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

Hi, welcome, everybody. I'm Meta Marshall, I head up the networking coverage here at Morgan Stanley. We're delighted to have Arista here with us today. I'm going to start with a quick disclosure. For important disclosures, please see Morgan Stanley research disclosure website at morganstanley.com/researchdisclosures. If you have any questions, please reach out to your Morgan Stanley sales representative. So from Arista today, we have Anshul Sadana, COO. We also have Curtis McKee, who heads up kind of IR, who's also assisting us on the call.

But maybe Anshul, just jumping in, Arista, for so many years, has just been a story of cloud and high-end enterprise, whereas -- and the sales force really just needed to get Jayshree and Andy in front of customers to show of EOS to close a deal. As Arista has continued to scale products and customer types, how has the sales force and R&D needs of the organization changed? And how will that continue to evolve as you move and make a bigger push into campus?

Sure. Meta, you're actually right -- correct. We have some great people at the company, and they're our best resources as well. So let's not tell anyone. It's a secret that Andy and Jayshree are actually part of the sales team. But the last decade or so has been phenomenal in winning marquee customers in pretty much every vertical. And that's needed to build a reference base and grow on top of that. And we have a great team, right? We have close to 2,500 employees at the company, and everyone contributes significantly for us to get to where we are today. But as we continue to grow and expand, you have to scale the products, you have to scale the sales force, you have to scale R&D. But certain aspects never change. You want to be the best product in your technology domain, and our platforms, our EOS, our CloudVision continue to do that. We have the best quality, and we do that with openness, right? We're not trying to lock in our customers with, I think, proprietary and so on, and it's still a very easy-to-work-with company. That culture, that focus on technology and quality matters to every customer out there, whether they are high-end enterprise or they are just an average company trying to compete in their own space, and we can help them do that. But when we go broader, you have to simplify some of these things for customers as well, right? You can't sell the most complex technology to an enterprise company with just 4 IT people in their staff. And that simplification is also happening, things like CloudVision, and helping us expand there.

Got it. Okay. Perfect. You've made a number of acquisitions kind of over the last couple of years as you've made this push into campus and routing, Mojo, Big Switch, Awake Security. How do you -- you just talked about simplicity and kind of keeping everything based on EOS. How do you get all of these incorporated while keeping the modularity cleanness of EOS and the value proposition the same?

Sure. It's a great topic for M&A. How do you acquire companies and integrate successfully, right? Because as in any industry, many people have made the mistake and suffered. We're not trying to acquire companies that build what we already built. And that's sort of the first difference. So we actually don't need to port EOS into Big Switch or Awake because these are very different offerings. These are not switches or routers. Mojo runs on the access point, and they have software that runs in the cloud for automating and managing your entire WiFi network all from the cloud. Big Switch is all for monitoring the network. It's not for the actual switching. It's a separate piece of software as well. And Awake is essentially AI technology applied to network security, network detection and response. And all of these are good synergies with Arista core offerings. They integrate very well. And I mentioned CloudVision in the previous note as well. CloudVision becomes one of the key points to integrate into an Arista leaf/spine design that customers might be using in the data center as an example. So I think that is helping quite a bit. But to put it bluntly, we're not acquiring companies that completely replace what we do today. These are adjacencies we're expanding into. So there's no question of having to put EOS there. EOS stays clean as is, that team continues to march on. And these other software offerings start to integrate through CloudVision.

Okay. Got it. And I just want to remind the audience that they can submit questions through the portal, and then they'll show up on this laptop, and I can ask them throughout the session. Given the recentness of the Q4 print, we'll focus on cloud titans just for a bit. You did a good job of managing expectations that Facebook and Microsoft would be a little bit more reserved in 2020, but I think the magnitude of the decline was a surprise to investors. How do you think about just what impact, kind of their upgrade timing impacted those results? And what do you think investors kind of miss about the opportunity with cloud titans maybe after the quarter?

Sure. Well, nothing seen since the earnings call, I would say. So the facts remain the same. The cloud titans' spend is going to be not directly correlated quarter-by-quarter based on their CapEx. And some people try to correlate to -- and so on. I think that's where it gets disconnected. We don't control when our customers spend on the network. But when they spend, I think we participate very well, and our business remains solid over there.

Got it. Is it more challenging, I guess, just figuring out timing of upgrade cycles when you think about cloud titan spending? Or is it more challenging just to understand kind of how much of their CapEx is going towards replacement or new builds in any given year?

Yes. I would say the next-gen cycles with the cloud are hard to predict on schedules, but it's given they are going to get to the next gen. Whether it happens in 1 or 2 quarters or 4 quarters, there are just many, many dependencies on their internal programs because they don't -- never try to make one change only. They try to change a lot of things in the stack at the same time. And if one of them pulls them back on schedule, then everything pushes out. And we've seen this happen a couple of times in the last few years. But the cloud titans do spend heavily on their overall infrastructure. It's hard to truly slice out how much of it go -- is going towards networking versus compute storage, data centers and so on. And then within that, how much is refreshed versus add-ons. Even we don't have that visibility. So it's very, very hard to actually get to that number.

Okay. Got it. Clearly, silicon diversity has been an area of focus for hyperscalers. And that was one of the strong attributes of EOS, it's ability to kind of swap in different silicon quickly. However, Broadcom has and continues to kind of have leadership position within this market. So is there anything Arista can do to foster more silicon diversity? And how does that -- how do you avoid kind of the market going towards having a dominant leader as we move towards silicon photonics?

Sure. Well, there are different parts of the market. There's the cloud then there's enterprise service providers and so on. And they all have varying needs. Largely, enterprise can leverage the technology the cloud is using. And others want to as well because they want to be innovators, too. But when you look at Arista EOS, we have a great foundation. It's a great architecture with centralized state, and then all of the driver layers are completely separated into user space agents. So it's very easy to build a modular operating system for us and have these multiple chips have supported seamlessly to the customer. We support the [ Beresford ] chipset today, which is now Intel, and we support lots of chips from Broadcom. Within Broadcom itself, by the way, there are at least 3 completely different architectures which need to be normalized for the customer, which EOS allows our customers to do. So EOS as a foundation will remain a strong pillar. When it comes to silicon diversity, a lot of these charts are made by customers. And it's one thing to talk about it, it's another thing to deliver on it, and none of our customers want to diversify for the sake of it. They want to diversify if there's a better alternative. And they generally don't find that better alternatives. So I would say Broadcom is in the position they are because of their great execution and others have been unable to catch up. And as long as that remains the case, it's going to be hard for others to compete. But having said that, we are actually more collaborative with our customers on this topic and let the customers drive this discussion rather than us pushing one versus the other.

And I guess the second part of that question, as you head towards silicon photonics or kind of onboard optics, how do you foster more of a merchant ecosystem that maybe has existed on the switching silicon side?

What you'll see is, in our view, a parallel track of a closed ecosystem of switching chip companies doing their own photonics co-package as well. And then more of for standardization effort to try and decouple that. And I think the actual results are maybe several years out. So no one knows what exactly will happen, but these battles are going to take place for the next 3 years or so before we reach any conclusion. Having said that, we don't believe the pluggables market has gone completely either. I think that will also remain strong. So it will be these 3 different choices customers will have. And clearly, the most power-efficient, the most cost-efficient, the highest-yield, highest-quality solutions will win.

Okay. Got it. And just -- it wasn't in the list of questions, but just since it's so topical right now and kind of -- to silicon, is just supply chain tightness that we're seeing right now, just kind of what are you guys seeing? When do you think that kind of the space can be at a healthier kind of supply/demand in terms of silicon?

Yes. So as you know, from our inventory position, we did buffer up components. We started this effort actually last year and have almost doubled the -- more than doubled the inventory we used to have earlier, so roughly the same amount of revenue. You can call us brilliant and very foresighted and smart people or just call us lucky, but I think we are in a better position than many other companies today to handle this shortfall. The supply chain industry is tight. When you look at the ecosystem, the chip companies and their supply chain with substrate vendors, all the way down to TSMC, there is a shortage -- acute shortage, I would say, with almost every technology right now, not just 7- or 5-nanometer, but even 16- and 28-nanometer are very, very tight. It's hard for us to see when these results -- our view is that this would last several quarters. So it's not going to be a quick fix for the industry. And this is not the first time we are seeing semi-cycle shortages and so on. This happens every decade or so. I think we'll have to get through this over the next year or so.

Got it.

I think you can plan better, and our customers will have to plan better around this as well given the shortages.

Got it. We'll give you the foresight award for now versus lucky. Moving on to enterprise, you guys have seen increasing traction, not just with high-performance verticals that you may have traditionally been in like financials, but general enterprise. What is their decision on investing in a new vendor versus transitioning to the cloud? And how do you think you've refined your sales message to kind of have so much success with these customers as of late?

I mean when you look at large enterprises, not just the Fortune 100, Fortune 500, but even beyond that, the next group of customers, they all have similar needs. They want the infrastructure that is resilient. They want security in their infrastructure, and they want it to be automated so that they don't have to keep adding stuff. IT systems are getting more and more complex, and the users are more demanding, right? The bar is extremely high. But if you drop a few packets and your call has some jitter, the user will complain about it. You can't easily get away with that. So people are investing more to stay focused and [Audio Gap] next year of enterprise customers are looking for that simplicity. They love EOS because it's rock-solid. But they may not go into the programmability features of EOS. But they do look at how easy it is to use CloudVision to manage their infrastructure to get that visibility. And as part of the sales effort as well, there's a lot of time spent on that topic. Today, we have customer meetings where more than half the time is actually spent on CloudVision instead of the EOS part of the product. And it's very well received by customers. I would say the market has also opened up a lot more to alternates to the incumbent compared to the past. Customers are more ready for that change now than ever before.

Got it. I mean in -- maybe not for campus, but for data center, just kind of what are you seeing in terms of -- I think with COVID, we saw maybe an acceleration of a move towards cloud. But now there's kind of a slowdown as they realize kind of those hard applications still need to stay on on-premise. Just what are you seeing as far as kind of customers deciding on the kind of on-premise versus cloud argument?

Well, if you have legacy apps that are not cloud-native, that have not been designed or compiled to be cloud-native, I think it's a mistake to move them out to the cloud because it's turns -- sort of be very expensive. To leverage the cloud, you have to use the right APIs, right libraries, containers, micro services, database layers that the cloud offers, and then you can get an advantage of being in the cloud. So cloud is not for every app. Number two, you have to be aware of the costs. And if you have a lot of data, you have a lot of compute, which is true for most large enterprises, it's actually quite expensive to move everything to the cloud. So we have seen large companies try and do that only to realize the bills are too high. And 2, 3 years later, they're busy repatriating the data. So we believe for large enterprises, this ecosystem will remain for many years to come where they have on-prem data centers and workloads as well as in the public cloud. And with offerings like CloudEOS from us, which allows to interconnect these with the same VXLAN technology with encryption, managed by CloudVision, you can actually get the best of both worlds. If you're a startup or a smaller company, it generally makes no sense to build your own data center. You want to be in the cloud. So I think we'd see the market split on large versus mid and small, I think large domains with a lot of on-prem.

Okay. Got it. And you touched on it a little bit there, but just how integral is that security and telemetry to this customer set? And Jayshree was just kind of highlighting your partnership with Palo Alto. Just where is that calculus of build versus buy on kind of security and telemetry?

Sure. Look, security is a very broad domain. So when we talk to our customers, there are aspects of security that apply to the network. We are not trying to build firewalls. We are not competing with Palo Alto. They're a great partner of us. But when it comes to the network, we want to build the world's most secure network and give that to our customers. So in doing that, we partner and collaborate. For example, we do micro segmentation that allows us to collaborate and stir flows through Palo Alto firewalls or others as well. And then with the recent acquisition of Awake, we are now offering network detection and response. And this applies a lot to the enterprise customers that won't have the skill set and the size of team needed to just stay up-to-date with every attack. And the Awake offering has an AI engine built in that takes all of the signals from the network then processes that their own machine language to come back with the predictive analysis of where the high-risk flows are and sometimes quarantine them automatically as well. In addition, when it comes to data center architectures and campus architectures as they're evolving, there's a bigger focus on segmentation rather than allowing all of the flows to mix together and talk to each other. You want to segment them so that IoT devices, especially -- or other nonuser traffic is completely separated out and doesn't have a chance to go breach any sensitive data. And some of the work we're doing there in the MSSG space that Ken Duda, our Co-Founder and CTO, spoke to on the earnings call as well, is very relevant to the enterprises. So we're doing lots of work here to make the network more secure and, in addition to that, partnering with companies like Palo Alto or VMware or Zscaler or others.

Got it. Maybe moving on to campus. You've exceeded your goals on the campus market since launch, even in a challenging environment that we've seen in the last year. How do you position yourself best as customers kind of reenter or reengage kind of on their campus office environments?

Sure. The campus space is a pretty large space, and we don't believe it will go away. I think the work space is certainly changing. Different companies have different approaches to this, and you will end up with more flexible work spaces where you have people coming with hot desks or collaboration areas where different teams come together and meet and so on. So that does impact some of the campus architectures, and we continue to lead that with many of our customers. But I would say the pause with COVID is a temporary pause. Some customers have told us that since there's no one in the office, what's the point of refreshing the campus now? Let's just wait until the workforce are ready to come back. Other customers have come back and said, since there's no one in the office, it's the best time to do the upgrade, and they are continuing with it as well. So I think it's a very mixed market, but it's, in the bigger picture, a 3- to 5-year cycle. This is very much a temporary pause.

Got it. And as you think about what helps spur that -- the continued growth of that business, there's kind of the 2 vectors, which are investing in more sales or investing in kind of a broader portfolio, more better integrations with WiFi, et cetera. Just how do you think about -- is there anything you need to do on the product side? Or is this really just a matter of kind of enhancing the sales force at this point?

We actually did expand our product offering quite a bit in 2020. We introduced WiFi 6 access points across the entire portfolio. The integration of the WiFi with CloudVision got a lot better. We introduced features like proximity tracer within our WiFi offering and the integration of WiFi to campus as the wired switches is a lot tighter already. In addition, we introduced new offerings, including a modular PoE switch, which works very well for large enterprises as well. And it's built on a next-gen architecture. So it's far ahead from an efficiency and a throughput and sort of stability, resilient architecture on the fabric and so on inside the switch, using data center technology applied to campus on the low end. So I think we are already pretty well positioned on the campus portfolio. That doesn't mean we don't have to build new products. I think every 2 or 3 years, you'll keep on seeing a constant evolution, like you would have a natural road map to any product line. But for us, it's not one or the other. I think we'll continue to do both. There's R&D investments constantly that are being made as well as go-to-market. If you look at our sales and marketing efforts, we continue to expand headcount in all of 2020 despite COVID. Yes, we got some benefit of having a lower or almost no travel expenses and so on. But in the end, we did grow sales. And I think that will continue for a couple of years to come, both on channels as well as direct sales. And I think that is a natural way to keep on growing in such a large market, where we're very well positioned with a great product offering and good reception from customers already.

Got it. I mean you guys clearly make a point about kind of being an alternative to a large incumbent in that space. There's also competitors of yours who kind of make noise about their refreshed campus portfolios. And so just what are you seeing in kind of the competitive landscape outside of kind of the traditional incumbents?

In the large enterprise, which is -- which has been one of our primary focus, I think the market has been dominated by one single incumbent for many years. And we run into that pretty much all the time. There are other companies that compete more in the mid-market, I would say. And they may be succeeding there, which is why you're seeing multiple companies say that they're doing well, but we actually don't see these other companies as much. We still largely see the primary dominant incumbent in this market.

Okay. Got it. Another reliable growth area for Arista has been kind of service providers, which kind of in part of your definition would include kind of the Tier 2 cloud guys. When it comes to Tier 2, there's also been this kind of build-versus-buy decision that they've had to make. How do you see that ongoing decision process for these customer sets? And is it maybe any different than you've seen with the large enterprises?

Sure. Well, just the Tier 2 cloud and just these service providers, both did well for us in recovering back to growth in 2020. The Tier 2 cloud specialty providers, many of them have finally found their specialty in how to win against an AWS and an Azure or a Google and are doing well there. And as you've seen in 2020, there's a lot of focus on digital services. So if any of these cloud offerings were in that space, they've all benefited as well. The scale is smaller. So it's harder for them to build everything on their own. Actually, they look for more quick-turn solutions that are very much like the cloud titans, but at a smaller scale, and we're a natural fit for those use cases. Many of them have people who were architects at a previous cloud titan company and are now running the architecture direct to cloud, so they already know us or know the architectures and the programmability to actually leverage the functions very, very well.

Got it.

The service provider market is also doing well. And I would say there's still a lot of work ahead of us, but we have made good progress over the last few years in the feature set, whether it's traffic engineering, whether it's QoS, helping them look at the subscriber traffic and segment it better and so on. So some of the new technologies like EVPN and so the legacy technologies like RSVP are finally starting to come together in this domain for us.

Got it. I mean when it comes to service providers, obviously, they've just had an incredibly taxing year on their networks with kind of all of the video needs that have kind of come upon their networks. And so is -- what are they asking you for maybe differently than they did a year ago? Is it kind of this ability to do more QoS on the subscriber traffic that you were just talking about? Or how is the discussion of the needs of that market changed?

I would say the video that we do -- sort of explosion was more of a bandwidth and capacity discussion in 2020. But what's really driving the architectures for that segment is 5G and NFV. And with 5G, a lot of the classic call functions that used to be in appliance type of boxes is all moving to virtualized services that are horizontal distributed scale architectures that can grow. And the proximity to the radio network is quite important for those types of use cases. So I think whether it's the metro backhaul, whether it's the telco NFV data centers and segmentation, traffic engineering using segment routing, those are sort of the new topics that are very relevant to service providers. And we participate very well in these discussions now. As I mentioned, we still have a few years of work left to actually catch up to all the legacy features.

Got it. Routing has clearly been an opportunity with service providers or that's where you've talked about it most. But as Broadcom chips advance, like are there other opportunities, you think, in the routing market for Arista?

Well, we already, I would say, do participate well in the routing use cases with the cloud titans and even the cloud specialty providers. And in those use cases, the customers are a lot more amenable to using programmability and extending the features on their own, connecting to their own controller for things like traffic engineering rather than waiting for all of the legacy stack. In the service provider market, it's harder because they have to interoperate with many, many third party devices, then they still want that stack so you get there. But there's also an opportunity, still, I would say, underpenetrated in the enterprise routing use cases, where they also have to connect to -- across data centers or to the Internet and so on. And I think that is still something that we can go after in the future.

Okay. Got it. And then maybe just to wrap up the conversation, what are the bottlenecks? What are the challenges that you feel like we're just starting to think about that will kind of determine the road map or would determine kind of how Arista can continue to gain share in the years to come?

Yes. I would say that there are 3 different parts of the road map that we look at. One is the platforms. And within the platforms, you always have a next-gen speed. We are at 400 gig today, which is based on the 8 x 50-gig technology. And in the very near future, you'll get 4 x 100 gig based 400 gig, which then goes on to go to 8 x 100 gig, 800-gig ethernet. And in the future after that, likely, there might be either 8 x 200 gig or some of the waiting, getting to 1.6 terabits. So the next 5, 7, 9 years, I think our -- these type of milestones where you have to keep progressing to keep up with compute storage and artificial intelligence that drive a lot of IO. They have to be part efficient, they have to be high quality and so on. Then you have the core software stack, including a lot of focus on routing scale and resiliency within EOS. And a lot of the road map is driven by customers and by us in that domain to stay ahead. As an example, we have customers that can upgrade thousands of racks worth of infrastructure without causing any outage to their tenants. These are single top-of-rack switches, with a single CPU and a single foreign chip that can be rebooted without dropping a packet, which is sort of a great way to provide resiliency without increasing cost. So there's a lot of work we do in those types of areas as well as routing, failover policies, things like that, that we are wheeling in. There's a third piece, which is automation and visibility. And this is where our work in the area of CloudVision or some of the open APIs, streaming telemetry work we do with our products is coming to bear fruit and is working very, very well already for enterprise customers, but I think there's a long road ahead. There's almost an unlimited amount of work you can do in this domain to solve people problems and organizational issues that you run into as well. I think that ends up leading to a great road map ahead for us and the industry, and we love to lead in this space.

Got it. Perfect. Well, this has been great. Anshul, Curtis, thank you so much for joining us today, and look forward to talking to you guys soon.

Thank you.

Thanks, Meta.