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

Good afternoon, everyone. Before we get started, if you are a member of the press or media, please disconnect at this time. This is a restricted line. Any unauthorized party in this meeting or any unauthorized use of the information communicated in this meeting is subject to prosecution to the fullest extent of the law. Any unauthorized person, including the media that is on the line at this time, Please disconnect. Please note, today's call is being recorded.

Perfect. Good morning or afternoon, everybody. Thanks for joining us. I'm Aaron Rakers. I'm the IT hardware and semi analyst here at Wells Fargo. I have the pleasure of hosting a fireside discussion with Anshul Sadana, the Chief Operating Officer at Arista. Before we go into questions, I just want to mention that if any of you want to ask a question, there's a ask a question button on your screen or feel free to e-mail me at [email protected]. So with that said, Anshul, I truly appreciate you taking the time and talking out helping us learn a little bit more about the Arista story. Thank you so much.

Thank you, Aaron. Thanks for having me here.

So I want to start with kind of open-ended question. Arista has done phenomenally well. The company is executing great. What I want to ask is that as you speak with investors, as you've even spoken with some customers, what do you still think is -- are there areas where you say, hey, people just don't quite get it yet. There's underappreciated elements of the Arista story. And maybe that will kind of segue into some other questions. But I want to start there because it's -- you guys have performed like I said, very, very well.

Sure, Aaron. Well, the whole market knows us for data center and campus switching. We do some routing, we do DCI. So a lot of comparison on speeds and feeds that the last 2 years, I can't count how many times people have talked about 400 gig and tools ahead and someone is building a new chip here and there and so on. But what people forget is the sticky part of the solution is EOS, the operating system, CloudVision, the automation suite. And that is really, really sticky. So as a result of that, customers love what we're building and they crossed their head and say, why didn't they use Arista earlier, what are we waiting for, and they rarely very, very rarely would ever go back. That stickiness, I think, is quite underappreciated.

And you would argue -- I'll touch on enterprise. I'll touch on cloud certainly in the discussion. But you would say that stickiness resonates in any end customer, any vertical that you obviously talk about in your earnings calls and disclose. But that resonates at EOS stickiness that CloudVision stickiness resonates the entire customer base? Or would you say there's areas where maybe that stickiness is a bit more pronounced and competitively advantaged?

I think it actually works across the board, across all of our segments and verticals. Everyone appreciates a good product that just works.

Okay. Talk a little bit about enterprise. I think that's one of, I would assume still as people -- I think one of the first questions you oftentimes get on an earnings call is what's going on in cloud and how quickly is cloud deploying 400-gig and so on and so forth. But enterprise seems to be the area that has definitely stood out this past 12, if not 24-plus months. Can you talk about what you're seeing in the enterprise customers where and why Arista is winning and how you're enabling the ability to sell a broader portfolio of solutions across that enterprise customer?

Yes. So Aaron, we have more than 7,500 customers today. We have maybe 5 to 10 sort of cloud titans. We have some service providers. We have some Tier 2 cloud. But once you remove the risk, you have [indiscernible] , you have other verticals as well, but once you remove the risk there's roughly over 6,000 enterprise customers that we have today, which is our largest customer base by country. And this investment has been going on for several years. The sort of feature set and the maturity that the enterprise needed for us to catch up to displace the incumbent they have been using for years, that journey has been on at least for the last 7, 8 years. But I would say about 5 years ago, we really started to focus from a go-to-market sales perspective as well. And I have kept on expanding that continuously. So today, Enterprise is our fastest-growing segment. It has been consistently for the last many, many quarters, and we believe that will remain the case even in the future. The cloud is volatile. There's growth cycles and there's slowdowns and so on. But Enterprise is a very steady growth journey. In addition, the enterprise is especially large enterprise has been underserved. There just wasn't enough competition. There wasn't a better solution. Many companies have come to market you remember the 4 Stens of the world and Belief BNTS of the world and many other companies have tried. But there's go-to-market for us, hey, we built a better widget for now. We may not be better all the time, but it's better, but it's cheaper, please buy it. You do not win in the enterprise just on price. You have to be competitive on price, but you really need a mature product offering that's better than what the industry is used to. And this is where things like CloudVision, do come into play. It's a significant automation suite for our customers. We have customers who have already come and told us that we have a semi company that runs their fab on Arista and they've been doing this now for over a year and their uptime has improved by more than 3x compared to where they used to be in the past. They just have too many outages, too many downtimes and so on. That has completely changed. They don't have to worry about downtime. And in today's market, you know how precious subfab cycle is. And they're thankful that they switched to Arista before this emergency situation came in front of them.

Yes. And you mentioned it a little bit in the answer to your question, but that build-out of feature functionality suite, CloudVision and layering on top of that, the software richness of Arista's [indiscernible]. Are there areas should I think about that further open up enterprise opportunities for you in the future as that continually deepens? Are there other big tranches of like, hey, this is an opportunity that maybe today we can address that pay as this portfolio evolves and it opens up additional enterprise customers?

If you look at enterprise in 2 very large buckets, one is data center, the other is campus. On the data center front, I think we covered it really well. There's always a next gen to worry about. So every 2, 3 years, there's next gen products, but we have over 50 configurations that we sell, all running the same Arista EOS operating system managed with CloudVision, either on-prem or in the cloud. Campus, as you know, we entered just a couple of years ago. And over time, we'll just keep on expanding that portfolio as well. So as we do that, I think going from very large enterprise to large enterprise to mid-market. And I think that journey will also continue to go on and we'll keep expanding our portfolio.as well as our go-to-market channels and sales coverage in the enterprise, especially for campus.

Yes. Yes. Shifting gears to the cloud business. You guys, I think -- people have talked about 50% cloud traffic growth per hyperscale CapEx budgets continue to increase. We're seeing Facebook, obviously, with a massive CapEx spend into next year, architectural evolution, AI, Metaverse, et cetera. How are you thinking about that in the context of could we actually see a situation where the cloud -- opportunity set of the cloud underlying growth drivers are actually accelerating here for Arista as we move forward?

I think the best way to look at cloud is cloud is a -- it was and is a phenomenal growth opportunity. But despite its size, there's still a lot of growth ahead. And I asked one of our cloud customers a couple of years ago, all the enterprises are moving to the cloud. So are we done building the cloud or there's more to build? And their answer was, well, no one really knows. But there's clear growth, at least for the next 10, 20 years and then we'll find out what other workloads can move to the cloud range. So it's a phenomenal market to be in. Cloud growth is often sensationalized by amount of CapEx and how the cloud companies are growing booming with digitization and so on. But you have to normalize these things, right? Before the growth projections for 2022 by meta or the other companies, we have to remember, there was this period of slowdown. So you have to normalize these things. If you normalize, I think they're all on steady growth but it's a volatile sector. So there will be periods of high growth, there will be periods of slowdown or even negative growth. But when you normalize all of it, it's just steady growth. And that's what I would expect as well for our business to do.

Is it hard to really maybe wrap our arms yet around the effect of metaverse and this idea of deepening AI narrative? Is it hard to say that, that would accelerate the bandwidth demands of the cloud traffic in the opportunity for you guys? Is that far-fetched? Or do you think that when you hear things like that, you immediately think like, hey, that could be an accelerant to bandwidth or cloud traffic growth?

I don't think it's an easy correlation. You can say that certainly because their traffic is going to grow 50% that someone's business will grow 50% like us because traffic growth of 50% means they only need to go from 100 gig to 200 gig, not even 400 gig, right? So we have plenty of technology coming to the market to enable all of this with 400 gig and 800 gig in a couple of years. and there's already work going on, active development work going on in the industry on 1.6 terabytes. So I think we just keep up with demand in this industry. The real question is AI, how big of a use case is this? And will this take over their market completely? Or will it simply be an offload and then you're still are relying on CPUs for most of your work? I think that's TBD. No one knows the final balance of this ecosystem. But we are seeing good growth with AI at multiple cloud companies, and they're already starting to find success in their businesses, their value add on why they were to use an AI cluster, and that certainly drives a lot more traffic. And also, drives a lot more demand for uptime in the network that has otherwise been hard for them to achieve.

Right. And I think it was your comments at a recent Analyst Day talking about the cloud titan demand and mentioning that the cloud titans are moving from build or buy to build and buy strategies. Can you help me appreciate what that means?

We went public in 2014. And since then, every meeting, investors ask us about white boxes and is this the threat? Will they displace us? So there's a lot of nasals early on in our journey. But what they don't understand or didn't understand is, in the early days, the cloud companies used to do an analysis of build versus buy. And buy may be more expensive, but maybe you get time to market, build a large to customize. And in most cases, the cloud companies never made a decision to build their own white boxes purely on costs. They did this because they felt there was no one else available to customize to their needs, to add their secret stores, to integrate with their own stack and so on. Today, the cloud companies have more than one choice. And as we've shown with some of our existing large customers, they not only can build their own switches, they also buy from us. In fact, we codevelop to the same spec. We first developed the spec with them, then both companies design switches for the same spec and the switches are interchangeable. We made an announcement at OCP with Facebook, where the next-generation fabric switches, they call it Minipack 2. We call it the 7368, they're essentially the same product on the 7368 and can be interchanged, but one comes from Arista. The advantage the cloud companies get is that when they build on their own only, they're actually locked into a single vendor and even worse, that vendor happens to be inside their own company. So that's a very different sort of ecosystem where you can't put too much pressure to accelerate time to market, what if there's a bug or a respin or a problem you run into your own stack versus buying to the market where it's an industry standard product and much more broadly available. So the cloud companies, think of this as a derisk #1, #2 as a development partner, and we bring a lot of value to the table. One of the new products we announced recently, the one I mentioned at OCP is greater than 20% power reduction compared to the standard design that everyone else in the industry was doing. That's a significant value to these cloud companies. And they love that kind of innovation that comes to market when we codevelop, which otherwise never existed, right? They were working in their silo. We work in our silo but now both the engineering teams are an extension of each other. We work as if they're one company, and we get tremendous value creation out of that. So I think this balance of build and buy is here to stay.

Yes. That's perfect. Very insightful. 400 gig, I would imagine any time you sit down with an investor or maybe even customers for that matter, 400 gig is first and foremost at the front of the discussions. With 400 gig materializing and while I think people seem to understood -- appreciate that 100 gig is going to be around for quite some time. I think the question comes back to Arista's competitive positioning. Do you think that Arista maintains or maybe even expand their market share presence at 400-gig relative to what you saw at 100 gig? Or any thoughts around that market opportunity for Arista?

I think there's a lot of data out there floating in the market and people looking at market share numbers and so on, and there are 2 ways to look at it. We have our existing customer base and growing. And within that customer base, we are very well positioned and continue to maintain and grow our footprint at all speeds 100 gig, 200 gig and 400 gig. There are especially 2 companies in the United States that build their own white boxes and they were actually the first adopters of 400 gig in the United States. So those 2 companies have migrated from their own white box to their own white box. But because they're very, very large companies, they dominate the 400-gig market share today. So if you just look at those statistics, there's all these headline news that white boxes are taking over 400 gig and so on, but the reality of the industry is just meeting status scope with those 2 companies for now. But the rest of the market, as I mentioned, the customers Arista caters to, we're doing very well, and we believe we'll continue to grow our position.

Yes. And I think you've mentioned 300, 400 gig customers to date, I think that's 75 a year ago. How do you think about that inflection of 400 gig as far as revenue contributions and to me, it's like this isn't a '22, it's a '22 to 2024, 2025 story, just a progression of 400 gig in the model and how we or investors should think about that.

Yes. The cloud companies think of architectures, architectural changes roughly every 3 years. The enterprises and some of the service provider customers, financials, think of these as every 5 years or so. But any company that's coming up to a next-gen architecture refresh, they are absolutely considering 400 gig. In many of the enterprise use cases, we're already seeing they might still have 25 or 50 gig down, but they're looking at 400 gig up to the spine. If they have a DCI connection, they're trying to upgrade those to 400 gig as well to remove any choke points. And you can see today's compute and GPU technology is able to push out a lot more IO well. So that journey is already happening. And you're correct, this will keep on happening at least through 2025 before the customers wake up and say, it's time for 800-gig.

Right. Right. And those architectural transitions, I mean, definitely -- and I think I've asked Jayshree this on conference calls in the past, thinking about server CPU cycles, that's kind of the leading edge, right? It's -- I think we get a little bit too tied up in CapEx numbers and stuff like that. But thinking about server cycle as the leading indicator, the Arista business, at least in the cloud guys, is pulled behind that. Is that a fair assessment, like 1 or 2 quarters after we start to see those CPU cycles which realized that we start to see that leaf spine infrastructure, investments start to move after. Is that fair?

It depends on the customers. So some customers do their DCI upgrades or installs independent of compute cycles. Some of them tied to compute cycles. But filling out of the racks inside a data center is always tied to adding more compute to that building.

Yes. You're the COO. So I think I have to ask you about the operational attributes of the supply chain right now, right? So Arista's purchase commitments, as we've been reporting, have been up dramatically, right? You and others over the past year, I think 2.1 -- over $2.1 billion exiting this last quarter. Just give us the current thoughts and the lay of the land of how risk is seeing the supply chain? Any thoughts on when you might see some normalization on that front and just what you're doing to execute through that?

Yes. Aaron, this is tremendous opportunity for us to grow. And it comes down -- a lot of people have asked in the past, what is your most or what's the biggest risk of your business. And it really is in a situation like this, a direct threat from a competitor. It's actually our own execution and we have to execute well. If we miss execute, we miss out on a large customer base. So we don't want to be that company that misses out. We want to capitalize on this growth opportunity. We are in a good product cycle, lots of next-gen technology. And then I know we spent a lot I'm talking about the hardware, but there's a lot of goodness in EOS with EVP and with the VXLAN with segmentation, segment routing for DCI and backbone connectivity. There's a lot of innovation going on in CloudVision and campus as well. So when you put all of that together, we have enough cash on hand. And a lot of people try to put together what can we do with it and so on. But one of the best uses is to buy the inventory, we know we're going to need. And lead times for many of the silicon components are at 52 weeks. Some of the smaller ICs are running at even today at about 72 to 76 week lead time. So if you're confident your business is going to grow, the best thing to do is to make those purchase commitments and get the supply. And in the worst case, we'll have some inventory that will use up over time because this is largely for next-gen products that have a long life ahead of them. In the best case, we'll just use up the components as soon as they show up. So our philosophy is this is the right time to invest in this area and keep on investing. And until lead times truly recover and we can build some basic buffer of components and the right critical resources, we should just stay on to this. And then -- which is why the purchase commitments as sort of an end function of sort of 52 to 76 week lead times. And as I mentioned, we're not trying to be hand to mouth. I think the whole just-in-time focus for the industry is what got many companies in trouble. And this is not the time to sit back. I think this is a time to invest and that -- we'll continue to do that.

Your best view on normalization, if you have one right now.

So my guess will be good for the next 7 days because things will change after that. But if you look at the Asian countries, places like Vietnam and they're still in a severe COVID wave with lots of shutdowns and impacts going on inventory. At some point, we have to get through that whether it's through vaccination or mass immunity, one way or the other, the world will start to recover it. But the Asian supply chain is critical for the whole world to recover in this infrastructure space. Then there's the fab capacity for silicon. The indications we are getting is labor demand, labor shortages and just raw materials, all of that is tied to countries, lockdowns and restrictions. Once they are gone, it will take about 6 months or so. So my guess is maybe late 2022. There's been things have recovered in that space. But again, it's a guess if things don't -- COVID keeps on growing with the new variants and then things get to stuck.

And at the same time, when you're asking whether it be enterprise customers or cloud customers, your visibility in the business has changed. I think in the past, you guys have talked about 1 or 2 quarters of some level of visibility. Now you're talking about not to put words -- a year's worth of visibility. I mean is that something structurally we expect to kind of continue well into it sounds like 2022 at this point? Is that fair?

It's a function of lead time. Most of the big silicon items are at 52 week lead times. So as a result of that, customers have no choice if -- they don't want us to guess for them, they want to plan better then we all look at 52 weeks' worth of demand, maybe even beyond that, sometimes just as a rough direction of the forecast. But once lead times come down, customers are anxious to actually reduce that future planning because it's taking a lot of cycles for them, and it's still guesswork, right, cloud companies are not going to be that perfect at predicting their future 12 months out either. So I think once the industry gets an opportunity, people want to go back to the preCOVID normal, which was 1 to 2 quarters for the cloud, maybe one quarter for the enterprise. On the silicon lead times, by the way, the recovery might be in late or sort of mid-2023 -- early to mid-2023. And as a result of that, people are keeping an eye out for that. I think once you see silicon lead times come down, then commodity items, that's when customers will reduce their forecasting as well.

Yes. I'm going to kind of fire through a couple of other quick questions in a few minutes we've got left. I get the question a lot from investors around Cisco Silicon One. I'd just love to hear -- I know we probably should have touched on this in the cloud in some of the prior discussions, but how are you seeing that competitively in the market, if at all?

So look, customers have looked for an alternative. They love Arista. They love our execution. They love the products. We have very good synergy with their teams. And at the same time, customers would like to have a backup option as well just in case. As a result of that, they've settled on Silicon One in some cases. In some cases, they look for other alternatives. And that's okay with us some -- a little bit of competition is. But yes, the growth opportunity is so tremendous ahead of us. I think the time is so large that we can just continue to grow despite all of this. In the end, the partnership we have with our silicon suppliers is great as well. The Broadcom execution has been stellar, largely, you see today the Silicon One is trying to just copy what Broadcom is doing, but that doesn't let you get ahead, right? We are still healthy with our customers and our products.

Very good. NetDL was announced at the recent Analyst Day, it definitely seems to be an important pivot point for EOS and just the strategy going forward. Can you kind of in a real quick way, touch on the key points or how we should be thinking about the importance of NetDL from evolution of the Arista story here as we look forward?

Yeah, there are lots of metrics our customers track. Do you know what's the most important metric a network engineer cares about? It's called mean time to innocence because if something is not working, they spend time proving that it's not the network. It's not their part of the network. It's something else. And it takes a lot of time. Enterprises -- large enterprises end up spending about 5 to 6 hours to get to that point once an incident has been reported. How do you help the network community get ahead? Compute has done this, applications have done this. So for networking, what's been missing is all of this data in real time at the right place to correlate, to analyze, to get to the right conclusions. And NetDL is the framework that leads us to that. There's an element inside every switch. It's part of EOS. It's an element of streaming, there's an element of a data lake that you end, then put the analytics on. And it becomes a platform to provide a lot more richness to the network than has ever existed before. I don't think I have to say that AI is happening in networking today, but I think it can in the future. But in order to get there, you have to enable all of these step functions and platforms before you can build real artificial intelligence.

Just maybe it's not the right way to ask the question, but when does NetDL proliferate across the product portfolio for Arista? And should I not think of that as an opportunity for Arista to extend opportunities of other adjacent monetization, be it up to software stack and other areas for the company? Or am I thinking about it incorrectly?

No, you are thinking about it correctly, but, NetDL as I mentioned on platform, you're essentially taking out turn in the middle of the night and your headlights can only see that far out. But when you have the right platform, there will be tremendous opportunities in the future to build on top of this. And whether it's an adjacent market or even our existing markets, we'll get there. But first, we have to enable this ecosystem and then you build all of the applications on top, including with some of our ecosystem partners.

Yes. And as far as the effect on the product portfolio or when we think about that having an impact on, be it customer demand or just across the Arista's strategy?

And you'll start seeing some of this mixture already integration with the product and more stickiness with EOS itself, and then we'll build the analytics on top.

Yes. Okay. And the final 3 minutes I've got left, I'd love to ask you about the campus switching opportunity. You've talked about $400 million in 2022 as kind of the new target. Just help us understand simplistically how is campus actually expanding the enterprise opportunity? Like is campus pulling in, the spine opportunities? Is it the reverse? Or how has that evolved for the company as we start to think about this next target you've laid out?

Aaron, in our first year of campus sales, especially with Wi-Fi ANC, which is combined we have seen many new customers just come to Arista. And I've spoken with many CIOs there and their response was asked them, hey, what happened? What changed in your mind that you're now ready to talk to us? You weren't ready to take our call a few years ago. And what happens is that there are small to mid enterprise, then their data center footprint wasn't large enough for us to be a strategic vendor to them. But campus was. And us entering campus opens the door for both parts of the discussion, and now they believe we are truly strategic that they can take us on as a key partnership and grow that relationship. Now today, we're also growing with our existing customers. So it took some time, existing data center customers. So it took some time, and now they're qualifying us in campus and deploying and they're growing with that too. And as you know, these are very, very large financials, very large enterprises. I think that will keep the momentum going on both new logos as well as the existing customer base on campus.

And is there any quantification or any kind of qualitative comments you can say about for those smaller data center footprint customers that brought you in as a campus supplier? Any evidence of if it's 12 months later, what have you, when they go into that next data center upgrade that it's the opportunity to take Arista holistically across campus and data center at that point?

There's too many other variables that don't get captured when you try to run a regression analysis on these sets of numbers. But subjectively, I would say in general customers are very happy. They're happy with the quality of the product. They're happy with the overall execution of the company. They're very happy with the automation they get with CloudVision, that itself opens up the door. And then it's a fact, dependency on when is the next upgrade in the data centers, then you build out, is there a new expansion opportunity. The moving that opens up, we get to play for that.

Okay. Well, I think we're pretty much at the time. Anshul, as always, I appreciate you letting us hear more about the Arista story. Thank you so much.

Thank you so much. Bye.

Thank you. Bye-bye.