2 to 3 meters would be good enough and more cost-effective less power. Co-packaged optics, we will become real fans of it. if we can make it open and standard spaced. We think that's super important. Otherwise, you end up having one vendor situations, where one vendor is doing one type of CPO and another and another. So we're not fans of that, but we will really embrace CPO when it's more open and standard spaced.

If I could just emphasize one thing that Jayshree just said there. the openness and interoperability is core to networking. And so many companies in our space attempt to capture the market with a single vendor proprietary solution. I can give you a long list of these IPX/SPX Network, DecNetLab,SNA. AppleTalk, LocalTalk, Net Pilots, Net Buoy on and on. It's a graveyard of technologies because open and interoperable wins every time in the market. And we're going to see InfiniBand go the same direction, by the way. And so the -- this is where taking CPO from a vendor proprietary technology to an open pluggable interoperable standard. Like all of the other optical standards are, I think is going to make the difference between it being viable and nonviable.

Got it. Alphabet soup of technology graveyards for those.

A few more for those who have pre-2000.

Yes, we'll see how the transcript capture all of those. All right. So we spent the vast majority of the time talking about the data center market. You guys have also been scaling the campus opportunity and the enterprise opportunity continues to be important. Just how are you seeing -- is the primary opportunity on the campus side still with existing data center customers? How are you taking advantage of kind of the HPE Juniper dislocation?

Yes. Actually, the dislocation started post COVID. It wasn't a vendor-specific one. We chose to enter at a time where suddenly no offices were being built. And there was -- we used to talk about the carpeted headquarters and the remote branches and none of that mattered anymore. It was one homogenous system where everybody was working remotely and the user had to carry their credentials, whether they were in a hotel or Starbucks or at home or office in the post-COVID era.

So I think that was a unique disruption that, therefore, gave us an opportunity to really build a wired and wireless homogenous system and take advantage of that same leaf spine architecture. Before I turn it to Ken to give more details of some unbelievable innovations you have done to make Layer 2 function better, which is a large part of campus. What I would say is we're starting -- initially, we saw a lot more of our existing customers, adopt our campus, but we're starting to see a nice blend of both. I would say we see about 40% new customers and 60% existing right now and the shift to more and more new customers is happening

Yes. No. I mean we're seeing also plenty of campus first customers, meaning customers who first purchased from Arista is for the campus. Campus is not a follow-on or just an add-on. It's a first-class initiative. And one of the reasons for that is the innovation that we brought to campus by integrating the wired and wireless management, being able to deploy into whether it's a branch or a small campus or even a large one, the wired and wireless together.

Too many of our competitors view them as these completely separate things, where wireless comes with its own complexity, its own set of controllers you have to deploy. We have a controller-less solution. We've taken the wired side and wireless side and made them work better together, both from a management point of view, but also just from a sheer control playing point of view, using the wired routed infrastructure, as essentially like a mobile IP infrastructure for mobile campus clients.

Because support hundreds of thousands of campus clients in a completely flat IP space, so anyone can go anywhere across a large campus with seamless fast routing across the whole infrastructure. And we've really raised the bar, I think, on...

We've also done some pretty incredible things like SWAG, the stackable open architecture. So I think we again brought more open standards and innovation to a somewhat stagnant set of technologies in the campus. It's been kind of fun to see that. And security. A lot of segmentation and Zero Trust security inside the campus as well.

Okay. In Q3 -- on Q3 results, you noted that there were some kind of constraints on the part of your customers that kind of limited some upside. Just what are you seeing as the biggest bottleneck that your customers are facing?

What did I say in Q3?

Just that there were some bottlenecks to your customers that were in selling equipment that was just kind of making it back towards you.

Oh. Look, I don't know if I was referring to supply chain or what the context was, but while we have solved many supply chain issues, I think the one in front of us in Q3 and actually, we realized it more acutely in Q4 is memory. Obviously, we're in a conference where I'm sure that's discussed a lot. And memory goes through these cycles. But I think because of the high adoption of memory in automotive sector in AI sectors, specifically servers, we're seeing some real shortages there. It's affecting our customers. We're doing everything we can to lean in and overcome it. I think it's going to be a 2-year problem that actually separates the wheat from the chaff. And so we're making significant investments and purchase commitments in chips in silicon and memory to make that possible.

Got it. Any questions from the audience as we wrap up? All right. You're all too busy.

All right. Well, this has been a great conversation. Jayshree, Ken, thank you so much for being here today.

Thank you. Thank you for having us.

Thanks, everybody.

Good morning, everyone. I think we're about to get started. Just a brief intro. I'm Mark Newman, Bernstein's U.S. IT hardware analyst. I'm joined on the stage on the far end by my colleague, Daniel Zhu, who is Bernstein's new networking analyst. And Daniel and I are delighted to welcome Arista's, John McCool. John is Arista's Chief Platform Officer and Senior Vice President as well as Special Adviser to the CEO. John has been at Arista for 9 years and prior to Arista had more than 35 years executive networking experience, including being SVP of Dell EMC and 17 years at Cisco. And on my left here, Rudolph is Arista's Head of investor advocacy and has been at Arista for 9 years as well as another 13 years of networking and cybersecurity experience prior to Arista.

So I think given the huge shifts we're seeing in the space, it's a great honor to have both of you here today.

Thank you for having us.

So I'll get started, and then I'll hand over to Daniel. So perhaps a very high-level question just to kick off on AI. The AI opportunity is clearly huge. Can you talk about how you're viewing the AI opportunity, how it's evolving and risks and opportunities for Arista?

Sure. So I think, clearly, Ethernet has a place in this AI opportunity. I think if we went back a couple of years ago, there was a question, what role does Ethernet have connecting what we call the back-end network, connecting GPUs and clusters. We've made substantial progress, not just as Arista, but as an industry in the standardization and development of that back-end network. And then the interconnect of these clusters, we call the front-end network also grows tremendously along with the challenges of power consumption. People want to connect multiple clusters across large numbers of physical locations to get power effectively. So that's kind of the framework that we think about the Ethernet opportunity.

Yes. I mean I think the other thing that is clearly manifest is the networks are getting larger, right? I mean you can't throw enough compute at the problem apparently. So now how do you build these larger and larger networks that are still just as efficient. So I think lots of room for innovation there, and we've done well for the last 20 years by innovating and really continue to drive the performance bar higher and higher. And I think in the AI networks, it shows itself up even more.

I guess kind of adding on to that, how are you seeing the evolution of the AI opportunity between front end and back end and also between scale-up, scale-across and scale-out?

Sure. Do you want to start?

Yes, I can start. So one is it's -- we're, I suppose, lucky in the sense that our products are very fungible, right? The same products can work in a front-end network, in a back-end network. And what it gives customers is that flexibility. So it's a little bit harder for us now to parse out like what is front end versus back end. And customers, I think, are increasingly finding that these networks are kind of blurring together, right? I used this analogy in one of our group meetings earlier today, where if you're on your Instagram app and you're trying to look like a Cowboy, is that front end? Is that classic data center? Is that back end? Like where does that line get drawn, right? So that's kind of one thing.

The good news is all of that growth has been good and great for our business. Then talking about scale-up, scale-out, scale-across. So a couple of things to remember. Like scale-up is really within the context today of the rack, but it's starting to go beyond a rack, right? And this is kind of the highest speed network, if you will. It connects to the memory, but it is also, in some sense, the simplest network, right, because there's not as much complexity kind of involved there. That's an opportunity that today is really captive since the bulk of the accelerators being used out there come from NVIDIA, it's really an NVLink is the protocol that's used a proprietary protocol. So it's not much of an opportunity today, but there are some efforts called Ethernet for scale-up networking is a term that you hear quite a bit.

So we view that as an opportunity for Arista probably in '27 at the earliest. So not something that we include in our TAM right now. Scale-out has been the bulk of that AI revenue. That's where you're connecting across racks, maybe even across buildings that are relatively close to each other, right? And then scale-across is something that's come up fairly recently, let's say, over the last year, but it's frankly a concept that's existed for a while, right? Because if you have data centers across multiple locations, how do you interconnect those data centers. AI brings some unique challenges there, but you've got to think about things like routing and encryption and traffic engineering once you start going. So I think that for us is a very exciting opportunity because it takes a unique set of platforms, which there's not that many companies that have those.

It plays to our strength that we've had in traditional cloud. What we saw in traditional cloud was the same phenomena. How can I connect as many CPUs as I can in the data center? And once we were able to scale ECMP very wide, those data centers provided not enough power. And then we had -- that was actually our impetus to develop routing to add another layer across. The same thing is happening with GPUs and the power consumption is more constrained. So you want to have a broader number of data centers that look like one logical GPU cluster.

Got it. And just double-clicking on that because I think -- sorry, I think the scale-across opportunity is a little bit less well understood by investors. How do you think investors should think about sort of dimensioning that opportunity? And how do the competitive dynamics differ from the opportunities within the 4 walls of the data center?

Sure. Maybe I'll start with the competitive dynamics. These are highly sophisticated networks, need strong routing convergence time, need capability like encryption end-to-end, need multiple optical type of support, so variations in the type of optics dependent on the physical locations of these different data centers. So it's extremely sophisticated. It plays to the strength that we have in our modular platforms, which have a virtual output queuing architecture with deep buffers to deal with the distances involved and plays to our traditional strength. And as Rudolph said, very few suppliers can compete in that kind of environment.

Just a reminder for the audience, we do have an app. You can submit your question there. We'll get to some time at the end, questions for the audience, you can enter in the app or if you prefer to ask live, we'll offer an opportunity to do that with a microphone as well.

So I think kind of speaking a little bit about architectures, I think within AI, we've seen a shift from AI model training being very dominated by pretraining to one that has more of a component of kind of post-training and test time compute as well as inferencing ramping a little bit more. How does that impact sort of the network architecture design that you guys are seeing? And how does that impact your networking opportunity?

Yes. I think that the increase in this pretraining activity brings broader storage and more machines into play, which really drive that next layer of the architecture and that front-end network as well as pulling data from different data centers and that scale-out capability. When we move to an inferencing environment, we think a little bit about the user interaction, which adds a latency component to the question. So you input something, it has to be computed back-end network and then get back to you. So it really interacts across the whole dimension of that network. And we see enterprise customers thinking through their next-generation campus networks and how that will impact them.

We had one discussion in the early days of training where one of our large hyperscalers talked about the things they hadn't thought about. So they were so focused on that back-end network. They really hadn't thought about the dimensioning of the front-end network and how that was impacted by pulling in the storage data. And one thing I thought was interesting was on the wide area network. If all the content that we're getting from AI is individualized now. We're not all watching the same cat video. We're watching different cat videos. It can't be cached. So some of these wide area networks were designed with caching capability to kind of eliminate some of the bottlenecks on the back end. So there'll be pressure all the way to the user as this becomes more of a widespread opportunity on inferencing.

There's also things like RAG and stuff like that, that are kind of expanding the aperture of what connectivity is needed, right? So it's not as simple as, look, I've got these or 5,000 or 50,000 GPUs that I'm just trying to interconnect, which is, in a sense, a simpler problem. Once you start having to go beyond just the resources that you control and you own and you're trying to connect the service on the Internet, maybe you're trying to connect to your database that sits in a different location. It just opens up more complexity. And I think one thing that I'd say as a takeaway is complexity has been Arista's friend, right, because more complex networks require higher levels of R&D, higher levels of engineering. And all the way from Andy and Ken, who started this company, I think we've shown that we can innovate better than most.

So is there still a lot of experimentation going on with the cloud customers in terms of how they deploy GPUs and networking to -- in response to power and cooling requirements. Is there still a lot of experimentation? Or are you seeing more of a consensus for the direction forward in terms of how it is done?

Experimentation on one angle, but the intensity of the power consumption and the cooling becomes a question even within a single customer of where I'm deploying and what kind of capability I have in that data center. So if you have a modern data center with liquid cooling, maybe that customer -- and a lot of power, I'm trying to affect the highest level of GPU density within that form factor. I may have legacy data centers that I don't have any liquid cooling. I will still want to air cool, but still have some capability in there. So there's a broader diversity of physical form factors to meet the market needs than I think we've seen. And I don't know whether that gets better. People started to standardize in cloud on not only just leaf-spine to directions, but the width of the rack, x86 GPUs, there just seems like there's going to be more diversity because we're on the edge of that power and cooling dynamic.

And some might take offense to the word experimentation, I guess, but the space of innovation, which is I think the point you're making is not slowing down, right? Like to give everyone an anecdote, we introduced our 400-gig platforms in 2019. We introduced our 800-gig platforms in 2024. So call it, 5 years. It's not going to take 5 years to go from 400 gig to 800 gig. It's definitely not going to take 5 years to go from 800 gig to 1.6. So what that is leading to is a lot of experimentation and trying out different things because ultimately, what they're trying to optimize is power utilization, time to first token, job completion times, being able to generate more tokens per dollar, more tokens per kilowatt, like all of these things are -- so they're willing to try pretty much anything.

The utilization of the GPUs, you want to maximize the utilization of the very expensive assets.

Exactly. Exactly.

I guess since we've talked a little bit about sort of the technology road map and 800 gig versus 1.6., can you talk about sort of the time line of that rollout and kind of what you guys are seeing with that sort of transition?

Yes. I think you just hit it. It's become more compressed than we saw 100 to 400, 400 to 800 compressing. It's going to go faster. We've seen a lot of announcements of silicon for 1.6. I think that one thing that we have learned, we've gotten this question, by the way, I came into the business, we had really secured a substantial position in market share, 100 gig. And everyone said, what's going to happen at 400 gig? And then what's going to happen at 800 gig. And we've been able to layer on those technology transitions based on a consistent software architecture and methodology and development.

We'll always have to add new -- some new things and capability as the speed increases, but it's been pretty straightforward. And I think we've led that deployment. The critical time is from first silicon to thousands of GPU connections. A lot happens within that. So you see us tend to wait until we really have deployments to make announcements around the next generation.

Got it. And continuing with the technology road map a little. We're starting to see a lot more talk about co-packaged optics. But in the past, I think you've talked about how the failure rates of optics is a big bottleneck for CPO adoption and how replacing the DSP and retiming chips of linear drive pluggable optics actually captures a lot of that is a pretty effective substitute. Is that still true with the 1.6 cycle? Or what are you kind of like starting to see more evolution here?

Yes. I'd like to answer this question by kind of coming back to why co-packaged optics began and the concept. There's really 2 things that are intriguing about co-packaged optics that we're trying to solve. One is power consumption. So if I can shrink the distance, but also the number of chips between the switch chip and the wire, either cable or optics, I save power. So that's a good thing. The second problem was in each one of these speed transitions, the laws of physics decrease the distance that I can run on a printed circuit boards between the switch and the front end. So the ultimate shrink is really putting optics on the die.

So I think we would say there's some inevitability to the trend of shrinkage, but we've been able to defy that in a couple of ways. The first observation we had is the improvement in the switching chips DSP technology that was designed for co-packaged optics allows you to eliminate the DSP in the optics component itself and preserve the operating model that the hyperscalers use of pluggable optics from multi-vendors. If something breaks, I can replace it. So we were -- and get all the power -- most of the power consumption savings that you have with CPO. So we were able to push out that transition 1 -- and we believe 2 generations. But there will ultimately be a time where to get the laws of physics, bringing that optics closer to the switch chip is inevitable.

One thing I would add is the biggest kind of concern around co-packaged optics is optics fail most often, as you pointed out. And if the optics fails and the optic is actually on the switch or on the board itself, you not take the whole switch out of commission, right, which means every workload that's connected to that switch is now paused or shut down or has to restart, et cetera. So the reliability requirements of the switch have just gone up exponentially, right? The hardware engineering -- and again, complexity is our friend at Arista. We've got the best and sharpest engineers. So we feel really good about engineering for that co-packaged era. But I think what customers are telling us is kick that can down the road as much as possible to John's point.

Fans, power supplies and optics tend to be the things that people want to replace on the fly. So you make in a situation with co-package, maybe there'll be some improvements in the overall reliability of the lasers. Maybe networks will be designed to be redundant of some path failures, so you can delay the serviceability requirements of those switches, but there'll have to be some operational adaptations.

Got it. And I guess continuing with the theme of emerging technologies, sort of optical circuit switching is a technology which has actually been around for a while but remained fairly niche. But at least we're starting to hear more talk about it. And I think in particular, there's talk about scale adoption by one of the hyperscalers. Can you talk a little bit more about what you guys are seeing in the space and sort of how that would impact Arista's business?

Yes. I think you framed it well. I think there's been one customer in particular that's used optical switches and some large-scale deployments. And you can think about them having massive scale. It's not equivalent to switching or routing. It's not a per packet-based decision on which way the packets go. It's more of a construct that interconnect some ways thinking about replacing patch panels, being able to allocate and move different elements of your compute or GPU environment at a more granular level. And as some of these deployments and other customers get larger, they can have the benefits of that kind of technology. So I think there is some elements of TAM growth because more people will be able to use those technologies, but it's still kind of a different aspect than traditional switching and routing. And if you want to...

Yes. I mean I think to Mark's point, this fits in that experimentation bucket, right? Like is there some pound of power savings? Is there some pound of performance improvement I can get. But even at this customer that has been using optical circuit switching for a while, once they get beyond a certain layer in the network, it is all Ethernet, right? Because the flexibility that Ethernet gives you is just tremendous. The variability in the supply chain, the diversity of suppliers, all of that, I think, brings its own advantages. So I don't think we view those as necessarily competing. I think the TAM for optical circuit switching is significantly smaller than Ethernet switching. So...

Got it. And we've kind of been talking about a few technologies that sort of are real, but maybe less disruptive than some investors might hear. Are there any technologies that sort of are on your radar that you think kind of are being underappreciated by Wall Street?

I'd say we touched on this earlier. The scale across opportunity, I think, for us is very, very exciting, right? I mean to kind of repeat a little bit of what we said earlier, like it takes a unique set of technologies. It requires these deeper buffers. It requires the virtual output queuing. It requires encryption. And that's something that I think probably the biggest misconception I've heard out there is people assume that a switchboard is a switchboard is a switchboard, right? But to John's point from earlier, a low latency switchboard is much more different than a scale-across switchboard. And so that nuance, I think, sometimes gets missed.

Got it. And I guess if we zoom out and kind of like talk a little bit more about the platform, right? I think at the Analyst Day, I actually asked Jayshree, why is it that Arista was the only -- essentially the only networking company that sort of entered the market after the dot-com bubble and saw sustained success. She highlighted that data center switching has really been kind of the foundation of Arista's strength for a long time. That's a category which has been one of the fastest-growing and most important components in the data center and probably all the more so now. So can you talk about how sort of you're leveraging that strength in switching to building a platform that supports other products and services?

Yes. Maybe just to amplify Jayshree's answer. In 2010, '11, it's hard to imagine how small these hyperscalers were. So it was a classic case of an underserved market. And a new competitor coming in with hyper focus, not trying to take enterprise networks or service provider networks and serving the emerging hyperscalers. Their problem was they had very smart people, sophisticated operators, but we're trying to connect together millions of compute nodes with a very small team. So how can I put my agent on your networking device like I do with my servers on a Linux operation and build a centralized management stack? How can I have resilience and scale and being able to scale out without bespoke different operating systems? So they were sort of the architectural premises that we developed at Arista. The hyperscalers had their own management stack, a centralized management stack.

They encouraged us to be able to stream data to that management stack. We took that concept and built something called CloudVision, which is a management stack that we use for people who don't have the wherewithal to build their own. So back to your sort of question on AI, this being able to scale and manage is happening as we speak into campus environments. If you think about a university, it's very easy to think about a new student coming in with 3 devices they roam with, right? Maybe a pad, laptop and perhaps a watch. And they're moving around campus. You imagine them all coming to see a sporting event that seats 20,000 people, all of a sudden, you're up to 60,000 Mac addresses in a very small place.

It starts to feel like connecting GPUs and CPUs. We've taken in our WiFi stack concepts that have been developed and standardized in the data center like EVPN and VXLAN, and with our VESPA architecture using Wi-Fi 7, being able to build these large Layer 2 roaming networks that fit into campus. It is a similar problem except these people move around. They're not standard like racks. The operating principles of a single management stack and a single operating system in your data center and campus are consistent with what we've done in the data center. So very much plays this whole consistency and scale piece that we've focused on.

Yes. One maybe thing that I'd add is our customers -- what customers really liked about us in the data center also was just the cost of ownership of an Arista network is fundamentally lower, right? Because it fails less often, it's easier to maintain, it's easier to operate the level of automation, et cetera. And so if you think about it, who has even less money to spend on networking, right? It's the enterprises because they might have one networking person. If you're a large bank, for instance, which many of you probably work at banks and you've got 5,000 ATMs, you don't want to be sending a tech to reboot a switch at every ATM every few weeks, right? I mean it's just not cost effective. So you want that reliability, you want that centralized management, you want that automation. And I think that's the advantage we've been able to bring from our cloud learnings, if you will, to the rest of the world, right? So we call it the modern operating model, and it's been very, very effective for customers.

It's hard to imagine in networking, customers were very scared to upgrade because they can run into regressions, something that used to work, I've upgraded it, now I broke it. And with kind of the concerns around security and security alerts, being able to move quickly is really important. It was really critical in the data center, but it applies to the entire network.

Got it. And now that we're talking sort of a little bit more about enterprise, where are we seeing sort of enterprise AI adoption inflect the most? And how do those use cases differ from sort of what we've seen kind of with the cloud titans?

Yes. Maybe I'll start and you add in here. I think definitely, we see different verticals with a different approach. I think the consistency is an enterprise may have certain amounts of data that they want to monetize or take advantage of and build out their own AI pieces. It could be in the financial vertical, maybe around security and credit card and credit card handling and fraud. In health care, it could be detection of anomalies around your health or radiology. So we see these vertical applications emerging in AI, specifically where a company may have a lot of data that they want to take advantage of.

Yes. I mean I think the health care example is a great one, right? Like we have customers that will have conversations with us about, look, we want to take advantage of this AI wave, but we're concerned about -- our patient privacy, right? So like how can we ensure data is segmented correctly and there's encryption on the wire, and we've got threat detection on the wire, et cetera. So we've got a whole slew of services now that -- and software capabilities that we've built within the switches that can provide that, right? I'd say the biggest trend we're seeing maybe with customers is fewer of them are trying to build their own training clusters because I think they recognize that the amount of CapEx investment that it takes for that is incredibly high, right?

But what they also realize, and this touches on something John said earlier, is inferencing is very much a latency-driven game, right? If you're going to ask a question, and it's going to take 2 hours to come back with a response, you might as well not have it, right? So how do you optimize that? Well, you can try and bring AI to the edge, if you will. So you're already starting to see questions about, okay, is there going to be an accelerator on each of these devices you have running in front of you? Can we bring the inference compute as close to the edge, right, whether it's putting it in a data center that doesn't belong to you, but closer to you or putting it in your own data center. So those are some of the trends we're seeing.

But I mean, we talked about over 100 800-gig customers that we have now and just from about a year of the product being out there, right? Obviously, there's not 100 hyperscalers. So there's a small number of hyperscalers in there. There's some of these new clouds and specialty clouds, but a fair amount of enterprises in there in the verticals that John mentioned, right, insurance, financial services, health care, the educational sector is up there, manufacturing, some of those folks are trying to think about the whole AI factory concept.

Got it. And we spent a lot of time on AI, and we kind of have touched on this already, but campus networking has obviously been a huge area of focus for Arista. Can you talk a little bit more about sort of value proposition and kind of the go-to-market in scaling that business?

So yes, I think it's important to think from an Arista perspective, campus is part of our enterprise go-to-market. So we have a very intimate engineer-to-engineer relationships with the large cloud providers. But we've had from the beginning of the company, starting with the financial vertical an enterprise go-to-market. The only thing they could sell was low latency switches when we started. Then we added more broad data center products, and we've been continually to add to that. We don't necessarily go to a customer or target a customer because they're a campus player. They buy a lot of networking equipment. Networks is important to them. And then our sales team will look for whatever opportunity exists within that account.

Maybe their WiFi is coming up first or maybe it's a data center opportunity, but they have more arrows in the quiver, if you will, to go build out an enterprise network. So that's how we go about it. We were pushed from our data center customers. Why aren't you going into campus because they want the same benefits of a single operating system, the same operating model into the campus pieces. So we took that to heart. When we were ready, we built our campus LAN switches organically. We did add a Wi-Fi component that I would say we've [ Aristafied ] into the architecture and going about it that way.

Yes. I mean I think it's important to keep in mind that the campus dynamics are also slightly different. I mean the refresh cycles can be 7 to 10 years. Data center, typically, even 3 to 5 years, you see a refresh cycle. So that's kind of one thing. It's also very rare that someone says, okay, you know what, come in and replace my entire campus my switching, my WiFi, my SD-WAN, right? It's much more piecemeal, right? Like where they'll say, okay, you know what, you won this business, but we're going to start with you in one building and then maybe a few months later, you get the second building. So it's a much more slower grind in that sense, right?

With that said, I think we're very happy with our growth, right? I mean we told the Street that we were targeting $800 million exiting the year. It was $750 million, and then we acquired VeloCloud, which John can talk in more detail about since he's heavily involved there. So we bumped that up to $800 million. We were able to meet that target. And now for this year, we said $1.25 billion, right, which we actually got this question earlier. I mean that's pretty aggressive growth in a market that isn't growing as fast, right?

And the reason we can do that is we're only about 3% market share in that campus market. So it's much easier to grow at a faster clip when you're a share taker, right? And it's -- I think we feel very confident about being able to become, frankly, one of the larger players in the campus in the next year or so. Again, there's one big player in the room and the name talks to the C, ends with an O. And then there's HPE Juniper, which obviously with -- especially with the merger now, they're in the probably in the teens percentage-wise market share.

But beyond that, it's been quite fragmented, and we feel like we've got a pretty good opportunity to kind of grow that business. So it's an exciting opportunity. It's a different sales motion. It's a different pace than the data center. But in many ways, I think customers care about the same things, right? They want a network that just works, right? You're not having to reboot stuff. You're not having to throw people at it over the weekends, things of that nature.

I wonder if I could just follow up on -- you mentioned HPE Juniper, obviously, a big merger there. Has that created an opportunity for Arista? Or what's the -- how is Arista benefiting? Or is there any risks from this acquisition that you see?

It's a discontinuity in the market. So that always presents an opportunity. There's been customers who used Juniper and HP as alternate sources. So now they don't have an alternate source, particularly around the Wi-Fi piece. There seems to be 2 things moving forward. So customers think about that. So it's an opportunity to create a conversation or have a conversation. But it hasn't fundamentally shifted our strategy. We're not sort of reacting to that event. We're still focused on the bulk of the market share and what we're bringing to market as opposed to changing our direction.

And do you see the combined scale and breadth that the new HPE has in networking as a potential threat to Arista potentially them coming in, gaining a bit more share in your areas or...

No, no. I mean what we've seen, it leaves us really as the largest pure-play networking company that can focus on every need in an enterprise. And that's a pretty big TAM for us. And as we said, particularly in the enterprise on the campus side, very low market share, so we can be a disruptor in that space. While on the other side, there's some rationalization and there's other pieces with storage and compute that require some focus for networking. So I think overall, we view as an opportunity.

Yes. I mean I think, again, this market, even if you go back to the data center market, size is not always an advantage, right? And frankly, it's one of things like Jayshree always keep us honest, right? Even though...

We disadvantage it.

Yes. Even though we've grown, we've got to keep that nimbleness and that start-up kind of mentality, which is why we tend to operate very efficiently and have stayed focused on pure-play networking, right? Because when your attentions are divided across storage and across compute and things like that, it's very easy to kind of move resources to whatever is the flavor of the day. And customers don't have patience for that, right? So I think in some sense, it helps create opportunity. It hasn't -- to John's point, it hasn't fundamentally changed the dynamic. And the largest player by far is still Cisco, right? So I think the opportunity is still there for, frankly, them and us to take share.

Great. So I had a follow-up question on memory, a bit of a hot topic these days.

We waited 30 minutes to get there.

Well, we have to ask about networking, too. So your CEO, Jayshree, talked about memory being the new gold and pricing environment being horrendous and lots of talk about exponentially higher pricing. My question really is how is Arista positioned versus other OEMs here? Are you able to secure enough memory? Or are you memory constrained right now?

We're not memory constrained. But as she mentioned, pricing has gone up, and we've made some targeted changes on our pricing to address that and also changed our internal price structure to make sure we have continuity of supply.

Yes. I mean it's -- if you noticed our purchase commitments have been going up. So it is -- and this is not specific to memory. I mean memory is a relatively small part of the switching BOM, right? We're not a server manufacturer where switching -- I'm sorry, memory is a bigger part of the BOM. So impact-wise, it's less of an impact. And also, I think one nuance that sometimes maybe the Street doesn't entirely appreciate, the more complex the switch the more memory. So for instance, the chassis switches, there's more memory in those. But as a percentage of the BOM, it's actually a lower number, right, because there's a lot more into that switch than just memory.

So what -- my point being is we are not maybe as impacted as a server maker. With that said, it's clearly an industry-wide constraint. I think our customers get it. So to John's point, where we need to, we have the ability to do these targeted price increases on these more memory-intensive SKUs and the customers get it. It's always a negotiation. I mean this is not a -- we'll say and they just do it kind of thing. But at the same time, I think because we've got that free cash flow, because we've got a lot of cash on the books, it's allowed us to make these investments to get in line? Because right now, I think the challenge is how do you get in line for that supply. But we don't believe -- pricing aside, we don't believe we are supply constrained in terms of being able to meet the demand we anticipate and our guidance is still valid.

Do you have a rough guidance on what portion of the bill of materials is memory?

We haven't -- we haven't disclosed that, and it varies.

We never disclose that.

Servers is going from 10% up to like 30% now.

Just because of the price. Just because you have the price.

PC is also from 10% to 30% roughly.

Yes.

So I mean, you're definitely in the single digits, I think.

We haven't disclosed that. But the percentages are going up just by the nature of the price increases. I would also just add in terms of being able to be in line, our relevance in the cloud market with top-tier customers helps some people understand where we fit in memory. They're also cognizant of the golden screw. If there's a lot of GPUs, but they can't connect them with switching, that wouldn't be great. So I think that suppliers are aware of that position. And obviously, we have a strong balance sheet that helps us make those multiyear kind of commitments where needed, right?

And your strategy is to pass on these price increases to customers and try to maintain your...

Well, some of it. I mean we've been able to absorb some of it. I think Jayshree said this or Chantelle said this on the call, right? Like Jayshree actually signaled the memory thing at our last earnings call, right? And I think everyone is a little bit surprised why she brought it up. But in many ways, I think it shows that she was almost questioned about where this is going. Now it has gotten actually worse since then. So we've been able to absorb for a while. But at some point, we can't, right? I mean the good news is our customers are some of the biggest buyers of memory, right?

Like if you talk about the hyperscalers, they need more memory than pretty much anyone else. So they get it. But no one likes a price increase, we don't like a price increase either. So it is always a tricky thing. But the question is really about our margin guidance. We feel very comfortable with that 62% to 64% gross margin guidance for the year that we can fit right within that using a variety of mitigations, right, whether it is absorbing some of it, obviously, purchase commitments and then some times price increases are needed.

Great. Thanks for that. Before I -- before we hand over to audience for questions, just one question I want to ask you, is there something -- what do you think that Wall Street is missing in your company? Is there something you'd like to highlight that you think Wall Street is not paying enough attention to?

When we have these conversations and even the discussion today, we think about from an investor point of view, TAM and opportunity, and we tend to segment things, right, front-end network, back-end network, campus. From an Arista perspective, it kind of -- especially the enterprise, loses track of our value proposition. Our value proposition, if we walk into an enterprise customer is, I can now deal with or execute any use case you have today, whether it's distributed enterprise wide area network connection, GPUs, connecting together your virtual hybrid networking environment. And I can do this all with a single operating system that is proven to be extraordinarily low in defects versus our competition and simplify your operating model.

So we go to market with that value proposition to an enterprise customer. They give us some thing about what's coming up in the next 18 months in terms of their changes, right, whether it's a management network in the data center or now with Velo, we're interconnecting a wide area network in a distributed fashion. So I just wanted to make that point and connect these different network use cases and area to really what our sellers are going to market with.

Yes. I mean maybe from my perspective, what I would add is I think networking is incredibly technical and complex. And maybe we need to do a better job of kind of simplifying that for the audience. But I think people sometimes assume like I was saying earlier, right, a port is a port is a port, like it's just a pipe, like how much complexity can be there. But there is a lot. And because of that, I think people assume sometimes that networking is also a zero-sum game, right? Like if -- like the question earlier about optical circuit switching, just because optical circuit switching is seeing some growth doesn't mean that Ethernet-based switching is declining because of that, right? It's not a zero-sum game. And I think especially in this AI era, like it is definitely a rising tide kind of setup. So you're going to see all kinds of companies, I think, do well because of that experimentation thing because of the pace of innovation, et cetera.

Great. Appreciate that. Any questions from the audience? I don't currently see any in the app, but feel free to stick up your hand, and we've got a microphone here and can come to you. Any questions? It's all super clear. We have one right here.

Question, things that are out of your control. What is the biggest thing on the medium-term horizon that is out of your control that will constrain growth? Is it power? Is it memory? What are you doing about this? How are you thinking about it? What mix can you put in place?

And power is definitely up there, right? Because I think customers want to build these as quickly as possible and power is probably becoming the biggest constraint. So to some extent, you're right, we don't control it. But what we do control is can we get more power efficient and power efficiency out of their clusters, right? Because the network is not a big consumer of power, but it is a consumer of power. And so some of the stuff that John talked about earlier in terms of reducing the power utilization by optics -- from optics by 50%, right, with our linear pluggable optics.

And again, we don't sell a ton of optics, by the way, which is I think another thing that the industry misses, but we are definitely thought leaders in innovating when it comes to optics. Andy Bechtolsheim is considered essentially one of the smartest minds when it comes to optics. And he's been pushing this idea of linear pluggable optics, eventually co-packaged optics, et cetera. So that's kind of one thing. But even if you look at the switches themselves, right, look, I love this comparison where even with the same Broadcom chip within our switch, right? So let's take a Tomahawk-based chip, put it in our switch, put it in another branded vendor switch. Our switches tend to be 15% to 30% more power efficient than the competitors. So that 15% to 30% that you can now save, you can apply to more GPUs.

Why is that? How can you save power?

Because we just -- it's kind of the stuff that John touched on earlier, like for instance, better signal integrity on the boxes, being able to design the hardware so that you're being as efficient as possible, getting rid of some power hogging components because you've got that better signal integrity. So it's just a variety of things.

You can lower the output power to the effect of the channel and minimize that, more direct control over the SDK on the software side to do some optimizations, thermal integration with how we do heat sinks and cooling to run the fans at a lower power. Lots of little things that add up to some pretty substantial changes. And sort of back to the memory, not 100% direct control, but definitely, we have an opportunity to influence that with some of the things we mentioned around our engagement with the large cloud customers and our momentum in the market is influential in helping us get some memory.

Thanks for that. Any other questions? We have one question over here.

They're making you walk the microphone today.

The other end.

So there was a large AMD meta deal announced yesterday. Is that additive to your TAM? I don't know if you can comment on that specifically. But then more broadly, in these -- once a large compute deal like that gets signed, let's say, is someone like a risk that consulted kind of as that discussion is happening? Do you come in later in the process? How does that sort of time line work of when the networking piece would come in?

Yes. I would say, look, I think we've been talking about the diversity of GPUs playing to Ethernet as the environment for running AI. We largely believed even when things were back in the InfiniBand days that, that would happen. AMD didn't even have a GPU in the market at that place, but we've just seen transitions like this that people want a diversity of endpoints and they want a consistent way to operate and connect those endpoints. So I think it's validation that there's going to be a multi-GPU environment. It's an external event. It's probably more long term in terms -- you see us making longer-term supply agreements. I think people are having to think through longer-term agreements in this environment to make sure that they have the capacity to grow. So I think it's all consistent with what we have baked into the model.

Yes. I mean I guess, in general, the biggest advantage, I think, the diversity of suppliers in any area gives -- it eliminates strategic locks that single vendor might have had prior, right, where they can influence buying decisions beyond just the GPU. So like someone saying like, hey, if you buy my GPUs, the network comes for "free" or you get the GPU sooner because you bought the network. So I think that -- it opens up the market, right? I mean the good news is all of these third-party, whether it's the branded accelerators or the homegrown kind of ASICs that are getting built by some of the larger players, they're all Ethernet-based, right? So that is a market that we can compete with.

We feel very good about winning in a best-of-breed kind of pipe. And it eliminates some of these maybe more go-to-market type impediments that we would have run into otherwise. So I think it's a good thing. It's directionally something that we've always planned for because I think customers always tell us that, look, we don't want to be locked into one vendor for anything. You guys included, right? Like even with us, right from the early days of the cloud, we've known we've had to coexist with other networking vendors. And it's not at all surprising that these customers would not want a single source something like compute either.

All right. Any other final questions from the audience? We're almost out of time. Maybe have time for one more if there is any other question. I don't see any hands, last chance. Daniel, do you have any other question on your side?

Yes. I mean I think sort of -- I think kind of we kind of just go back to sort of that high-level sort of like AI discussion, right? Jayshree has really talked about sort of addressing a $100 billion TAM. And as part of that, Arista has guided to AI networking revenue doubling from $1.5 billion in 2025 to -- I think the number is $3.25 billion in 2026. And I just want to get a sense of like how much of that expansion is coming from hyperscalers spending more versus Arista being able to capture kind of a bigger share of that opportunity compared to competitors than you might have anticipated before?

It's hard to parse it. It's a combination of both. Definitely hyperscale growth and our architecture growing within that is a key component. But I think we've also been effective at picking up some of the new companies and people that are starting to emerge in AI. I don't know if you have anything.

Yes. I mean we've talked about potentially even 1 or 2 more 10% customers, right? So I think we are definitely seeing a diversity of customers coming to the table. But also like we're very happy like Meta was referenced earlier, right? Like we're very happy with our partnership with our existing 10% customers. We don't name them anymore. They're customer A, customer B, but we still love them, just to say, right? And so yes, it's -- I think business is good on all fronts and not forget the enterprise, right? I know this was more AI focused, but even within the enterprise, we've got customers, like we said earlier, that are starting to be relatively significant, not as large as the hyperscalers, but the significant amount of dollars associated with AI coming out of the enterprise in the nontraditional places that you might imagine.

Great. Well, we're right out of time. Thanks, everyone, for joining us, and thanks very much for joining us on the stage today.

Absolutely. Thanks, Mark.

Thanks, Mark.

Thanks, Daniel. Take care.

Thank you.

Welcome to the Fourth Quarter 2025 Arista Networks Financial Results Earnings Conference Call. [Operator Instructions] As a reminder, this conference is being recorded and will be available for replay from the Investor Relations section on the Arista website following this call.

Mr. Rudolph Araujo, Arista's VP of Investor Advocacy, you may begin.

Thank you, Regina. Good afternoon, everyone, and thank you for joining us. With me on today's call are Jayshree Ullal, Arista Networks' Chairperson and Chief Executive Officer; and Chantelle Breithaupt, Arista's Chief Financial Officer. This afternoon, Arista Networks issued a press release announcing the results for its fiscal fourth quarter ending December 31, 2025. If you want a copy of the release, you can access it online on our website.

During the course of this conference call, Arista Networks management will make forward-looking statements, including those relating to our financial outlook for the first quarter of the 2026 fiscal year, longer-term business model and financial outlooks for 2026 and beyond. Our total addressable market and strategy for addressing these market opportunities, including AI, customer demand trends, tariffs and trade restrictions, supply chain constraints, component costs, manufacturing output, inventory management and inflationary pressures on our business, lead times, product innovation, working capital optimization and the benefit of acquisitions, which are subject to the risks and uncertainties that we discuss in detail in our documents filed with the SEC, specifically in our most recent Form 10-Q and Form 10-K and which could cause actual results to differ materially from those anticipated by these statements.

These forward-looking statements apply as of today, and you should not rely on them as representing our views in the future. We undertake no obligation to update these statements after this call.

This analysis of our Q4 results and our guidance for Q1 2026 is based on non-GAAP and excludes all noncash stock-based compensation impacts, certain acquisition required charges and other nonrecurring items. A full reconciliation of our selected GAAP to non-GAAP results is provided in our earnings release.

With that, I will turn the call over to Jayshree.

Thank you, Rudy, and thank you, everyone, for joining us this afternoon for our fourth quarter and full 2025 earnings call. Well, 2025 has been another defining year for Arista. With the momentum of generative AI and cloud and enterprise, we have achieved well beyond our goal at 28.6% growth, driving a record revenue of $9 billion coupled with non-GAAP gross margin of 64.6% for the year and a non-GAAP operating margin of 48.2%. The Arista 2.0 momentum is clear as we surpassed 150 million cumulative ports of shipments in Q4 '25.

International growth was a good milestone in both Asia and Europe growing north of 40% annually. As expected, we have exceeded our strategic goals of $800 million in campus and branch expansion as well as $1.5 billion in AI center networking.

Shifting to annual customer sector revenue for 2025, cloud and AI titans contributed significantly at 48%. Enterprise and financials recorded at 32% while AI and specialty providers, which now includes Apple, Oracle, and their initiatives as well as emerging neoclouds performed strongly at 20%. We had two greater than 10 customer concentration in 2025. Customer A and B drove 16% and 26% of our overall business.

We cherish our privileged partnerships that have spanned 10 to 15 years of collaborative engineering. With our ever-increasing AI momentum, we anticipate a diversified customer base in 2026, including one, maybe even two additional 10% customers.

In terms of annual 2025 product lines, our core cloud AI and data center products built upon a highly differentiated Arista EOS stack is successfully deployed across 10 gig to 800 gigabit Ethernet speeds with 1.6 terabit migration imminent. This includes our portfolio of Etherlink AI and our 7,000 series platforms for best-in-class performance power efficiency, high availability, automation, agility for both the front and back-end compute, storage and all of the interconnect zones. Of course, we interoperate with NVIDIA, the recognized worldwide market leader in GPUs but also realize our responsibility to broaden the OpenAI ecosystem, including leading companies such as AMD, Anthropic, Arm, Broadcom, OpenAI, Pure Storage and VAST data, to name a few, that create the modern AI stack of the 21st century. Arista is clearly emerging as the gold standard terabit network to run these intense training and inference models processing tokens at [ tariff locks ].

Arista's core sector revenue was driven at 65% of revenue. We are confident of our #1 position in market share in high-performance switching according to most major industry analysts. We launched our Blue Box initiative, offering enriched diagnostics of our hardware platforms dubbed NetDI, that can run across both our flagship EOS and our [ open NOS ] platforms. We saw an excellent uptick in 800-gig adoption in 2025, gaining greater than 100 customers cumulatively for our Etherlink products and we are core designing several AI rack systems with 1.60 switching emerging this year. With our increased visibility, we are now doubling from 2025 to 2026 to $3.25 billion in AI networking revenue.

Our network adjacencies market is comprised of routing replacing routers and our cognitive AI-driven AVA campus. Our investments in cognitive wired and wireless, zero-touch operations, network identity, scale and segmentation, get several accolades in the industry. Our open modern stacking with SWAG, switched aggregation group and our recent VESPA for Layer 2 and Layer 3 wired and wireless scale are compelling campus differentiators. Together with our recent VeloCloud acquisition in July 2025, we are driving that homogenous secure client to branch to campus solution with unified management domains. Looking ahead, we are committed to our aggressive goal of $1.25 billion for '26 for the cognitive campus and branch.

We have also successfully deployed in many routing edge, core spine and peering use cases. In Q4 2025, Arista launched our flagship 7800 R4 spine for many routing use cases, including DCI, AI, spines, with that massive 460 terabits of capacity to meet the demanding needs of multiservice routing, AI workloads and switching use cases. The combined campus and routing adjacencies together contribute approximately 18% of revenue.

Our third and final category is the network software and services based on subscription models such as a-care, CloudVision, Observability, Advanced Security and even some branch edge services. We added another 350 CloudVision customers a day, almost one new customer a day and deployed an aggregate of 3,000 customers with CloudVision over the past decade. Arista's subscription-based network services and software revenue contributed approximately 17%, and please note that it does not include perpetual software licenses that are otherwise included in core or adjacent markets.

Arista 2.0 momentum is clear. We find ourselves at the epicenter of mission-critical network transactions. We are becoming the preferred network innovator of choice for client to cloud and AI networking with a highly differentiated software stack and a uniform CloudVision software foundation. We are proud to power Warner Bros. distribution network streaming for 47 markets in 21 languages in the pan-European Winter Olympics that is happening as I speak. We are now north of 10,000 cumulative customers, and I'm particularly impressed with our traction in the $5 million to $10 million customer category as well as the 1 million customer category in 2025. Arista's 2.0 vision resonates with our customers who value us for leading that transformation from [ in Congo ] and silos to reliable centers of data. The data can reside as campus centers data centers, WAN centers or AI centers regardless of their location.

Networking for AI has achieved production scale with an all Ethernet-based Arista AI center. In 2025, we are a founding member of the Ethernet-based standards for both scale up with ESUN as well as completing the Ultra Ethernet Consortium 1.0 specification for scale-out AI networking. These AI centers seamlessly connect the back-end AI accelerators to the front end of compute storage, WAN and classic cloud networking.

Our AI accelerated networking portfolio consisting of three families of Etherlink spine-leaf Fabric are successfully deployed in scale out and scale across networks. Network architectures must handle both training and inference frontier models to mitigate congestion. For training, the key metric is obviously job completion time, the amount of time taken between admitting a job, training job to an AI accelerator cluster and the end of a training run. For inference, the key metric is slightly different. It's the time taken to a first token basically the amount of latency it takes for users submitting a query to receive their first response. Arista has clearly developed a full AI suite of features to uniquely handle the fidelity of AI and cloud workloads in terms of diversity, duration, size of traffic flow and all the patterns associated with it.

Our AI for networking strategy based on AVA, autonomous virtual assist curates the data for higher-level functions. Together with our published subscribed state Foundation in EOS, NetDL, or Network Data Lake, we instrument our customers' networks to deliver proactive, predictive and prescriptive features for enhanced security, observability and agentic AI operations. Coupled with the Arista validated designs for network simulation, digital twin and validation functionality, Arista platforms are perfectly optimized and suited for Network as a Service.

Our global relevance with customers and channels is increasing. In 2025 alone, we conducted three large customer events across three continents, Asia, Europe and United States and many other smaller ones, of course. We touched 4,000 to 5,000 strategic customers and partners in the enterprise. While many customers are struggling with their legacy incumbents, Arista is deeply appreciated for redefining the future of networking. Customers have long appreciated our network innovation and quality, demonstrated by our highest Net Promoter Score of 93% and lowest security vulnerabilities in the industry. We now see the pace of acceptance and adoption accelerating in the enterprise customer base.

Our leadership team, including our newly appointed co-Presidents Ken Duda and Todd Nightingale have driven strategic and cohesive execution. Tyson Lamoreaux, our newest Senior Vice President, who joined us with deep cloud operator experience has ignited our hyper growth across our AI and cloud titan customers. Exiting 2025, we are now at approximately 5,200 employees, which also includes the recent VeloCloud acquisition.

I am incredibly proud of the entire Arista A team and thank you, all employees for your dedication and hard work. Of course, our top-notch engineering and leadership team has always steadfastly prioritized our core Arista way principles, of innovation, culture and customer intimacy. Well, I think you would agree that 2025 has indeed been a memorable year and we expect 2026 to be a fantastic one as well. We are amid an unprecedented networking demand with massive and a growing TAM of $100-plus billion. And so despite all of the news on the mounting supply chain, allocation, rising cost of memory and silicon fabrication, we increased our 2026 guidance to 25% annual growth, accelerating now to $11.25 billion.

And with that, happy news, I turn it over to Chantelle, our CFO.

Thank you, Jayshree, and congratulations to you and our employees on a terrific 2025. As you outlined, this was an outstanding year for the company, and that strength is clearly reflected in our financial results. Let me walk through the details.

To start off, total revenues in Q4 were $2.49 billion, up 28.9% year-over-year and above the upper end of our guidance of $2.3 billion to $2.4 billion. It was great to see that all geographies achieved strong growth within the quarter. Services and subscription software contributed approximately 17.1% of revenue in the fourth quarter, down from 18.7% in Q3, which reflects the normalization following some nonrecurring VeloCloud service renewal in the prior quarter. International revenues for the quarter came in at $528.3 million or 21.2% of total revenue, up from 20.2% last quarter. This quarter-over-quarter increase was driven by a stronger contribution from our large global customers across our international markets. The overall gross margin in Q4 was 63.4%, slightly above the guidance of 62% to 63% and down from 64.2% in the prior year. This year-over-year decrease is due to the higher mix of sales to our cloud and AI titan customers in the quarter. Operating expenses for the quarter were $397.1 million or 16% of revenue, up from the last quarter at $383.3 million. R&D spending came in at $272.6 million or 11% of revenue, up from 10.9% last quarter. Arista continued to demonstrate its commitment and focus on networking innovation with a fiscal year '25 R&D spend at approximately 11% of revenue. Sales and marketing expense was $98.3 million or 4% of revenue, down from $109.5 million last quarter. FY '25 closed the year with sales and marketing at 4.5%, representative of the highly efficient Arista go-to-market model. Our G&A costs came in at $26.3 million or 1.1% of revenue up from $22.4 million last quarter, reflecting continued investment in systems and processes to scale Arista 2.0. For fiscal year '25, G&A expense held at 1% of revenue. Our operating income for the quarter was $1.2 billion or 47.5% of revenue. This strong Q4 finish contributed to an operating income result for fiscal year 2025 of $4.3 billion or 48.2% of revenue. Other income and expense for the quarter was a favorable $102 million, and our effective tax rate was 18.4%. This lower-than-normal quarterly tax rate reflected the release of tax reserves due to the expiration of the statute of limitations.

Overall, this resulted in net income for the quarter of $1.05 billion or 42% of revenue. It is exciting to see Arista delivering over $1 billion in net income for the first time. Congratulations to the Arista team on this impressive achievement.

Our diluted share number was 1.276 billion shares, resulting in a diluted earnings per share for the quarter of $0.82, up 24.2% from the prior year. For fiscal year '25, we are pleased to have delivered a diluted earnings per share of $2.98, a 28.4% increase year-over-year.

Now turning to the balance sheet. Cash, cash equivalents and marketable securities ended the quarter at approximately $10.74 billion. In the quarter, we repurchased $620.1 million of our common stock at an average price of $127.84 per share. Within fiscal 2025, we repurchased $1.6 billion of our common stock at an average price of $100.63 per share. Of the $1.5 billion repurchase program approved in May 2025, $817.9 million remain available for repurchase in future quarters. The actual timing and amount of future repurchases will be dependent on market and business conditions, stock price and other factors.

Now turning to operating cash performance for the fourth quarter. We generated approximately $1.26 billion of cash from operations in the period. This result was an outcome of strong earnings performance with an increase in deferred revenue, offset by an increase in accounts receivable driven by higher shipments and end of quarter service renewals. DSOs came in at 70 days, up from 59 days in Q3 driven by renewals and the timing of shipments in the quarter. Inventory turns were 1.5x, up from 1.4x last quarter. Inventory increased marginally to $2.25 billion, reflecting diligent inventory management across raw and finished goods. Our purchase commitments at the end of the quarter were $6.8 billion, up from $4.8 billion at the end of Q3. As mentioned in prior quarters, this expected activity mostly represents purchases for chips related to new products and AI deployments. We will continue to have some variability in future quarters due to the combination of demand for our new products, component pricing, such as the supply constraint on DDR4 memory and the lead times from our key suppliers.

Our total deferred revenue balance was $5.4 billion, up from $4.7 billion in the prior quarter. In Q4, the majority of the deferred revenue balance is product-related. Our product deferred revenue increased approximately $469 million versus last quarter. We remain in a period of ramping our new products, winning new customers and expanding new use cases, including AI. These trends have resulted in increased customer-specific acceptance clauses and an increase in the volatility of our product deferred revenue balances. As mentioned in prior quarters, the deferred balance can move significantly on a quarterly basis, independent of underlying business drivers.

Accounts payable days were 66 days, up from 55 days in Q3, reflecting the timing of inventory receipts and payments. Capital expenditures for the quarter were $37 million. In October 2024, we began our initial construction work to build expanded facilities in Santa Clara and incurred approximately $100 million in CapEx during fiscal year 2025 for this project.

As we have moved through 2025, we have gained visibility and confidence for fiscal year 2026. As Jayshree mentioned, we are now pleased to raise our 2026 fiscal year outlook to 25% revenue growth, delivering approximately $11.25 billion. We maintain our 2026 campus revenue goal of $1.25 billion and raised our AI centers goal from $2.75 billion to $3.25 billion. For gross margin, we reiterate the range for the fiscal year of 62% to 64% inclusive of mix and anticipated supply chain cost increases for memory and silicon. In terms of spending, we expect to continue to invest in innovation, sales and scaling the business to ensure our status as a leading pure-play networking company. With our increased revenue guidance, we are now confident to raise the operating margin outlook to approximately 46% in 2026. On the cash front, we will continue to work to optimize our working capital investments with some expected variability in inventory due to the timing of component receipts on purchase commitments. Our structural tax rate is expected at 21.5% back to the usual historical rate, up from the seasonally lower rate of 18.4% experienced last quarter Q4 '25.

With all of this as a backdrop, our guidance for the first quarter is as follows: revenues of approximately $2.6 billion, gross margin between 62% and 63% and operating margin at approximately 46%. Our effective tax rate is expected to be approximately 21.5%, with approximately 1.275 billion diluted shares.

In closing, at our September Analyst Day, we had the theme of building momentum, and we are doing just that. In the campus WAN, data and AI centers, we are uniquely positioned to deliver what customers need. We will continue to deliver both our world-class customer experience and innovation. I am enthusiastic about our fiscal year ahead.

Now back to you, Rudy, for Q&A.

Thank you, Chantelle. We will now move to the Q&A portion of the Arista earnings call. [Operator Instructions] Regina, please take it away.

[Operator Instructions] Our first question will come from the line of Meta Marshall with Morgan Stanley.

Great. And congratulations on the quarter. I guess in terms of kind of the commentary you had, Jayshree, on the one or two additional 10% customers. I guess just digging more into that, what are the puts and takes of -- is it bottlenecks in terms of their building? Is it -- like what would make or break kind of whether those become two new additional kind of 10% customers?

Thank you, Meta, for the good wishes. So obviously, if I didn't have confidence, I wouldn't dare to say that, would I? But there's always variables. Some of it may be sitting in deferred. So there's an acceptance criteria that we have to meet and there's also timing associated with meeting the acceptance criteria. Some of it is demand that is still underway. And in this age of all the supply chain allocation and inflation, we've got to be sure we can shift. So we don't know if it's exactly a 10% or high single digits or low double digits, but a lot of variables will decide that final number. But certainly, the demand is there.

Our next question will come from the line of Samik Chatterjee with JPMorgan.

Jayshree, congrats on the quarter and the outlook. I don't want to sort of say that the 25% growth is not impressive. But since you're doing 30% is what the guidance is for 1Q, maybe if I could understand what's maybe sort of leading to somewhat of a cautious in terms of visibility for the rest of the year? Is it the sort of one to two new customers and their ramps that you're sort of more cautious about? Or is it availability of supply [indiscernible] some of the components or memory that's sort of giving you maybe a bit more cautiousness about the visibility for the remainder of the year? If you can understand the drivers there.

Yes. Thank you, Samik. First, I don't think I'm being cautious. I think I went all out to give you a high dose of reality. But I understand your views on caution, given all the CapEx numbers you see from customers.

That's an important thing to understand that we don't track the CapEx. The first thing that happens in the CapEx is they got to build the data centers and get the power and get all of the GPUs and accelerators and the network comes and lags a little. So demand is going to be very good. But whether the shipments exactly fall into '26 or '27, to add, you can clarify when they really fall in, but there's a lot of variables there. That's one issue.

The second, as I said, is a large amount of these are new products, new use cases, highly tied to AI where customers are still in their first innings. So again, I'm giving you the greatest visibility I can fairly early in the year on the reality of what we can ship not what the demand might be. It might be a multiyear demand that ships over multiple years. So let's hope it continues. But of course, you must understand that we're also facing a lot of large numbers. So 25% on a base of now $9 billion when we started last year at $8.25 billion is a really, really early and good start.

Our next question will come from the line of David Vogt with UBS.

Maybe Chantelle and Jayshree, can you help quantify sort of both the revenue impact and potential kind of gross margin impact embedded in your guide from the memory dynamics and the constraints? I know last quarter, and you had mentioned in this quarter, obviously, the supply chain does have some constraints. When you think about -- I think, Jayshree, you said kind of the real outlook that you see, maybe you can help parameterize what you think could hold you back, if that's the way to phrase it? And just give us a sense for what upside could be in a perfect world effectively if you could share that?

I'm going to give some general commentary and Chantelle, if you don't mind adding to it. Our peers in the industry have been facing this probably longer than we have because I think the server industry probably saw it first because they're more memory-intensive. Add to that, that we're expecting increases from the silicon fabrication that all the chips are made, as you know, centrally with one company, Taiwan Semiconductor.

So Arista has taken a very thoughtful approach being aware of this since 2025 and frankly, absorbed a lot of the costs in 2025 that we were incurring. However, in 2026, the situation has worsened significantly. We're having to smile and take it just about at any price we can get and the prices are horrendous. They're an order of magnitude, exponentially higher. So clearly, with the situation worsening and also expected to last multiple years, we are experiencing shortages in memory.

Thankfully, as you can see, reflected in our purchase commitments, we are planning for this. And I know that memory is now the new gold for the AI and automotive sector, but clearly, it's not going to be easy, but it's going to favor those who plan and those who can spend the money for it.

I think the only thing I'd add to your question, David, and thank you for that, is that -- so we're comfortable in the guide, and that's why we have the guide and why we raised the numbers that we did. So we're comfortable we have a path to there within the numbers we've provided. The range of 62% to 64%. I think we are pleased to hold despite this kind of pressure coming into it. This has been our guide since September at our Analyst Day. So we're pleased to hold that guide and find ways to mitigate this journey. Now whether it ends up being 62.5% versus 63.5% in the guide in that range, that's where we'll continue to update you, but the range we're comfortable with.

Our next question comes from the line of Aaron Rakers with Wells Fargo.

Congrats as well on the quarter and the guide. I guess when we think about the $3.25 billion guide for the AI contribution this year, I'm curious, Jayshree, how much you're factoring if any, from scale-up networking opportunity, how do you see -- is that more still the [ 27 ]? And also, can you unpack like ex the AI and ex the campus contribution, it appears that you're guiding still pretty muted low single-digit growth on non-AI, just curious how you see the non AI and non campus growth.

Okay. Yes. Well, rising tide rises all boats but some go higher and some go lower. But to answer your specific question, what was it around?

The scale up.

The scale up. We have consistently described that today's configurations are mostly a combination of scale out and scale up were largely based on 800-gig and smaller ratings. Now that the ESUN specification is well underway. And Ken Duda, you can -- I think the spec will be done in a year or this year, for sure. So Ken and Hugh Holbrook are actively enrolled in that. We need a good solid spec. Otherwise, we'll be shipping proprietary products like some people in the world do today. And so we will tie our scale-up commitments greatly to availability of new products and a new ESUN spec, which we expect the earliest to be Q4 this year. And therefore, majority of the -- we'll be in some trials where a lot of -- Andy Bechtolsheim and the team is working on a lot of active AI racks with scale-up in mind, but the real production level will be in 2027 primarily centered around not just 800-gig but 1.6T.

Our next question will come from the line of Amit Daryanani with Evercore ISI.

Congrats from my end as well for some really good numbers here. Jayshree, if I think some of these model builders like Anthropic that I think you folks have talked about, they're starting to build these multibillion-dollar clusters on their own now. Can you just talk about your ability to participate in some of these build-outs as they happen, be that on the DCI side or maybe even beyond that? And by extension, does this give you an opportunity to ramp up with some of the larger cloud companies that these model builders are partnering with over time as well as you build out [ TP ] or premium clusters. I'd love to just understand how that kind of business scales up for you folks.

Yes. No. Amit, that's a very thoughtful question. And I think you're absolutely right. The network infrastructure is playing a critical role with these model builders in a number of ways. If you look at us initially, we were largely working with one or two models in there and one or two accelerators, NVIDIA and AMD and OpenAI was primarily dominant one. But today, we see that there's really multiple layers in a cake where you've got the GPU accelerators, of course, you've got power as the most difficult thing to get. But Arista needs to deal with multiple domains and model builders and appropriately, whether it is Gemini or xAI or Anthropic, Claude or OpenAI and many more coming. These models and the multiprotocol algorithm or nature of these models is something we have to make sure we build the network correctly for. So that's one.

And then to your second point, you're absolutely right. I think the biggest issue is not only the model builders but there are no more [ in silos ] in one data center, and you're going to see them across multiple colos and multiple locations and multiple partnerships with our cloud titan customers that we've historically not worked with this. So I think you'll see more Copilot versions of it, if you will, with a number of our cloud titans. So we expect to work with them as AI specialty providers, but we also expect to work with our cloud titans and bringing the cloud and AI together.

Our next question comes from the line of George Notter with Wolfe Research.

I was just curious about the product deferred revenue and how you see that coming off the balance sheet ultimately. Obviously, it's just been stacking up here quarter after quarter. So a few questions here. Does that come off in big chunks that we'll see different quarters in the future? Does it come off more gradually? Does it continue to build? Like what does the profile look like for that product deferred coming off the balance sheet and [ flowing ] through the P&L? And then also I'm curious about how much product deferred do you have in the full year revenue guidance to 25%.

Thanks, George. Thanks for the questions. Not much has changed in the sense of how we have this conversation. What goes into deferred is new product, new customers, new use cases, the great new use case is AI. The acceptance criteria for that, for the larger deployments is 12 to 18 months. Some can be as short as 6 months. So there's a wide variety that goes in. Deferred has balances coming in and out every quarter. We don't guide deferred and we don't say product specific. What I can tell you in your question is that there will be times where there are larger deployments, but will feel a little lumpier as we go through. But again, it's a net release of a balance. So it depends what comes in at that same quarter timing.

Got it. Okay. Any sense for what's in the full year guidance? I assume not much? Is that fair to say?

It's super hard, George. It's when the acceptance criteria happens. If it happens December [ 22 ], it's a different situation. If it all happens in Q2, Q3, Q4, that's a difference. So that's something we really have to work with the customer. So sorry that we're not able to be clairvoyant on that.

Our next question comes from the line of Ben Reitzes with Melius Research.

I guess my congrats to guys. This execution and guide is really something. So I wanted to ask about two things that I just was wondering if you could talk a little bit more about your neocloud momentum and what that is looking like in terms of materiality? And then also, if you don't mind touching on AMD with the launch, we're kind of hearing about you getting a lot of networking attached to the 450 type product or their new chips? Wondering if that is a catalyst or not as you go throughout the year?

Yes. So, Ben, as you can imagine, the specialty cloud providers have historically had a cacophony of many types of providers. We are definitely seeing AI as one of the clear -- in the past, it used to be content providers, Tier 2 cloud providers. But AI is clearly driving that section. And it's a suite of customers, some of who have real financial strength and are looking now to invest and increase and pivot to AI. So the rate at which they pivot in AI will greatly define how well we do that. And they're not yet titans, but they want to be or could be titans just the way to look at it. So -- and we're going to invest with them, and these are healthy customers. It's nothing like the dot-com era, we feel good about that.

There are a set of neoclouds that we watch more carefully because some of them are oil money converted into AI or crypto money converted into AI. And over there, we are going to be much more careful because some of those neoclouds are looking at Arista as the preferred partner, but we would also be looking at the health of the customer or they may just be a one kind. We don't know the exact nature of the business. And those will be smaller and they don't contribute in large dollars, but they are becoming increasingly plentiful in quantity, even if they're not yet in numbers.

So I think you're seeing this dichotomy of two types in that category or three types. The classic CDN and security specialty providers, Tier 2 cloud, the AI specialty are going to lean in and invest and then the neoclouds in different geographies.

AMD?

Yes, the AMD question. A year ago, I think I said this to you, but I'll repeat it. A year ago, it was pretty much 99% NVIDIA, right? Today, when we look at our deployments, we see about 20%, 20%, maybe a little more, 20% to 25%, where AMD is becoming the preferred accelerator of choice. And in those scenarios, Arista is clearly preferred because they're building best-of-breed building blocks for the [ NIC ] for the network, for the IO and they want open standards as opposed to a full on vertical step from one vendor. So you're right to point out that AMD and in particular, is a joy to work with [ Lisa and Forest ] and the whole team, and we do very well in that multi-vendor open considerations.

Our next question will come from the line of Tim Long with Barclays.

Yes, appreciate a little color. Jayshree, maybe we could touch a little bit on scale across. It's obviously gotten a lot of attention, particularly on the optics layer from some others in the industry. Obviously, you guys have been in DCI, which is kind of a similar type technology. But curious what you think as far as Arista's participation in more of these next-gen scale across networks? And is this something that would be good for like a blue box type of product? Or would that more be in the scale up? So if you could give a little color there, that would be great.

Right. Okay. So most of our participation today, we thought would be scale out. But what we are finding is due to the distributed nature of where and how they can get the power and the by-sectional bandwidth growth where essentially the throughput scale out or scale across is all about how much data you can move, right? As the workloads become more and more complex, you have to make them more and more distributed because you just can't fit them in one data center, both from a power bandwidth throughput capacity.

Also, these GPUs are trying to minimize the collective degradation. So as you scale up or out, the communication patterns become very, very much of a bottleneck. And one way to solve it is to extend this across data centers, both through fiber. And as you rightly pointed out, a very high injection bandwidth DCI routing.

And then there's a sustained real-world utilization you need across all of these. So for all these reasons, we are pleasantly surprised with the role of coherent long-haul optics, which we don't build but we have worked in the past very greatly with companies that do, and they're seeing the lift. And the 7800 Spine chassis as the flagship platform and preferred choice that has been designed by our engineering team now for several years for this robust configuration. So let's blue box there and much, much more of a full-on Arista flagship box with the U.S. and all of the virtual output queuing and buffering to interconnect regional data centers with extremely high levels of routing and high availability, too. So this really lends into everything Arista stands for coming all together in a universal AI spine.

Our next question will come from the line of Karl Ackerman with BNP Paribas.

Yes. Agentic AI should support an uptake in conventional server CPUs where you have -- where your switches have high share within data centers. And so given your upwardly revised outlook of 25% growth for this year, could you speak to the demand prospects you are seeing for front-end high-speed switching products that address agentic AI products?

Yes. Exactly, Karl. I think in the beginning -- well, let's just go back in time in history. It's not that long ago. Three years ago, we had no AI. We were staring at [indiscernible] being deployed everywhere in the back end. And we pretty much characterized our AI as only back end, just to be pure about it, right? Three years later, I'm actually telling you we might do north of $3 billion this year and growing, right? That number definitely includes the front end as it's tied to the back-end GPU clusters. And it's an all Ethernet, all AI system for agentic AI applications.

Now a lot of the agentic AI applications are mostly running with some of our largest cloud AI and specialty providers. But I don't rule out the possibility. You could see this in our numbers with loads of 8,800 gig customers that many of that is going to feed into the enterprise as well as agentic AI applications come for genomic sequencing, science, automation of software, I don't know -- I don't think can any of us believe that AI is eating software, but AI is definitely enabling better software, right? And we're certainly seeing that in [indiscernible] as well in our adoption of that. So the rise of agentic AI will only increase not just the GPU, but all gradations of XPU that can be used in the back end and [ funded ].

Our next question comes from the line of Simon Leopold with Raymond James.

I wanted to come back on the issue around sort of what's going on with the memory market. So two aspects to this is, one, I'm wondering how much of a tool has been price hikes, you raising your prices to customers or and/or whether or not within the substantial amount of purchase commitments you have, whether there's a significant aspect of memory in there. So you've prepurchased memory effectively at much lower prices in the spot market today.

Thank you. Okay. I wish I could tell you we did purchase all that memory that we needed. No, we didn't. But [indiscernible] in the industry have done multiple price hikes already, especially those in the server market or memory intensive switches, we have clearly been absorbing it and memory is in our purchase commitments. But -- so as everything else, the entire silicon portfolio is in our purchase commitments.

Due to some of the supply chain reactions, Todd and I have been reviewing this, and we do believe there will be a onetime increase on selected especially memory-intensive SKUs to deal with it, and we cannot absorb it if the prices keep going up the way they have in January and February. And I would tell you that all the purchase commitments I have in my current and Chantelle's current commitments are not enough. We need more memory.

Our next question will come from the line of James Fish with Piper Sandler.

Ladies, great quarter, great end to the year. Jayshree, are hyperscalers getting nervous now at all in ordering ahead? What's your sense of pull-in of demand potentially here, including for your own Blue Box initiative?

And Chantelle tell for you, just going back to George's question, are you -- I know it's difficult to answer, but are you anticipating that, that product deferred revenue is going to continue to grow through the year? Or just -- it's way too difficult to predict and you've got customers that could just say, "We accept, ship them all now." and so we end up with a big quarter, but product deferred down.

I'm going to let Chantelle answer the difficult questions [indiscernible] over again.

Happy. Thank you, James. I appreciate it. So I think for deferred, generally is -- so we don't guide deferred, but to try to give you more insight, there will be back to George's question, there will be certain deployments that get accepted and released but the part that's difficult is what comes into the balance, right, James. So I can't guide that would be a wild guess on what's going to go in, which is not prudent, I think, from my perspective. So we'll continue to mention what's in it. We'll continue to show you through the balances. We'll talk about it in the script in the sense of the movement. But that's probably as much as I can tell you with a responsible answer looking forward.

James, this is one of those times no matter how many times you ask this question in several different ways, the answer doesn't change.

I mean we're all -- [indiscernible] is doing the same thing over and over again.

Yes. So on the hyperscalers, I don't think they're getting nervous. You've seen what a strong business they had, how much cash they put out and how successful they are. But I just think they are working more closely with us. Typically, we had a 3- to 6-month visibility. We're getting [indiscernible].

Our next question will come from the line of Tal Liani with Bank of America.

I almost had the same question to you what I asked you last quarter because you grew -- increased the guidance. Yes. No, it's -- I'll explain. You increased the guidance, but the entire increase in the guidance is basically the cloud. The -- and if I look at -- it's very simple to the [ sector ] numbers. If I remove campus and I remove cloud and you provide these two numbers for both '25 and '26, the rest of the business, which is 60% of the business, you guided to grow zero. And in previous years, it was -- I can make estimate, it was anywhere from 10% to 30% growth. So the question is, why are you guiding this way that 60% of the business is not going to grow. Is it because it's just conservatism?

Now, can I pause you there because I know you like to dissect our math several different ways and come up with conclusions. We're not guiding that our business is going to be flat or we're not going to grow here or grow there. But generally, when something is very fast paced and growing, then other things grow less. And exactly whether it would be flat or grow double digits or single digits [indiscernible] -- it's February. I don't know what the rest of the year will be, okay?

No. But that's the question. The question is, is there allocation here? Meaning if you -- let's say you have only set a number of memory slots, so you allocate it to cloud and then the rest of the business doesn't get it or is it just conservatism and lack of ability to [indiscernible] availability?

It's neither of the above. It's -- we don't allocate to our customers. It's first and first served. And in fact, the enterprise customers get a very high sense of priority as to our cloud. Customers come first. So -- but allocation of memory may allow us to be in a situation where the demand is greater than our ability to supply. We don't know. It's too early in the year. We're confident that we could guide 6 months after our Analyst Day to a higher number, but we don't know what the next 4 quarters will look like to the precision you're asking for.

Our next question comes from the line of Atif Malik with Citi.

It's [ Adrian Colby ] for Atif. I was hoping to ask about for an update on Arista's four large AI customers. I know that, that fourth customer, you talked about was a bit slower to ramp to 100,000 GPUs. Just wondering if you can update us on their progress there. And perhaps what's next for the other few customers that have already crossed that threshold.

And lastly, is there any indication that, that fifth customer that ran into funding challenges might come back to you?

Okay. Adrian, I'll give you some update. I'm not sure I have precise updates, but we are in all 4 customers deploying AI with Ethernet. So that's the good news. Three of them have already deployed a cumulative 100,000 GPUs and are now growing from there. And clearly migrating now into beyond pilots and production to other centers, power being the biggest constraint.

Our fourth customer is migrating from InfiniBand so it's still below 100,000 GPUs at this time, but I fully expect them to get there this year, and then we shall see how they get beyond that.

Our next question will come from the line of Michael Ng with Goldman Sachs.

I just have one and one follow-up. First, I was wondering if you could talk a little bit about the new customer segmentations that you guys unveiled with cloud and AI and specialty. What's the philosophy around that? And does that kind of signal more opportunity in places like Oracle and the neoclouds?

And then second, with cloud and AI at 48% of revenue and AMD had a combined 36%, you have 12% left over. Is that a hyperscale customer, because it does kind of imply that you have a new hyperscaler that is approaching 10% because obviously, we thought that the next biggest one would have been Oracle, but that's moved out of cloud now. So any thoughts there would be great.

Yes. Sure, Michael. So -- Well, first of all, my math is 26% and 16%, so 42%, so I don't have 12% unless you had 58%. It's really only 6%.

So on the cloud and AI tightness, the way we classified that is a significantly large scale customers with greater than 1 million servers, greater than 100,000 GPUs, an R&D focus on models and sometimes even their own XPUs. And this can, of course, change. Some others may come into it. But it's a very select few set of customers, less than 5 or above 5, that's the way to think of it. right?

On the change on the specialty cloud, as I said, we're noticing that some customers are really, really focused solely on AI with some cloud as opposed to cloud with some AI. So when it's a $0.70 AI-centric especially with Oracle's AI [ Acceleron ] and multi-tenant partnerships that they've created, they have naturally got a dual personality some of which is OCI, the Oracle Cloud, but some of it is really AI -- fully AI based. So we -- the shift in their strategy made us shift the category and [indiscernible] to keep it too.

Our final question will come from the line of Ryan Koontz with Needham & Company.

Jayshree, in your prepared remarks, you talked about your telemetry capabilities. I wondered if you could expand on that and discuss where are you seeing that key differentiation? What sort of use cases you're able to really seize the upper hand competitively with your telemetry capabilities?

Yes. I'm going to say some, and I think Ken who's been designing this and working on it will save even more. Ken Duda, our President and CTO. So telemetry is at the heart of our -- both our EOS software stack as well as our CloudVision for enterprise customers. We have a real-time streaming telemetry that has been with us since the beginning of time, and it's constantly keeping track of all us, which is it isn't just a pretty management tool. And at the same time, our cloud customers and AI customers are seeking some of that visibility too. And so we have developed some deeper AI capabilities for telemetry as well.

Over to you, Ken, for some more detail.

Yes. No, thanks for that question. That's great. Look, the EOS architecture is based on stable orientation. This is the idea that we capture the state of the network and [indiscernible] out from the system database on the switches into whatever the television or whatever system we can then [ receive ] that, and we are extending that capability for AI with a combination of in-network data sources related to flow control, RDMA counters, buffering and congestion counters and also host level information, including what's going on in the RDMA stock on the host, what's going on with collectives, latencies, any full control problems or buffering problems in the host NIC and we pull those -- that information altogether in CloudVision and give the operator a unified view of what's happening in the network, and what's happening in the host. And this greatly aids our customers in building an overall working solution because the interactions between the network and the host can be complicated and difficult to debug when it's different systems collecting them.

Great job, Ken. I can't wait for that product.

Thank you.

This concludes Arista Networks Fourth Quarter 2025 Earnings Call. We have posted a presentation that provides additional information on our results, which you can access on the Investors section of our website. Thank you for joining us today and for your interest in Arista.

Thank you for joining. Ladies and gentlemen, this concludes today's call. You may now disconnect.

Hello, everybody. Tim Long, Barclays IT hardware comm equipment analyst. Thank you so much for joining. Very happy to have Arista Networks. With us, we have Mark Foss, who's Operations and Marketing; and Hardev Singh, GM on the Cloud and AI products. So we're getting a few different pieces of the business, a few different perspectives here. Thank you so much, guys, for joining.

Thank you. Thanks for having us.

So let's -- yes, we're going to bounce around a little bit, but -- maybe if we start more, Hardev, probably for you on when we're looking at Cloud and AI networking, there's a lot of debates about competition. So we've always been talking about white box and it's always been there, but it's not going away as a discussion point. And obviously, over the last several years, given what NVIDIA has done with Mellanox, they've increased their presence a lot first with InfiniBand, now somewhat with Ethernet. So maybe if you can just kind of how you view competitive differentiation and dynamics and what is it that helps Arista win?

Sure. I mean, I'll start by saying that the AI momentum is strong, right? There's a lot of activity in this AI space, and we're very excited about it. I think our differentiation and our value hasn't changed from the past. We are -- with AI, you just have a different set of requirements. you have the scale up, let me frame the -- where we really play today, right? So if you're looking at scale up with NVIDIA, that's a closed system right now we don't play. So most of where we are playing now is the scale out which is the network to connect all these accelerators.

And now also, you hear about scale across, which is once these cluster sizes become really big, you need to connect them to each other. It's DCI in the traditional world, and that's another segment of that AI space where we play very, very strongly because that's routing, that's a chassis product for us. But what we're also excited about is the non-NVIDIA accelerator ecosystem picking up, right? So that opens up more opportunities for us for scale-out for sure as well as scale across, but also potentially in the future for scale up, right? So these non-NVIDIA accelerators you have the scale up, which is going to be Ethernet, whether it's ESUN or UALink the flavors of Ethernet, that opens up a new TAM for us. Obviously, I don't think it's material in '26, it's more of '27. '26 will be more trials, pilots, but that's another opportunity we're excited about.

And in terms of our differentiation, Arista is a software company, and it all starts with EOS, which is our software. And it's a really unique operating system. It all starts from the architecture, multi-process state sharing architecture very, very high quality. It's been deployed in some of the largest networks on earth, automation, visibility, single binary image, that's our main differentiation.

And then the hardware side, we use off-the-shelf chips, but the way we design our hardware is very efficient. So our power draw is generally about -- could be about 25% less than the equivalent products out there, which adds up when you're deploying a lot of products at scale and power being a big top of mind issue. It really helps to have a lower power draw switch.

Okay. you mentioned scale up and scale across. Maybe let's start with scale across. I was talking to you before, Ciena talking about like some more optical pluggable wins. How do you see Arista from a product standpoint playing? And maybe talk about whether it's latency, speed, distance, where is it where there's going to be a higher content of Arista in that scale across architecture?

So scale across is a new term, but traditionally, we call that DCI, which is data center interconnect. In an AI world, it is connecting a bunch of AI data centers that are physically in a metro region or across. So very excited about that opportunity. We play very strong there with our EOS feature set as well as our 7800 chassis platforms. You mentioned coherent technologies. So once the distance has become large, you need coherent optics like a ZR technology is what most customers use and are looking at today at 400 gig going to 800 gig.

Yes, if the distance becomes very large, then you have a latency impact, so that needs to be kind of architected well that your AI workloads are not going across these long links too much, but that can be solved from an architecture perspective. But yes, I think from a scale across, I think we are a very dominant player there.

Okay. Great. And then maybe the second topic is scale up, and we'll talk a little bit as blue box because I think that's part of the target audience for that technology. I know you guys have had this blue box type of deployments already. But maybe walk us through how those products are different from a full OS switch and how they're different from an EMS, ODM switch from like a Celestica or whatever. Talk to us about the software and hardware differentiation with both the higher-end switches and the lower-end switches?

I mean, I touched on the scale up slightly earlier in terms of the opportunity for us would be from the non-NVIDIA accelerators. In scale up, it's really -- the complexity would be in the hardware design because every accelerator has different technical characteristics, right, the link budget, the signal integrity, the HBMs that are connected to these accelerators. So it's really a custom switch. You can compare it to an Envy linked switch, right? It's like that's what that product looks like.

From a software feature set, you could say it's slightly light, but then the few features that you need reliability, congestion control, latency are super important. So slightly light in feature set, but those have to be very robust, those features and way more complex from a hardware design. So way we think about it, I think it plays into Arista's advantage because we are good at doing complex hardware designs.

The blue box today is mainly deployed in front-end networks at really a handful of players out there. And the real differentiation there is you have diagnostics, better signal integrity, better power draw. There's probably a few others I'm missing, but that's generally the consensus there then they can use a third-party operating system on top of that.

Okay. But I'm -- so I mean back to the scale up. So this is as far as guideposts when we see all this TPU growth or Trainium, that's going to most likely push -- that's going to be a big impetus to push scale up to Ethernet, and that opens up. Is that the right way to think about it?

So net positive there will be a pull-through effect for Ethernet vendors, yes.

Okay. Got a lot more in Cloud, but maybe we'll go over to campus for a little bit.

And I have that one campus question...

Here's your first. Let's see if we can do it. Really good guidance, 50% plus growth for next year. So kind of accelerating after a few good years, which is normally not what happens in the campus world. So maybe walk us through how that's happening. And still very low market share. So how much runway is ahead?

Yes. Sure. Arista started when we started shipping product in '08, we were focused on data center and large Cloud, largest networks on earth. And when we got to like 2018 time frame, many of our customers came to us and said, "Hey, we love the quality of your software, the automation is unprecedented. We love CloudVision. Do you have the visibility that we have in the network." Can you please develop some products for the campus to connect the users. So we listen to our customers. And in 2018, we started delivering POE switches. We acquired a company that did WiFi. And we slowly but surely, we've been shipping away at share and we're targeting kind of the high-end enterprise.

So if you look at the campus market, which is about $30 billion market, about half of that is kind of the large enterprise, which is kind of the Global 2000, if you will. So our product set is really focused on that area -- that segment of the market. And we slowly chipped away there. And we found that customers were really looking for a viable second vendor alternative. Generally, Cisco is the big player in the large enterprise, and they've got about 75% share there. And customers are really looking for a good viable opportunity -- a viable alternative there that had high-quality software and automation and visibility. So Arista, they kind of came to Arista with open arms.

And -- we've -- we're slowly gaining share. We're probably about -- of the overall campus market, we're probably like 2.5%. If you look at the large enterprise segment, we may be approaching 5%. If you look at our data center share, right now, we're about 3%. So could we eventually get there in campus? Anything's possible. It probably will take longer to gain that much share in campus just because in the data center market, there's -- if you win a couple of these big clouds, you can get 10% share pretty quickly, but there's no cloud titan in campus. So it's kind of -- you've got a nickel and dime your way up there. But right now, it's been high growth and primarily because of share gain is really what it's been.

Yes. And there was a new announcement this week, incorporating a little bit more AI? Talk about that a little bit?

Yes. No. Just it was a couple of things. It was more about increasing the domain for mobility. So I think we could support up to 500,000 endpoints. And then also we have this autonomous virtual assistant, which really came from -- we got that from our -- 1 of our security acquisitions, and we're applying that now to campus network. So -- it's been really good.

Okay. All right. Great. We touched on the $1.25 billion campus, maybe back to the AI and the $2.75 billion target, another one with really good growth. Maybe a little bit about -- obviously, you have 2 very large customers and 1/3 that's probably getting very large. You've discussed these 4 deployments, the scale deployment. So maybe give us an update there and how do you think about Arista's positioning as we go to more dense deployments and bigger GPU counts, does that advantage the EOS? Give even more advantages in those type of deployments?

Yes, I think the growth we're showing from this year's number on AI and then going to $2.75 billion is a very healthy increase. Yes, you have the large guys building these massive clusters, 100,000 plus. But I think there's also equally strong momentum on like the Neoclouds or the Tier 2 clouds, baby cloud, whatever you want to call them, as well as enterprise. I think some of the verticals like financials or automotive, research, they are seeing use cases for building AI clusters. Yes, they're going to be significantly smaller in size compared to the big guys, but they all add up.

I think Jayshree in the last earnings call mentioned we have 20 to 30 new customers in this second-tier AI customers. So strong engagement. With these guys, the value of EOS becomes even more important, right? They are going after the same AI use cases, whether they're providing AI as a service or they're doing their own AI workloads, the robustness of the network is key, and they want to pick the best-of-breed network. And they don't have probably an army of engineers like the hyperscalers do. So they depend on like, for example, Arista even more to kind of help them build that AI network at the quality they want to build in.

And for these large cloud networks, we've been working with these guys for like a decade or even more on building out large scalable clouds. And so we were really the logical choice to when they started investing in AI, then they needed an Ethernet portion of that for interconnect and their GPU. So they know us very well and they love our operating systems. So we were the logical to be selected for the Ethernet interconnect.

And just to add, the engagement model is also slightly different with the Tier 2 clouds or enterprise, they have a requirement, there's a time to market, all right, Arista or anybody else, do you have the network? Can we get this up and ready and get it going. The hyperscalers is very different. They're always at the cutting edge, bleeding -- trying to adopt bleeding-edge technology. So with them, it's close partnership. You work with them on these requirements. You kind of almost co-develop with them. then you kind of reach a stage of, okay, it's an RFP because they need to have dual vendor in whatever role there they want to go after, and then you can probably get a design win then you kind of codevelop the product with them and then you can ship, deploy and all of that.

So very different engagement model. And I think the partnership we've had with these hyperscalers is close trust goes a long way. This value, not just in the product when we talk about hardware and software, but there's a lot of value they see in the partnership and the history of sometimes maybe guiding them in the right direction of choosing technology or defining new products with them.

Okay. Great. Yes, I would love your perspective, one of the things that -- to the casual observer goes unnoticed is comparing revenues with other companies, and you guys obviously have a very large deferred product revenue balance. If you can talk a little bit -- I've covered this space for a long time have seen this stuff before with newer technologies. But can you kind of just level set with us what is causing that? What are the features or specifications that the customer is looking to see in the network before it could be recognized?

I think every customer is different, to be quite honest. And as we've been growing and we're deploying, I guess, larger projects. We're getting the acceptance terms can be a little bit longer sometimes like our acceptance terms can range anywhere from 30 days to 18 months. And the deferred balance is growing, but there's stuff that's going in and out every quarter. So it's like the same accounts that are in there are the same projects which are in there every quarter. Everything goes in and out. But I think it's a function of the growth, and it's a function of just larger, more complex projects what it is.

Okay. And then maybe sticking on the data center and AI, a little bit on the go-to-market. So -- could you touch on maybe like Neoclouds and sovereigns that are getting a lot more attention. Are those harder to get those at bets? How are you making sure that you're involved because some of the competitors there like a Cisco or an NVIDIA might have a country presence or whatever for a long time. So maybe walk us through that part for some of the newer customers that could be some decent scale?

Yes, you've got to keep your eye out for these because sometimes these could pop up quick. You've everyone heard of these guys. So -- but our -- we have some salespeople that are really savvy. They are constantly looking at the news, they're constantly doing research and figuring out where this spend is happening and where these Neoclouds could be popping up. So we do a pretty good job of jumping on these. But yes, these can come up pretty fast and they can -- but our model is really direct touch to these guys because these are oftentimes a very complex deployments. And -- but yes, we just -- you got to keep your eyes open because they come up fast.

And then how about the -- obviously, NVIDIA has the benefit of selling a full rack and particularly with probably less technical savvy customer bases. It's easy to just -- even if their networking is not as strong as yours to bundle it. How does that dynamic work? How do you fight that dynamic? I'm sure the more sophisticated ones are making independent decisions.

Yes. Yes. No, for sure. And it's been a very valid strategy of NVIDIA. I think generally, at the hyperscalers, they're generally -- it's more difficult to kind of deploy a bundling strategy with them. They're very savvy and they've got a lot of buying power, so they will generally go with what they want to. I think the smaller companies definitely are more prone to -- with the bundling strategy. But a lot of these Neoclouds and stuff, a lot of these engineers, they've been around the block. They've come from the hyperscale a lot of time and they move around.

So these guys know what they're doing. So oftentimes, they don't allow that. But it definitely -- the bundling strategy definitely can work. But generally, the more sophisticated and experienced the networking people are generally, they will generally choose to disaggregate the Ethernet.

And I think the trend of InfiniBand now really going down, Ethernet...

Yes, it's going to...

That I think helps Ethernet players, including us, and gives customers the smaller ones, even more confidence to adopt Ethernet.

Now is that fully because the consortium work together to match technical capabilities with InfiniBand and maybe customers don't want a proprietary type solution? And is it the sense that there still will be places where InfiniBand makes sense. But in general, it's going to be a down ward trend?

Combination, a bit of that, a bit of them seeing the big guys deploy it and scale and talk about it publicly. So it's a combination of a few things, that they feel very confident. And I think also NVIDIA, talking more and more of Ethernet spectrum. So there's a change in the tune there as well. So I think it's a combination of things, like Mark said, you guys are smart. They move around.

These guys like the dual source as well. So if you have a proprietary technology, you can't dual source it and they don't want to get locked in there. They want to have choice to be able to have leverage against their vendors and stuff like that.

Okay. Maybe, Hardev, talk to us a little bit about kind of speed migrations, 400 to 800, 800 to 1.6. How do you see -- where are we in that continuum now? I know some customers are going at different paces for sure, led by some of the very large hyperscalers. Sometimes it's led by hyperscales that you're not a major supplier to, so it might look different. But like how much do those transitions matter? And do you generally see those transitions as opportunities to take more share or get involved in more use cases. Maybe just walk us through the dynamic?

I'd start by saying 2025 is done now. 2026 is probably the year of heavy 800-gig deployment, right? AI, and I'm talking specifically AI. And -- but I think it's even more important with AI to adopt the next gen as fast as you can. So we, as a company, can be first from a time to market to get the next speed out that actually gives you an advantage, a head start because you will get into those qualifying cycles early, big or small or medium-sized customers and then we just grow from there.

So I think, yes, we are -- Andy talks about it. So we're heavily investing in our next-gen 1.60 based products. There are new technologies that come along with that. There's rack scale infrastructure, there's liquid cooling. There are technologies around how to bring the power down. So LPO at 1.60, we still feel confident we can have LPO to work, this co-package copper, the CPO, all valid technologies that are under evaluation. And then I think that's one way to show differentiation. So I'll sum by saying, yes, getting to the next speed first and fast is a differentiation.

And one thing I'd like to add is it's really good to be qualified at the lower end speed with your software already because most of the 90% of qualification is related to software. So if you're already qualified at say, 400 gig, when you move to 800 gig, that goes way fast, right? Like we saw -- as soon as the 800-gig products were available, we saw a really fast transition from people deploying 400-gig in AI to 800 gig, it almost happen over night for the new deployments as soon as it was there, people just moved over. But -- and that was because people had already qualified the software.

And we believe in shipping first and not announcing first. So that's been a consistent strategy.

Interesting angle to take. You did mention liquid cooling. I'm curious. I think I know I was at your Analyst Day, and I think there was a slide that showed like a liquid cool switch. It was probably up there. Briefly what's the timeline there? What's the savings? What's the power savings? What's the differentiation? And do you think that's going to become as big topic for switching as it has for, say, servers?

I think liquid cooling is going to ramp. It's going to ramp fast, but I think it's a function of the data center builds. So if I'm a customer, for example, who's planning to build these data centers, they can take time, right? Location, power, construction, all of that. And if I'm building my next gen data center that's going to go live in '27 or '28, I probably want to have it 100% liquid cooled because I can save all the ambient air con, all that effort and design in different. At that point, the network has to follow compute. The compute today is already their liquid cool, the NBL 72s or even the other accelerators.

So at that point, the network rack will also be -- there'll be 2 things. We'll go at a rack scale level. There's no more a switch. It's a rack of switches, fully liquid cooled and then they'll complement the compute, which is already there. So you'll have 1 liquid cool infrastructure in the data center and the network is there. So don't confuse by saying everything is going to be liquid cooled. I think there'll be a combination of air cooled and liquid cool because their data center investment is already there where air cooling is there. So it'll probably start off with air cool and liquid cool and then eventually liquid cooling takeoff.

Okay. You mentioned LPO and CPO. We get questions a lot about the impact of those technologies as well as OCS. If you could just talk about, one, the different interconnect and then, two, OCS, how you see that impacting the switching routing markets?

LPO, again, is probably thought leadership from Arista to the industry, right? Andy Bechtolsheim is the one who came up with this concept technology, got the industry to adopt it. And then now in 800 gig, it's in high-volume deployment, right? You're probably talking to the optic suppliers. And it has a big benefit to the customer from a CapEx cost, which is they're cheaper because there's no DSP, and OpEx cost because it's lower power, so I spend less money on power, which I could actually put more compute in, right? We are confident that even at 1.60, we can have -- we can get LPO to work. So we're working on that.

CPO is another promising technology. It's been there, this is probably second or third or fourth generation. It's again, addressing the same thing, which is how do I reduce my cost and power. There are a few things to solve around serviceability and stuff. And I think the industry will get there. So we are -- we will embrace when that maturity comes in. And there are other technologies too, co-package copper. CPO also has use case and scale up in the future, not just scale out, right? So promising technology.

You mentioned OCS. OCS is -- Google has done a great job at probably years of investments to get to the stage they are. How we see that adopted outside of Google to be seen. I would be a bit -- yes, I don't know how much time that would take because there's a heavy engineering lift from a software side to get the technology and more importantly, you need to have the scale to make that work, right?

Okay. Yes. Great. Maybe we'll probably have time for 1 more, Mark. for you with your operations hat on. Company was early on with purchase commitments. And I know, 5 years ago, some of us are saying, "Wow, what these numbers are massive. What are they doing? And they were needed and they're still needed." So maybe talk to us about where you are in capacity and component availability and what's the risk that there's some disruption in the next year or so because the demand is just so high.

Yes. Yes. We try to do our best to mitigate risks. I think one of the biggest purchase commitments we have right now is with the chips because the chips are like anywhere from 38 weeks or 52 weeks. So we're constantly having to order these preorder these. And I think one of the big challenges is bigger in the mix a year in advance. You got to figure out because all these different ship architectures got to figure out which is going to be the highest volume 1 a year from now and it's kind of an art more than a science.

Then the rest of the supply chain, there's -- compared to like when we saw the whole COVID situation where the demand went down and there was supply chain issues, now the demand is really high, so it's a little bit different. So there's -- there's constantly little challenges here and there in the industry. There'll be a shortage here shortage there, and we jump on those, and we'll buy from third-party brokers every now and then if we need to, and we just do our best. But yes, it is -- there is continuing to be a lot of demand out there, and we're just managing the best we can.

Okay. Great. Well, I think we're up against time. So thank you both so much for joining. Thank you, everybody.

Thank you, thank you.

Thank you.