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

Great. Thanks very much. Good afternoon, good morning for people -- for whoever is joining us. Today, I'm hosting John McCool. I've hosted John McCool many, many times in the past in this conference under different hats, Cisco before and now Arista. And John is -- serves as the Chief Platform Officer and SVP of Engineering and Operations in Arista. We have many questions for John. And first of all, I want to welcome you and thank you for participating on our call.

I want to start maybe with, if you don't mind, just to present yourself kind of your areas of responsibilities just briefly so people get to know you if they don't know you already. And then discuss the current environment. I'm going to start with a very general question. The current environment for spending on equipment, if you can touch on kind of how do you see the environment in cloud, the environment in data centers and the environment in campuses. And I'm talking, of course, about COVID-19, implications of COVID-19.

Great, Tal. So first, thanks for the opportunity. It's a pleasure to be here. As Chief Platform Officer at Arista, I have a variety of responsibilities, which include development and design of our hardware platforms, oversight of our supply chain and manufacturing as well as our system test and development initiatives, pretty much all the engineering areas out of the software design. Look, I mean, in terms of the environment, it's just been extraordinary in terms of, first the distortion on the supply chain with the closure in China that we saw in February as well as then the ongoing closures that followed in March, which led to a significant disruption. And at the same time, I think as we're all working from home almost on a global basis, the urgency and importance of the network, its availability, the ability to operate it consistently and remotely have never been more at the forefront of people's minds. So just, I think, a very dramatic change in the way people are viewing the landscape in multiple dimensions. I don't think that we've really understood or sorted out the long-term importance of this period.

Got it. Maybe I'll start with soupe du jour: slowdown in cloud spending. How do you see the environment in cloud evolving? What are the issues we've noticed again -- maybe take a step back, what are the issues we've noticed before and you have noticed before and how the industry is dealing with these items and where we stand today?

Yes. Sure. So I think maybe just overall, in terms of cloud, even pre-COVID, it was phenomenal growth we saw in '17 and '18 in that segment. And we took advantage of that building up our 100-gig market leadership during that transition. As those companies were beginning to absorb that build out, we did see a slowdown in '19 as people were absorbing those assets and continuing to build more complexity of those networks and readying ourselves in that market for an ultimate transition to 400-gig. I think in the near term, certainly, with COVID, it's really placed more emphasis in the immediate time frame of operating those networks and dealing with the capacity increase, more to an eye of stability and certainty in these times where we don't have that much. While people are still continuing to engage in 400-gig, we're still seeing development of the core technologies and optics that are necessary for that transition. People are still continuing to evaluate the technology. So that kind of baseline is continuing to move forward.

Now there are 2 types of cloud companies, and this is all public. We know that Amazon, Google are using white boxes. We know that others, Microsoft, Facebook are not. So the question I have is twofold. Number one, what is your value to corporate, to cloud companies who are using white boxes for switching? And what is your value for companies who are not? And how is that going to evolve?

Yes. That's a great question. So it's important to frame it the way you did. If you look at folks like Amazon and Google, really started to build out their networks well before, I think, the industry had any recognition or appreciation that cloud networking was a different thing than either a service provider or enterprise market. Their needs, I don't think were well served or met. There wasn't a lot of focus on that market, and they had really, in some ways, no choice, but to build out some of their own technologies to address their internal needs, which ended up relying on merchant silicon and then the white box supply chain to deliver the hardware. More recent entrants have taken more of a hybrid approach. And in those accounts, large accounts that have the wherewithal to do their own development, there's both a competition and cooperation with the entire white box strategy. And we've taken advantage of that being very intimate in our engineering engagements with those customers. It goes beyond just a traditional supplier-customer relationship, but really more of a codesign. I think the 7368 that we introduced that runs FBOSS as well as EOS last year is a good example of those kinds of intimate relationships. And then you have this third segment that you pointed out where -- and what's something that Arista is evangelizing is these concepts that have been used in the cloud effectively were to build out networks that scaled well beyond your operational capacity because of automation. And we believe that every enterprise and service provider should and could be able to adopt those principles, but they don't have tons of software engineers to develop a management stack. They do still want to participate at some level depending on their size and integration with other products and technologies. And that's where we've invested heavily in CloudVision to effectively be that point of control for those enterprise networks that don't have the wherewithal to build out their own controllers and management stacks that the cloud providers have developed.

Got it. If I go back to kind of the question I asked you about the cloud and I say, do you see a trend or do you see a risk that -- I'll ask you differently. Is there any cost advantage of using white boxes versus Arista? And if that's the case, do you -- what do you -- how do you address the risk that existing customers, whether it's Microsoft, whether it's Facebook, existing customers will adopt white boxing just because it's going to be cheaper for them. What's your view on the topic and et cetera?

No. There's a perception that this is a commodity hardware, so it should be cheap, et cetera, et cetera. The reality, and I think we've seen this already play out in the enterprise where there was also the belief that things would disaggregate into a separate software stack. There's complexities, a, in bringing in the team and people and overhead that you need to manage that supply chain. People presume that it's a traditional supply chain, that inventory is there in 4 weeks when you place an order. You're dealing with raw component lead times and a purchase order-based commitment. You're dealing with obsolescence. You're dealing with your own RMAs and transactions. So you have to hire a number of people to deal with that, which makes sense at a certain size. But as you move off that size, that becomes a more difficult thing to amortize across fewer boxes. And then there's the software component as well. So what do you run on top of those white boxes, right? This is a raw piece of hardware. And we've also found that the standardization, while people are using the same merchant silicon, how they operate the peripherals, the fans, the power supply, requires integration, and it requires testing. So when you add up all those components, and we've seen customers time and time again take a look at that path, because it is interesting and appealing. And even the folks that are using white box today, I think, appreciate that expense that's maybe not obvious to some at first glance.

We recently published a very in-depth report on white box routing. And I even spoke to executives at your company about it. And the question is, you're playing in routing. You launched products. How do you play -- first of all, what's your view on white box routing with carriers, with the cloud guys? And how do you play in this trend?

Yes. I think the white box routing concept is kind of coming at the heels of what the cloud folks did. And I think that a lot of the carriers have done experimental deployments. There's certainly a play there for NFV and disaggregating some of the functions and more of the stateful functions onto the servers themselves as opposed to a switch. And clearly, we're supporting a number of those designs with our data center technology today. But fundamentally, I think the same challenges are there in terms of management of the supply chain and actually getting everything to work. So if you're on the other side of white box, you are the system integrator, the validator, making sure the software works on top of that stack, and you have to take that on. So it will be interesting to see how that markets develops. And we kind of know what happened on the cloud side of this.

Got it. I want to go back to COVID-19. And just one follow-up question to what you said on COVID-19 and demand situation. So how do you distinguish -- you have a campus switch market, and you have a data center switch market. How do you distinguish between what happens, the impact of COVID-19 to campuses versus data center switches? Meaning, do you see less willingness to deploy in campuses because consumers or employees are not in campuses now versus do you see growth in willingness to deploying data centers because there's more traffic on the network? Or does it work in a different way?

Yes. I think that's a great question. It's really difficult to give a long-term impact to that versus kind of a short-term impact answer to that question. Certainly, in the short term, people are having trouble just getting into their buildings, and campus rollouts take one building, one closet at a time. You can't send 5 people in there and stand up the network. You have to go floor by floor, space by space and upgrade physical assets. So that becomes, I think, problematic for the campus in the short term. Then there's just this aggregate question is how do we view space and offices in the future and does this dramatically change how people approach coming to the office and is it -- it won't be as important in the long-term because people are shifting to work from home? I don't know the answer to that. I don't think people can speculate on that. There certainly will be a role for the office in the future. I actually came to the office for this call today, and I'm excited to be here, maybe more than I've ever been. But on the other side, I think people are really looking at their data center, the data inter-center interconnect and its robustness right now. You can also, in the enterprise, get at those physical assets easier. So you can roll out 5 racks in the data center and stand up an enterprise data center without the same endeavor of a campus rollout. So that's how we view this in the short term. I would say no matter what happens and how long this goes on, the campus networks are still beckoning to be upgraded and modernized just from an operational construct.

Got it. So I want to go back to pre COVID or ignore COVID kind of and discuss the industry. The data shows -- and this is third-party data, the data shows that both campus switching market and data center switching market slowed down in the last few quarters. We've been showing it every single quarter for the last 3 quarters. We've seen it also on Cisco's results. So that's exclusive. That's -- I'm ignoring here issues with timing for you, timing of the customer or timing of certain deployments. This is kind of industry-wide data. And the question is why it is? Why do we see campus slowing? Why do we see data center slowing, especially the surprises on the data center side? If traffic goes up, why -- and there's more deployment, why do we see it slowing? And then the follow-up question is what's going to turn it? What's going to turn it around? I know it's a wide -- it's kind of a wide question, but I want to address both sides of the equation, campus and data centers.

I'll start with the data center. So in the data center, the cloud has been a huge part of that. And as we started the call, we talked a little bit about the dynamics we saw in '19 around the cloud growth and how that was moving off really kind of a hyper-growth in '17 and followed by strong growth in '18. So that's a big component of the data center and that high-speed data center market. In terms of campus and enterprise campus, specifically in large enterprise where we're focused, that's really dominated by the incumbent, probably some speculation but momentum out of the refresh cycle that they were driving as well as the increase in ASPs initially. But it might be more of a flattening out as that goes forward. Our value proposition really around the enterprise has been more of the lack of distinguishing between campus and data center and more of a premise that this is how you should run your enterprise network from your hybrid cloud environments where you may have virtual machines, other than Amazon or a Google or an Oracle or running your data center interconnect or connecting your users to the campus. So that's kind of how we put forward our value proposition, not segmenting these 2, but really trying to show how you can unify and simplify your operations by running one network.

So let's focus on data centers, and then I'll focus on campus. So data centers went through a big 1-gig cycle. If I look at the data by port size, clearly, the growth is driven by 100-gig. But also the growth-is decelerating by 100-gig. Do you think there is some kind of fatigue in the market just because we've been deploying 100-gig for multiple years and now the market is waiting for 400-gig? Is that -- could that be the reason why data center is slowing?

It's hard to speculate. Certainly, there's enough product and things to evaluate in the 400-gig world, and we do see that activity taking place. Whether that has impacted the market, it's hard to speculate. We don't think so. We continue to see a good business in the cloud. It's not -- it's still very substantial, and 100-gig continues to roll out. Just to give you maybe some sizing here. Q4 -- 400-gig market was happening but it was something like 5,000 ports compared to several million in 100-gig. So we're still many, many orders of magnitude different. But there's no doubt that when the economics get into place here, especially around the optical components, it will be more cost-effective to operate at 400-gig. And then we'll start to see that transition kick in, which could drive another cycle of growth and replacement of some of those 100-gig switches.

And do you bring -- on the 100-gig side, do you bring -- or in data center switching, in general, do you bring the same differentiation that you brought before to the market? Meaning, customer that has to take a decision today, do they still see the same advantage of using Arista that they've seen before versus competition? Or did others catch up, opened up, changed strategy to a point where now it's more about price than differentiation, technology differentiation?

Yes. We still fundamentally believe we have extraordinarily strong differentiation, and I'll go through a couple of points there. There's certainly more mimicking of our marketing lingo around cloud and Open and EVPN and then being able to put agents on your boxes. But the reality of it still has a long way to go. In terms of EOS, it still is the gold standard in terms of quality and scalability. The published subscribed model that Ken and his team have developed around the state streaming is extraordinarily fundamental. Maybe what's less known is the level of automation that we've invested in, in software testing. We were able to put out a software release, incredibly high-quality standards, without even having complete what the traditional physical level system testing that my team does. Obviously, we find issues there because we invest in it, but we're not relying on it to the extent that legacy vendors will be. And I don't think that's something that you can fundamentally tape up on an architecture or a system. It has to be built in there from day 1. I think that we do a really great job at differentiating ourselves in terms of the customer engagement around requirements. This is engineer-to-engineer discussions. There's no intermediaries rating product requirement documents that are passed down to engineering. This is fundamentally, in many of these efforts, a codesign and a level of technical data sharing on test results and things that they haven't seen in the past in my world. And then surprisingly, even supply chain is a differentiation, having a different type of supply chain than white box being able to go to multiple contract manufacturers or cater that supply or risk level, especially in these environments. And then having sort of a uniform test standard no matter where things are built. Knowing how products are tested, how they're built is a piece of this kind of pyramid. And then last is merchant silicon. So we -- they won this company, invested in merchant silicon. Our customers, we talked about white box, one of the value propositions is same underlying silicon in an Arista box or a white box gives them agility in terms of managing risk, managing a dual source supply strategy. And also, those -- we believe those silicon providers are more intimate with us in the details and access to their technologies because we're not fundamentally competing with them at the silicon level. So that's really, I would say, kind of the elements of the core differentiation we bring to cloud.

Is 400-gig going to be as significant as the 100-gig cycle? Meaning 100-gig is deployed at the server level, top of rack level and the roll level aggregation, I mean, all levels. Is 400-gig addressing kind of the same market size? Or is it much smaller just because it's a different technology, bigger throughput, it addresses only portions of the 100-gig market today?

Yes. I think the way maybe to reframe that a little bit, if I could, is will the speed of that transition -- I think ultimately, those speeds will propagate through all the tiers of the network. We definitely see the most interesting tier being the data center interconnect, especially what's going to happen around optics and ZR optics to interconnect data centers at regional or even across country level, it's just phenomenal. So we think that, that will be one of the first places this begins to take off. It's also where the speed per fiber connection that you have is very important and has a lot of value. And then the technology will begin to trickle down. 400-gig ports can also be broken into 100-gig ports with breakout cables. So even though something is 400 gig, it might be operating as multiple 100-gig ports or broken down in that manner. So it will take some time. 100-gig is very well established. So the transition may take longer in all those tiers than we saw with 100 gig. And I still think there's some work to do on the optics economics to make that happen.

Got it. I want to switch to campus, and I want to ask -- start with a general question. You've touched on it, and you touched on what you're bringing, what kind of innovation or differentiation you're bringing to enterprises. So the first question is what drives an enterprise to migrate from their legacy vendor that has been there for many years and there is no employee growth? I mean the employee base is growing in line with GDP, let's say. What drives them to migrate to a new architecture or a new product and disrupt their network and do it with Arista?

What we're finding in the enterprise has been a lot of frustration with quality. The networks are extremely brittle and, probably, people don't realize the implication of that is people don't touch them. So we were just talking the other day about a customer we came across in the last kind of decade. They did one major upgrade to their enterprise network. That's not good, right? Because you have increasing number of security alerts that you have to respond to. And people should be upgrading their networks, maybe not as much as we upgrade our phones, but on a relatively frequent basis, to take advantage of bug fixes, security patches and to keep current. And that's just not happening in the enterprise and even more so in the campus. And people realize that, and they always are weighing this balance between the risk of exposure and the risk of the upgrade. And when they decide to upgrade, going through a very contorted mechanisms of qualification, requesting certain fixes on old branches, it's really nothing that you see in the server market or the storage market in terms of its complexity. We go in and taking kind of this approach that we've seen the cloud providers do is have a centralized management scheme that can push out configurations, control configurations, monitor the network from a single point, whether it's the campus or the enterprise or your data center interconnect router. And that is having a ton of appeal and sometimes even skepticism because how could this be? We've been operating these networks for 20 years, and they'll enter into it with their toe in the water, maybe with us in a certain use case until they experience it and believe it. So that's fundamentally how we see people flipping their networks to Arista in the enterprise.

Got it. Then do you -- early success that you have, is it with existing customers? Or does it open up also opportunities with new customers?

Yes. I think on one of the calls, we did mention that we were actually surprised that we saw about half the customers were new to Arista. Now these could have been people who were following us and maybe the timing of their data center build out wasn't such that we fit. Maybe we had to get to a certain size of an organization before we earned their trust. So it's a mix. It's a mix of both. We've seen deals that we've actually won through Mojo that then become an Arista customer.

So I -- my next question, you just mentioned your enterprise WiFi acquisition, Mojo. And my question is exactly there. How does it add to your portfolio? Meaning, why do we see -- and it's a general question, but I want it to be specific to Arista. Why do we see most of the switching companies going big on relatively small acquisitions of WiFi solutions? Why is that strategically important for a switching company to own this asset? And what's the experience you've had with Mojo?

Our calculus when we looked at Mojo was we knew we were looking to enter the campus access market organically but did see that customers had put, even in the past, value around having an integrated approach or taking even just a supplier seriously on an RFP if they could serve both the wireless and the wired component of their networks. So it really was driven by customer preference on how they look at suppliers for the campus and the credibility that took place with that. And it has brought us into some new areas. It's also -- for us, part of Mojo was around the management plane. So we strongly believe that the controller model in WiFi serves people well in the 90s and 00s but that the cloud-based model was going to be more effective. It was the highest growth segment and very analogous to what we had done with CloudVision and centralizing the management of multiple assets. Mojo brought this aspect of the help desk, right? If you think about Arista in the data center servicing large customers that have the wherewithal to build their own networks, even design their own networks, the campus market is kind of how the person takes the first call when your laptop doesn't work and figure out whether it's the network or the laptop. And Mojo brought a lot of that DNA into Arista that's been very helpful.

Got it. So in general, to conclude the campus, what are the hurdles that you need to work on in order to drive up adoption of your campus environment? What's the focus of the company in order -- you have a product. You have a pretty good portfolio. It brings value. What drives greater adoption?

Yes. So I think for investors, I'd say a couple of things. Think of the campus as part and component to the growth of our enterprise segment, not as some separate go-to-market initiative that we have a separate campus sales force. The value we bring to those enterprise customers is to be a true technology superior alternate than the incumbent. That's something I don't think anyone has tried to do before. Folks have gone on the enterprise either on a point basis or with a good enough kind of value proposition at the edge. We're saying this is a fundamentally better way to run your networks. We've demonstrated and have references around people who have used CloudVision and transition from the incumbent and can attest to the simplicity of those environments. We have to continue to evangelize that. I touched on it earlier that when you've operated in a certain model and mind frame for such a long period of time, it's hard to imagine that things are better. So we can get -- when we get a customer and engage in a customer in a trial, a POC, maybe a certain use case, like visibility or their management network, they start to see, oh, this is really different, once they experience it. That's probably the thing that we have to overcome is just that inertia in that market to really believe that you can do something better.

Got it. Question from the audience is about routing. What kind of innovation do you bring to the routing market? And why haven't we seen your routing portfolio growing more meaningfully in the last few quarters?

Yes. So we've been very disruptive in positioning merchant silicon built around kind of serving the switching market into the routing market, and that leads to a very disruptive economic factor. And I think we've been candidly a big factor in that transition, and we've seen traditional routing vendors reply to that. As those silicon kind of can cover a couple of Internet route tables, it is well poised to replace routers in the traditional environment. And that has served us well in the cloud market, in data center interconnect where they've been more willing to adopt new design methodologies. It's also done extremely well for us in the enterprise. But the number of routers in the enterprise compared to service provider are much smaller. So where we believe we need to do a better job is really the penetration of routing in the service provider environment. We had some initial design wins in that segment that were more revolutionary than evolutionary where they could exploit some of the cloud networking technologies and less around legacy. But as we go into more traditional service providers, that legacy component as part of the transition and software functionality became more important. We have not given up there. I think that's an important message. We've continued to develop a significant amount of software and functionality, especially in the release that we just came out with this quarter that can address new use cases and that interplay between new technology design environments and interconnecting with legacy that I think we'll be able to do a better job.

Got it. And John, unfortunately, we ran out of time, but we did cover most of the topics I wanted to cover. So I want to thank you sincerely for joining us today and driving all the way to the office and wearing a suit or a T-shirt instead of wearing your pajama. And we'll take it from here. Thank you so much.

Thanks, Tal. Always a pleasure to talk to you. Bye-bye.

Thank you. Bye-bye.