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

John McCool, who's the Chief Platform Officer for Arista Networks. I'm Jason Ader from William Blair. And I've known John for probably 15 years from back in his Cisco days. And it's good to see you again.

Great to be here, Jason.

Thanks for being here.

Nice to be here live.

And before we begin, I'm required to inform you that a complete list of research disclosures or potential conflicts of interest is available on our website at williamblair.com. And we're going to just do a fireside chat here, but feel free to ask a question at any point in this session. So -- and then we're going to do a breakout right after here right after this. And remember, we'll get to the name of the place, but it's not going to be [ up until ].

So to start out, John, for investors less familiar with Arista, could you give us a quick overview of the history of the company, what you sell and who your customers are?

Sure. So Arista was really founded with the notion of building more of a software-centric approach to networking and initially focused on the cloud segment. I think looking at where we are now, it's hard to believe the cloud was underserved and very small segment when the company started. And they were using service provider and enterprise oriented solutions and adapting them to their needs and it didn't fit quite well. So Arista, with a software-based approach, really focused on how they can centralize the management that could add agents into the operating system and really scale what had been very small teams to a very large deployment of millions of servers. We then took that approach into the enterprise, starting initially with the financials and really promoting this idea of a cloud-centric networking approach and the data center. From there we expand it into new technology segments from data center into routing and campus and built more of an enterprise end-to-end portfolio, and that's really spurred the growth that we see in the enterprise segment.

Great. And would you say that from a -- just who your customers are standpoint, are you generally focused on the higher end of the market? Is that fair to say?

Yes, we would categorize as more Fortune 2000, so really technology differentiation in the high end of the market, not so much in the mid-market. There are points of mid-market that might be more technology-centric, maybe in media entertainment or the financial pieces. But people that really value the network and generally they're generating revenue from the deployment of their networks.

And then can you talk about your role at Arista. What does Chief Platform Officer mean?

Sure, I have a couple of different responsibilities for new platform development in our hardware side and manufacturing and supply chain. And the last piece I'm skipping, a little busy recently.

Good. Well let's talk about that. You've been in the industry for a couple of decades. Have you seen anything like this with respect to supply chain?

No, I think we've seen kind of supply issues, maybe a tornado or a tsunami or a specific plant, specific to different parts, maybe some capacity issues. But this is extraordinary both in terms of the demand. I think that we have seen either enterprises that maybe had some digital transformation projects that, I think, really recognize more of the importance of those as COVID hit or on the cloud side, really reacting to the scale that was needed to deal with the pandemic. At the same time, capacity did increase. I think semiconductors actually did increase and grow, but not to the extent that demand is not being met.

And can you separate the issues for us, like what are the, I don't know, 2 or 3 or 4 major areas of constraint right now in the market?

Yes. No, it's really kind of transformed over the last 2 years. If we were here a year ago, things were pretty broad-based in terms of supply challenges, everything from metals to resins to semiconductors. You also had a lot of issues with mobility, control orders in different regions, a specific plant or factory make a component, well, may went down for an outage and then had to be quarantine. With the exception of China, I think you're seeing a general opening up of economies and movement of people. So you're seeing less of that. But the semiconductor portion of the supply shortages continue with, I think, a really specific area being challenged around power and analog and logic devices that have very broad-based use, not just in networking or even IT, but can be used in consumer devices, which are broadly distributed.

Got you. So it's power and analog, what about like the network processors? Is that related?

Supply is still tight. If you think about larger components like CPUs and complex systems on ships, still very tight. But I think in this area of power analog logic, pretty acute. And I think an area typically that had been underinvested in terms of capacity. These are devices that are very low -- generally in low cost and very used in everything from power drills to sophisticated networking equipment.

Yes. Got you. As related to Arista, how have you guys been able to navigate the situation? It feels like you've done better than most.

Yes. It's certainly not without its challenges. I think we were fortunate that when we entered into the pandemic, we were in the middle of a refresh of a lot of different products. So at the beginning of a product life cycle, which I think gave us the fortitude to make investments in the supply chain and component purchase commitments around those because we know they have a long life. I think we've also done a good job with some agility around redesign and alternative design substitutes to work around this. So that's really been how we've been navigating it to date. But I don't think we -- we're in the same choppy waters as everybody else.

Yes. Got you. And as you think about some of your competitors that maybe have had bigger issues with supply, what do you think that is?

I really can't speak to what the competition is or isn't doing. I think we were fortunate that our direct manufacturing was not located in China. It's been in other locations. And again, kind of making those bets on some of those new platforms, certainly definitely helped us.

Good. And then has anything materially changed since you guys reported, on the supply chain?

Yes. I think -- we don't give guidance within the quarter. So we kind of had some puts and takes. On the supply chain side, I think all eyes are on what China is doing with the zero COVID and how that opens up and it opened up in June. I understand there's some more questions around that, so we're watching that closely.

Now it's just a question of what you can get on a plane and how many planes?

How many planes you can get out, right, exactly.

Got you. Good. Any questions on supply chain? Otherwise, I'm going to -- yes, go ahead, Mike?

[indiscernible]?

Yes. And yes, and I'd add a third one to that, also substrates and lead frames. And we hear all 3 of those, and it might differ on a different product at a different time. And when we ask that question, yes, its wafer fab capacity, but if that's resolved, it's going to be assembly and test.

[indiscernible]?

I think everything is constrained and it might be a little different on each different product depending on what test -- what specific tester it needs or where it's being assembled or what the wafer is. But it's just tight across all 3 of those elements.

[indiscernible]?

I think because these parts are fairly well distributed, there's sometimes opportunity to buy excess from different places or [ broker bus ], but it's hit or miss.

So it's like whack-a-mole basically?

It's -- yes. So you have to be agile and…

Any other supply chain questions before we segue to other things? Okay. Wanted to talk about the cloud vertical. This is obviously a really important business for you guys, it's seen a nice rebound. What's that been driven by and then how do you help people think about sustainability? And I used to cover Foundry Networks, which is an old competitor in this space. And the thing with Foundry was like they sold a lot to public sector, right? And it was like, okay, you sold the stock in Q3 because that was when things were peaking and then you bought it in Q1 because that's when there was no public sector demand, but you knew it was coming. So is there a kind of a dynamic with Arista where you're sort of peaking right now with this cloud spending cycle and investors maybe be hesitant to buy it because it's like, okay, we're peaking now. And then in 18 months, we're going to…

I'm glad we're not on the government procurement cycle. But I guess, I would say in this segment, there is going to be an element of cyclicality to the cloud, no doubt, as they build out. I think if we look in retrospect, you saw things starting to flatten out in 2019. Our belief now, maybe with hindsight, is they were a little bit underinvested during that timeframe, anticipating the new technology cycle that many people call the 400 gig cycle. We call it the 400, 200, 100 gig cycle because it -- the new technologies can be deployed in many ways. So we think coming into this, there was a little bit of a depressed demand and in anticipation of the new cycle. The technology cycle coincided almost exactly with COVID. So products were coming out, and they were in proof-of-concept in 2020 and early deployments in 2020 and I think we're in that cycle. And I think that COVID kind of forced some decisions along the way, decisions that would have happened had to be made earlier to deal with the supply chain issues with maybe some more forced uncertainty on how they would deploy. And I do think we're in early innings of this cycle. We're also seeing new use cases build out around cloud. Specifically we've seen before build-outs at the edge, but now AI clusters which is new TAM within cloud, where they'll put together a set of machines around Ethernet switching to do AI and analytics.

And historically that would have been an InfiniBand play?

Or not existent.

Or not existent.

So InfiniBand would be a choice for these. And I think that we've figured out how to make these Ethernet-based AI clusters more deterministic and deal with some of the differences in packet and latency that go around the AI use case.

Got you. Okay. So maybe just remind folks on what you've said about your cloud visibility?

Yes. I think, definitely, it's extended from what we typically would say from an architectural view being 6, 9 months out. I think we have better visibility given that viewpoint and where people are in their cycle and what they're deploying.

I think Jayshree said like first half of next year. right, through the first half of next year, visibility, something like that?

Fair enough, yes.

Yes. Whatever she said.

Whatever she said, yes, she's the CEO.

Okay. And then just on the enterprise side, how much of your business is coming from enterprise today? And what are some of the main drivers of your success there?

So in 2021 we wrapped up the year talking about enterprise being equal to our cloud business, which was a first for us and that was a big milestone. In the quarter in Q1, we didn't break out specific numbers, but we talked about the Cloud Titans being #1, Enterprise #2. And then service provider and financials, which is a component of the enterprise being the third vertical. So it's been very strong. I think that we've done a good job with the breadth of the portfolio, picking up new logos and additional share there.

Do you think it's been just your technology is -- it's the value proposition of your technology is just really resonating with enterprise customers?

It's the technology that goes into really what we're selling both to the cloud and the cloud really drove this view that operational simplicity and efficiency was important. And we took that into the enterprise and took many technologies that the cloud deployed maybe on their own, like network management. The cloud folks do their own thing around management, they centralize the data so they can manage the whole cloud centrally and efficiently. We built our own product, CloudVision to bring that experience into the enterprise. And we have some inherent capabilities in the operating system that makes that happen. Building out the portfolio, I think we were seeing as a data center best-of-breed supplier in the enterprise. We added routing and then campus, so we can really put forward a whole solution for an enterprise network to the enterprise. And it also gives our sales teams more opportunities to insert. So not just waiting for the next data center refresh, if there's an opportunity for a network management system or taking an old MPLS backbone and converting it, we have those opportunities as well as the campus refresh, so more at that.

Okay. Yes. I mean you've been building out your channel there for…

Yes.

…more than 5 years.

Yes. And so we've been -- we continue to be very direct led because we have to show the differentiation in the technology. And also it's an experience. So our account teams have done a good job finding an opportunity to get us into account because I can tell you, our quality is better and operational experience is better. But they'll insert something in and then try this out and use this for a small deployment. Please call our technical assurance group, see how that goes and then we're able to land and expand from there.

How big a factor in terms of your competitiveness and your ability to win, especially displace competitors, incumbent players has come from, #1, like having better supply than some of your competitors or #2, the fact that you don't have a software subscription, which some of your competitors do have like on switches. You have to buy it now a mandatory software subscription. Have those 2 things been a factor?

I think on all sides, there could be some opportunistic wins because of supply on a specific product. And I don't think we're emphasizing that in any way. We're looking at the sustainable advantage of the ease of operation and making it just work. And the subscription fits into that. There's a certain fatigue of my subscription are complex, I'm constantly being asked if I value a specific feature. We're not competing on features. We're competing on ease of use. The quality is poor. If I have a defect or a security -- I'm afraid to upgrade because it introduce new bugs, but I know I have security issues and alerts that should force me to upgrade, but I'm going to wait. We do things very differently from a software perspective. So the customer's experience is much better, just running it and operating it and upgrading it.

Yes, you have to have single operating system, right?

It's a single operating system. And one thing that's very different is our developers design the tests for that software. So if you're a developer at Arista, if you came in today, you probably would inherit a test harness for your EVPN code, but if you want to modify it, and add new features, you add features to the test and you submit it and you run it in software like you would do a piece of software, which has typically not been the case in the hardware-centric networking world. You get people to physically test the box and insert test cases. We run about 64,000 tests a day on our software and that's improved the quality. The architecture is different. It's more of a database model that allows you to prevent deadlocks and other issues. So it's a completely different development approach.

Can you talk about sort of the next-generation software that you guys introduced, I think it was a couple of months ago, to your EOS offering?

I'm not sure which specific.

Not NetDB, it's called…

Network Data Lake?

Yes, Network Data Lake.

Yes. So the -- I'm trying to think of a way to simply explain this. The way we built our operating system is it has an internal database inside our switches and routers and all the processes, publish information of that database. And if you're consuming it, you can subscribe to it. That made it easy for us to export all of that data and telemetry out to a network-wide database and we've extended that. We're calling it a Network Data Lake and then open that data infrastructure up to third-party. So you could build network management platforms that subscribe to our network information. You can integrate with Splunk and other type of third parties. And it makes it easy for them to consume network visible information, where previously they might have to go out to box #1 and pull box #2 and box #3 or set up complicated network traffic flows to steer the network information into their devices. So it's a very foundation technology, I think, one, I think that's a little bit invisible to people today, but something that I think we're going to exploit pretty well in the future.

And then we've seen obviously a convergence of networking technology and security technology over the last decade or so. Talk about what you guys were doing in security?

Yes. So we did acquire a company called Awake that's in the segment of security that's focused on network detection of anomalies. And it plays into this concept of this Data Lake, right? The more information you have about what's going on in the network, the more you can run AI and sophisticated analytics. So the ability to pull data in from the network element and then run something on top of it creates the capability to do that. So we're at the beginning of that, and we're definitely emphasizing that security element in our campus piece because I think that's an area where people have a lot of concern, who is accessing my network, who's coming in? The concept that you can have a physical perimeter for your corporate network and secure it and maintain it from just who has badge access is long gone.

I know the campus side has been a focus for the company. You guys have grown really rapidly there. Maybe just talk about what's allowed you to see that growth. And then on the kind of components of the campus, you have switching, you have some software, you talked about, and then you have the wireless product, which is also an acquisition, maybe talk about where you are with the wireless product?

Sure. Yes. So the campus piece was probably a couple of different elements, taking our EOS operating system that runs on our switches and then our routers and then building out the feature set and capability to run power over Ethernet switches. There was some new hardware development there. We developed some fixed switches as well as a modular chassis and probably more importantly than any of that was integrating that into a single management playing with CloudVision and then simplifying the deployment. We also understood, especially as you're going into campus, many people view that piece as both wired and wireless. And without our wireless entry point, you're just locked out or sometimes even having the conversation. So we did an acquisition of a company that had a cloud-based approach that we integrated into our campus portfolio, very strong around cloud MSP kind of deployments. We've been able to execute a very large service provider deployments and have pivoted that technology into the enterprise as part of that offering. And we're doing further stitching of that into our CloudVision. So you have a single management framework for both the POE portion as well as the wireless.

And do you feel like the campus portfolio right now is fully flushed out or there are more things that you could do on the campus side from a portfolio standpoint?

I think the basic pillars are there. There are some more filling out. We introduced a very low-end switch last quarter. So there are some piece parts that I think can be filled. I think there was also a question a year ago whether we could handle -- the question I would get from investors, there are a lot of features in the campus to deploy voice data video. I think we've satisfied kind of the base deployment piece. Now how we differentiate the offering is what I think the security piece is starting to add. And we've been at that with the Awake offering and that's well received. The piece where we're highly differentiated today is that you can run the campus and the routing and the data center all the same way. This is hard to believe, but the approach that's been taken historically in networking is these were all 3 different things. You have a data center network, you have a routing overlay and then you have a campus. Each one had its own operating system, its own management platform. So if you're a person…

Sales people too, different sales people, right?

Different sales people and different general managers, as you know.

Right.

So incentivized often to win share. So this is just easier to consume because there's typically either one Director of networking at a company or one VP of networking and they buy all the same stuff. They don't want to train their people on 3 different things. And we can go in and offer an alternative that maybe today you want data center from Arista, but tomorrow you want campus, you don't have to retrain your people or change your operating procedures.

How do you get past the whole CCIE barrier where you have a lot of enterprises that are sort of both Cisco shops and you have people that have trained and got certified on Cisco engineering and maybe -- they don't want to sort of lose that professional certification? I don't know how -- is that -- maybe talk about how that's maybe changing because it doesn't seem that it's changing?

Sure. That's a great question. I think that certification process has been definitely, I think, our competitor has done a good job certifying people and bringing them along with new use cases. It's expensive to maintain. It's time consuming. And where we've seen a shift is new people entering the market have more of a software kind of approach where they not only want to learn networking, but they're familiar with Linux and compute and playing to that model and making it simpler also helps because you don't have to learn all of these bespoke protocols and FCoE and LISP and all of these things to deploy. So that simplicity element and then orienting it more towards, you're getting out of school, you're new with this career. You already know Linux. You know some of the open source networking protocols playing into that and we have a certification, but it's really around how you get up and running with Arista as opposed to deep learning of some network protocols. So some of its just becoming less relevant.

And a generational turnover?

Yes, absolutely, yes.

Yes. Okay. Great. Any questions out there? I guess we're in the last few minutes here. I'm going to ask you, from your perspective, you've been in the industry a long time, I know you don't talk to investors that often, but talking to some today and you've talked to some over the years since you've been at Arista. What do you think is the most underappreciated aspect of the Arista story?

I think the hard part -- networking is very esoteric and the type of things that we manufacture, people typically don't see. They're in a closet somewhere or in the basement. So it's hard to explain. I think there was a lot of questions, okay, you did data center. Can you do campus, can you do? I think the piece right today as we sit, is the aspect of network visibility. We have a very unique technology stack that can drive visibility into what's happening in the network. The relevance of that is in this Data Lake concept, which would allow better integration with third-parties that provide the whole data center solution. The other one is just around security. So I'm able to get the information from what's on my network and analyze it. That really provides the foundation for some things we can do in the future. So I think there's a lot ahead of us on that side that we're just creating the kernel today that is going to manifest over the next 5 years.

Got you. So it's kind of an architectural…

Highly architectural differentiation from the way our software is built and constructed and tested, yes.

And Awake, what is -- there's a lot of different elements of security out there. What does Awake specifically do?

Really looking at network anomalies and trying to predict, is this normal? Is this not normal? Is this something different in a way that it would be difficult for a human -- things are moving fast and even collecting many streams of 100 gig data effectively and compressing it and looking at what headers is difficult?

So it's essentially behavioral?

Behavioral analysis, yes, and forensics.

Excellent. Well I think we're out of time here. But thanks, everybody, for joining, and thanks, John, for being here?

Jason, always good to talk to you, thank you.

Thanks. And we're going to do the breakout, yes, [ Bill Maher ] yes.

[ Bill Maher ]?

Real time with [ Bill Maher ].