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

I am Meta Marshall. I lead up the networking coverage here at Morgan Stanley. We're pleased here today to have Jayshree Ullal, CEO of Arista. Jayshree is actually going to start off with a couple of minutes of prepared remarks, and then we're going to jump into questions. So I'll let you stand at the podium and then will kind of come over here for questions.

Is this turned on?

Oh, yes. It is.

Okay.

Yes.

Well, so first of all, good morning, everybody, especially for all you brave ones who showed up to public conferences. I know the topic of the day is health. And it's great -- always great to be here at the Morgan Stanley conference. How many of you don't know Arista? That's wonderful. And so when I talk about Arista, usually, it's always about the technology first. We are very much a company built on very disruptive innovative technology. And in the 12 years I've been here, we've always been best-of-breed, and we're always leading the pack. And we are driving the future of merchant silicon, software-driven networking at multiple innovation planes, the data plane, the management plane, the control plane. And today, Arista is no more just a data center company, but I'd like to say we're really a cloud networking company that's furthering various use cases, whether it is the data center, which has been our center of gravity, going to routing and data center interconnect use cases or then moving more and more closer to the users and IoT with our efforts in cognitive campus. So over the last 10 years, our TAM has increased from approximately $10 billion to today well north of $25 billion to $30 billion. We ended last year at about $2.4 billion in revenue with good solid operating margins. Five verticals we typically focus on, continue to be our main thrust. The number one vertical that many of you may have heard is the hyperscalers or cloud titans, tends to be our largest one, very lumpy, a lot of volatility and customer concentration there, but very, very good customers. And really gave birth to Arista as a meaningful company because we defined cloud networking very different from the classic silo networking. The Tier 2 cloud providers are a combination of customers who still rely on the public cloud providers, but also build their own data centers in a significant manner. The enterprise, which is our fastest-growing segment now and really led by a number of multimillion-dollar opportunities we have both in the data center and campus. The Tier 1 and Tier 2 service providers, we've done okay in that, but not as well, and that sector has been struggling, both in general as they migrate to more modern principles, but have to still rely on their legacy operational environments. So these end up being our main verticals. And as part of the enterprise, I should mention the financials has always been one of our strongest verticals and so we track that separately. And so with that, I think, I have -- I could say a lot more and that we could be here an hour, but maybe you'll open it up to questions.

Yes. So why don't you...

Okay. All right. I'll move over. Thank you, Meta.

Great. And so you kind of highlighted some of the -- what makes -- what Arista's special sauces. But why do you think kind of 10 years into your journey, either other networking vendors or even your customers haven't been able to replicate what you've done.

No. I think that's a good question because it's very hard to do it. It's extremely hard. Like if you think of how long we've been doing it, we founded the company in 2004. It took us 5 years to build the operating system, and it really was a distributed computer science problem to build a state-oriented software stack for networking. And nobody had ever done that before, and nobody's still done it. They can do bits and pieces of it, but no one has done the level of published, subscribed, state-oriented in a very consistent fashion. And then you have to build all the Internet features and cloud features on it. So it -- I always like to say, it took us 5 years to build it, and it's still taking us 10 years to perfect it. If you look at the foundation of our software, today, we have had to gut out almost everything we started with and add more and more streaming telemetry capabilities, automation. So in our world, typically, in our industry, hardware changes every 3 years and software has not changed even in 3 decades. And what Arista's done is make software relevant in networking. And I think that's one of the big reasons. And the traffic patterns have changed. It used to be that everything went up and down, north-south. And now it's really a combination of east-west traffic and some north-south.

Got it.

And so you need that new software to do that.

Got it. And so you mentioned your cloud titans contributing upward near 40% of revenue in 2019. What do you still see as their biggest networking problem? Is it just -- is it handling all the east-west traffic? Is it how to manage multiple tiers? Just where is their pain point today?

Yes. No, there are multiple pain points. Maybe I'll go back a little bit in history to explain how we started with them and how we're still continuing with them. So if you look at the initial pain point, some of the early cloud titans had to build their own networks. They couldn't even rely on vendors. When you look at some of the early days of Google and Amazon, they said, "All the traffic is moving east-west. I'm unable to keep up with my I/O, the network piece of my server because the server is spitting out more capability than my network can handle." So they had to go build their own proprietary boxes often with OpenFlow and older technologies, right? So in comes Arista greatly inspired by what they were doing to build something. And the first thing we built was something that could scale their servers and deal with this east-west traffic. And not just build something that was blocking and oversubscribed, but a fully flat, [ N-way ] leaf-spine architecture. So the first thing we did was handle the server to network ratios with racks of scale. So the first problem was racks. Then as we started building these racks and data centers, we found that all of these cloud titans, who are building out not just racks, but their regions, so that the servers and storage was a scaling problem, but now that it had to be expanded from a capacity point of view to the region. So we started with racks, we ended up with regions. And the definition of region also changes. Do you put in routing? How do you deal with the interconnect of the optics? How do you deal with for security and encryption? So even the region isn't one definition. It's multiple things factoring in there. And then today, I think we're also dealing -- besides racks and regions, we're also dealing with refresh. How do they go from 10 gig to 40 gig? 40 gig to 100 gig ended up being one of the fastest transitions they made in record time, which put a risk in the map not only as a leader, but the #1 market share. And now we have to deal with not just the 100-gig transition, but also the advent of the 400 gig coming in. So the use cases are multiplying, and there's 3 problem spaces that we're continually dealing with and the common theme across them is scale and capacity.

So probably the number question you get from investors is just why is white box not a threat? And you just went into what the special sauce is.

Yes. Yes.

Whether it be some of your cloud titan positives, it brings up that question, again. And so maybe just reiterating why you still view -- what Arista has done or why you still think that white box is largely kept to a couple of customers?

No. Absolutely, Meta. First of all, the white box question is one we've been answering ever since the company was formed and -- especially when we went out on our IPO show. So the first thing I want to reiterate is whether it's a white box or a blue box or a no box, Arista's commitment and investment in engineering is fundamentally software, right? So we can run on our platforms, and we can run on third-party white boxes, too. So embracing a box is a very small percentage of the problem. Making the box do something is really the big problem. And most enterprise and even many -- most of the cloud titans expect a turnkey solution. There are very few companies that can do it themselves. You would need thousands of engineers to do it yourself. And you need development engineers. You don't just need IT engineers who keep the network. So when we look at the white box, if you will, we look at it as, "Okay, what packaging does the customer want. Do they want a virtualized package? Do they want a containerized package? Do they want it on their own hardware? Or do they want us to build the hardware." So that's kind of the portfolio or continuum of choices. And Arista's not married to the box. Arista's married to making the box work with our software. And we also have examples where we've worked with our customers' software. You may know that we co-developed with Facebook on the 7368, where we run Arista EOS and FBOS, Facebook OS, concurrently. We've done a lot of work with contributing drivers to Sonic, which is Microsoft's effort. We've done an extreme amount of work with Google on OpenConfig. So it's not just a white box, it's an evolving box, if you will, of hardware and software that the industry is constantly changing, and we have to keep up with that change.

Got it. Part of Arista's value proposition has been being silicon agnostic. Yes, that merchant silicon market has largely consolidated to Broadcom, kind of leading to this silicon diversity argument that either Cisco or Juniper are making. How do you counter kind of that silicon diversity argument versus kind of the proposition you guys have had?

Well, I actually don't counter it. I would love silicon diversity. I wish there were more Broadcoms. So I think the important thing to remember is Arista supports at least 5 families of silicon today -- merchant silicon. In the Broadcom family, there's a couple of them: Jericho and Tomahawk and Trident. We also embraced now Intel, prior Barefoot, P4 programmable silicon. And we have a consistent set of FPGAs to do more layer 4 to 7, our low-latency packet inspection with our MetaMako silicon. So the whole embracing of merchant silicon, we are glad to see that everybody is following us, too. But remember why we started it. We started it because people were building their own proprietary ASICs, proprietary software and developing a closed box, and we have disaggregated that to merchant silicon, software and management. And I think even today, less than half the adoption of networking is on merchant silicon. So we still have a long way to go. And proprietary silicon, ASICs versus merchant silicon, it's a question of what gives you the best power, density and performance, and that's what we embrace. And we've always found that merchant silicon, not just the first time, but through cycles of evolution over multiple years, just keeps getting better and better, everything from transistor geometry to performance and scale, right? When we started our journey, we were building 1 terabit chips. Today, it's 5 to 10 terabits, and I can predict the future of merchant silicon going to 50 to 100 terabits. So we just don't see that with proprietary ASICS.

Got it.

You see interesting points in time, and then not a continuum in family.

Got it. Maybe turning to the Tier 2 cloud vendors. These have been major customers as well. Yes, they're trying to determine whether it makes sense to have their own data centers, whether to move them to public cloud. How have you seen that decision process playing out? And what role does Arista make in helping them with that decision?

Yes. No, I think what's clear is whether you're an enterprise or a Tier 2 cloud or even a Tier 1 or 2 service provider, everybody now is embracing cloud principles. Cloud principles are not just for the titans anymore, right? Of course, the titans have a certain scale that everybody else doesn't. That's the difference. They all want high availability. They all want fast recovery from their failures. They all want agility of their workloads. Automation. Everybody doesn't just talk about it, they want to implement it. They all want analytics. They also want the right programmatic APIs to go into the multi-cloud architectures. So going back to your question on Tier 2, it's -- the Tier 2 cloud provider is not trying to be a public cloud provider. I think it's pretty established today that the top 3 are Amazon, Google and Microsoft, maybe a couple more in China and others. So when you look at the Tier 2, what they're thinking of is, "I have these unique applications, whether it's my Exadata or my large mainframe applications or I have a large enterprise set of customers who want a cloud journey. To give them that experience, how much of the cloud do I need to build in the premise? And how much do I rely on the workload flowing?" So by definition, they're all moving to a hybrid strategy. They're not any more trying to be their own cloud vendor. However, they are building a more specialized cloud. They're building workloads that are unique to their applications and experience rather than a public cloud that's building scale and offering the mid-market every choice of storage and compute. So I think the two can coexist. It's important to understand that we don't look at Tier 2 clouds as those just building hundreds of servers. So these are still mass scale. They're building large numbers of regions and data centers. And I think while they will rely on the public cloud, they will continue to build their data centers as well.

Okay. Got it. Capacity of the data center has grown. Optical is, obviously, becoming more of a consideration not only for timing of upgrade cycles, but just kind of cost and meeting the tighter integration. So how do you see that optical and switching combination like evolving? Whether it be pluggables, onboard optics? And what will Arista's role in that be?

Yes. Our founder, Andy Bechtolsheim, he was even trying to go to the OSC conference, even though the conference may not happen. He's still passionate about this that I couldn't speak nearly as eloquently as he. But the one thing we have all seen, whether it's at 1 gig, 10 gig, 40 gig, 100 gig or 400 gig is you've got to have a flexible optical form factor of pluggables. And the beauty of that architecture is then you can short -- support short distances. You can go to Metro. You can go long distances, long haul. You can build in the right encryption. So Arista has always been a big fan of pluggable optics at any speed. And we'll continue to support that flexible nature because an electrical switch has to be somewhat -- has to have the choices of what they connect to, to make the personalities change. Now having said that, there will be times, especially in the 400 gig, where the cost of optics and the maturity of optics is still happening. So in some cases, it may be pluggables. In some cases, it may be onboard connected to give some advantages of power. Arista is a big fan of standards. The 400ZR and ZR+ is one you'll absolutely see us embrace and offer optical options there and also work with partners, including companies like Acacia or Inphi or Ciena, et cetera. So our view on optics is it's that last mile problem. It's too important to ignore, but it's still complex to think you can do it all yourself. There are lots of laws of physics. So we'll partner in some cases. And for most part, we'll have standard interfaces that we can connect to a variety of options.

And do you think there'll still be a big enough diversity of kind of that merchant market, just given some of the acquisitions we've seen over the past 2 years.

Yes. Acquisitions create other companies. So I think there'll definitely be diversity. But I think the diversity has taken longer on 400 gig than 100 gig. So we're big fans of both QSFP-DD and SFP. OSFP has a beautiful power profile to not just do dense 400 gig, but also be the pathway or road map for 800 gig as well. But I think the important thing is as new -- as old players, consolidating players emerge and even the old players who consolidate are understanding the importance of serving their customers and their vendors in a stand-alone fashion. So I think optics will always be a separate market than IP networking.

Okay. So you're going to buy Acacia products from Cisco.

We're going to buy Acacia product, period. If the customer drives it and if Cisco supports it and we see that in action, we're here to help our customers.

Got it. Maybe turning to enterprise and your growth into the campus market. You noted at your Analyst Day you're really being pulled into this market by customers. And what has surprised you by your entry so far? And what do you think your attach rate to enterprise customers could eventually be?

What has surprised me is how much they value the quality of our software. Of course, we always knew we had a very high quality, robust experience. But I think the value to it that we attached was, "Okay, it will happen slowly because enterprises move slowly." But there's so much fatigue in the customer base, so much frustration and so much dominance of one vendor that I think the acceptance of Arista as the natural and only alternative in many cases. And the rate at which it has happened is literally within the last year or 18 months, right? Now having said that, I think it's not going to happen at the same pace as cloud titans nor will they have the same scale as cloud titans. So we're touching many more customers -- many, many more customers, and we were pleasantly surprised that the adoption of million-dollar customers has doubled in the last 3 years. I think some of that's come from data centers, but also some of that has come from our new entry into the campus. So I expect, particularly with our entry to the campus that the enterprise interest is going to get even greater. Because over there, it's super clear that they need cloud principles and they need a cognitive architecture, but it really has nothing to do with whether they build the data center or not.

Got it. And so part of that entry into the campus market with your acquisition of Mojo, are there any other pieces you feel like you need as you enter into the campus market more fully? Or do you feel like you have the portfolio you need?

Yes. I think we've done a good job over the last 2 years of both laying out the architecture and introducing the products in phases. So the missing component -- we announced our entry into a cognitive architecture with making sure we built the same spine or spline for data centers. And then we came out with Mojo. Mojo was the missing piece. Arista did not have the radio management expertise. But what we saw is that there was a new way to enter and build that campus edge with uniformity for wired and wireless. You didn't have to have a separate architecture for wireless and a separate one for wired and snaking and out. So that same leaf-spine architecture could be emulated in the campus and reduce layers of complexity and topology. So 1 OS, 1 cloud vision, 1 universal spine and different edges. So we don't see that we have gaps, but we do see we have to fill out the portfolio and execute since we're the new kid in the block. And as you know, we have aggressive goals there, and we're executing to them.

Got it. In the enterprise market, that's clearly a market that's moving to kind of a hybrid world of having some on-premise, some on cloud. How will Arista kind of continue to gain share in that market and help customers as they look towards these hybrid cloud solutions?

Well, what's interesting for us is if they go to the cloud, we're there. If you go to the premise, we're there. And I think we do though have an advisory role and a strategic role to play there on what workloads go where. And there are 2 huge aspects to that. One is making sure that our cloud EOS, as we call it, has the right programmatic APIs not just for 1 cloud, but multi-clouds. And -- so Arista has been very committed to that, and we support all the clouds. But equally important is the visibility and security, and this is where some of the work we've done with our recent acquisition of Big Switch will really play. Because connecting the clouds is one thing, visibility, monitoring and security between multi clouds and the premise is where most people have not done enough work. So the classic enterprise customers repackage everything and call it hybrid cloud. The public cloud customers are, obviously, moving workloads. Arista is in a unique position because we've addressed both problems, and we have -- but at the same time, we know customers don't want 3 architectural models. They want 1 software model that can handle 3 use cases. And so we are definitely both in the consultative phase and deployment phase of that.

Got it. And so that brings up Big Switch and maybe kind of starting the conversation of, will that be kind of a stand-alone company? Or will that be something that you'll integrate into kind of a portfolio of hybrid cloud solutions? Just let us know how you kind of think about the extensive part of the portfolio?

Yes. Sure. So Big Switch is a fascinating acquisition because they've been around 10 years. And the first thing that I think it will give us a chance to learn from them on how to monetize software. So there will definitely be stand-alone parts to them, especially the network packet broker market, which, we believe, is over $1 billion in TAM. And Arista had a little bit of it with our data analyzer, which was our in-line switching and -- this in-line monitoring and the switching platform. But with the acquisition of Big Switch, we get a full stack of data analytics monitoring capability. So we're very excited about that. We also believe that the multi-cloud visibility and the topic we were talking about earlier is something that's extremely important that Big Switch is focused on. So the combination of our cloud EOS and then monitoring and visibility will also give us an entree. So these 2 use cases are exactly why we think the Big Switch and Arista marriage makes sense. And like all companies, we integrate them quickly into our portfolio, especially integration with EOS and CloudVision because our customers want 1 software, 1 image. But at the same time, we appreciate the depth and capability of the engineers. So we integrate, but we'll isolate and make sure the engineers can get work done as fast as they did before and we don't slow them down in any fashion.

Got it. You alluded to in kind of your remarks that the service provider market has been kind of slower to develop maybe than you expected. Just -- is it just the nature of slow-moving service providers? Or are there things that you could be doing to kind of speed any transition there?

Well, I think it is a bit of both. I think not to belittle them and call them slow moving, but they do have large-scale networks and customers that they have to serve operationally in an incumbent model. So they got the legacy problem in speeds. And so Arista comes in and says, "Here's a shiny thing." And their first natural reaction is, "Well, how do I operationalize this? I need this, this, this feature. Do you have all of this." A lot of times we do, but a lot of them -- a lot of times we say, "Hey, why don't you take a page from the cloud book and not try to do it the old way and bring in the new segment routing or BGP scale routing or DDoS mitigation features and don't just be stuck in your 10- or 20-year-old architecture." Easier said than done. So what we see with service providers is they like what we're doing. They want to adopt it. The rate of adoption takes time because of legacy, right? And so -- and I think we are being thoughtful about which pieces of legacy we will support them on and which we won't do. And I think that also slows us down.

Yes. Got it. And so we've talked a lot about where all the Arista's strengths are. We've clearly had some surprises in the past couple of quarters, perhaps around...

That makes 2 of us.

Perhaps around the cloud titans, more specifically. Where were you surprised? And where do you think investors just kind of got ahead of themselves around that this is always going to be a study up into the right market.

Yes. No, I don't believe the investors were thinking there should be a steady growth to this market. I think that is the TAM opportunity and execution that Arista's believes in long term. I think the volatility of the cloud titan spend was to our advantage in 2017 and '18, when they just grew faster than we imagined. So we were flat-footed then in a positive way. And then as they got into absorption and building out new regions and thinking about several upgrades, et cetera, the -- they went the other way and one particular titan paused in Q2 last year and another titan decided to postpone their server spend, and therefore, the forecasts changed. So we got positively surprised in '16, '17, '18. We got negatively surprised in '19. But if you step back and look at the average of the 4 years, it's still a beautiful thing. It's grown and gone up to the right. So if you take out the quarters and normalize it, I still believe the cloud titans will continue to spend. Obviously, their rate of spend, as they have built out a lot of the regions and we have crossed the first innings, will be big, but the rate of growth won't be the same as it was before in the 30% to 50% range. But I still think, in the long term, the refresh cycle of these data centers they put in will be important. It just doesn't happen in our court of boundaries. And that's the biggest lesson we learned that our visibility to their forecast is generally very good for 1 quarter, maybe 2, and not much more than that. And it's not just our visibility. It happens to be their visibility, too. So although we have a very intimate relationship with them, when they cough or sneeze, we catch a cold, sometimes pneumonia.

Or coronavirus.

No. Not yet, not quite.

Exactly. So maybe kind of dissecting the 2 of those, I mean, Microsoft is maybe an understood situation of waiting kind of 400ZR in order to make their 400-gig upgrade. Maybe on Facebook, just kind of helping investors understand maybe the pacing of servers versus switches? And just why to not kind of -- to buckle those 2 together necessarily when thinking about Arista?

Well, first of all, it's important to understand that not all our cloud titans have the same degree of mission-critical, customer-facing time pressures, right? We talked about the top 3. And Facebook has a very large-scale cloud titan like network. But if they pushed out a service cycle, it just isn't as much of an urgent situation as it might be for the cloud titans, who have mission-critical customer. So I think this is a case where they consciously decided to wait on their server cycle. And the minute you do that, then the network spend, the storage spend, everything attached to that gets pushed out. I also think they got to a point in their structure where because we were codeveloping a product, they also put a little bit more just-in-time forecasting and discipline into that. So before they were building a lot of inventory kind of like we're doing now because of the coronavirus and as we were consuming their inventory, they started saying, "Wait, we're pushing out our service cycle. We can postpone the network build. And by the way, why do we need to have more inventory." So the combination of the server cycle and their own inventory and forecast management practices, I think, led to a bigger forecast decline. Now both sides understand and the partnership continues to be strong, and I think the period of digestion is over. The understanding on the timing is much more clear. So I think we'll actually reach a more predictable state than the times before.

And maybe just kind of putting a fine point on that, should we think of server spend being 3 to 6 months ahead of networking center?

Yes, that's good. I didn't answer that question. Typically, 6 months, right? Sometimes 3, if they're in rapid deployment. But it service -- the network spend generally lags the server spend by 6 months.

Okay. Got it. And maybe just -- you kind of put a fine point on as far as maybe coupling the forecasting a little bit more, but just how do you think about how your visibility has changed over the last year?

Well, like I said, first of all, we don't have great -- we have good long-term partnerships and understanding of their projects. We don't have visibility beyond 1 or 2 quarters with the cloud titan guys. And like I said, it's not because we don't want to. It's because they also plan that way. And so maybe it just makes this sound really weird, but really true is we have a really solid understanding of what their 1- to 2-year project is. We have a pretty solid understanding of their 1- to 2-quarter forecast. And I'm sure Martin Hull is smiling here because he has to deal with this -- our VP of Products for the cloud has to deal with this on a quarterly basis. But the in between is less clear, between 2 quarters and 2 years.

I realize I meant to ask the coronavirus question upfront. So I'll kind of ask that now.

Is it the obligatory question?

Yes. The obligatory question of the conference. Of just -- you noted really no impact from coronavirus on your earnings call, with your supply chain largely managed. Are you starting to see any tightening of the supply chain or anything that we should know?

Well, I think at our last earnings call, and we haven't updated anything since, it was much more early stages. I think it was Feb 13, Ita? And so if you look at it now, since we've had about another 2, 3 weeks, what we can confidently say is, our contract manufacturers are not in China. So we're very comfortable that we can build product. However, a lot of our inventory and components are built in China, and we're seeing the return back of workers, not quite to 100%, but people are coming back, but we're definitely seeing a shortage in inventory and component supply that we're building into our lead times. Our lead times have extended by 2 to 4 weeks. So we think it's tight, but we think it's okay. And it's something we're monitoring. If it continues to improve and the workers come back and we get the components, Arista is very comfortable and committed to building more inventory on components that often have 16- to 20-week lead times, right? So we do that routinely. So if the situation doesn't degrade, and, in fact, improves, then I think we'll start to improve. If it continues for a long tail, then we're monitoring the situation. We'll have to see.

Got it. I want to open -- any questions from the audience.

I think Meta's has decided. Yes, the mic is coming.

Yes.

Jayshree, can you explain to me what the situation is with Microsoft in the upgrade to 400 gig? What are they waiting for? Is it feature set? Is it availability of specific other components on the server side, for instance? What -- I mean, what's the bottleneck for a 400-gig upgrade?

Yes, that's a very good question. First of all, I wouldn't say they're waiting, waiting. Majority of the market is still 100 gig. And if you look at the total available market of 100 gig this year, it's $4.5 billion, $5 billion. If you look at the total available market of 400 gig this year, it's about $200 million. So that puts it in context for you on -- it's almost 25x in difference, right? Having said that, I think what's happening is that -- maybe I'll go back to why 100 gig happened fast. The velocity of 100 gig happened because the need was there. The data centers were getting built out. The ecosystem of sever partners to support 25-gig or 50-gig mix was there. The optics was there. The entire thing just literally came together in a year. That has been more of a challenge in 400 gig, especially the optics, right? So Arista has 10 400-gig platforms waiting to be ignited and lit, and we've been shipping them from anywhere from 6 months to a year, Martin, is that correct? So I think the challenges are most of these networks for 400 gig are likely to start in the spine. I think the sever layers will remain the way it is. And in the spine, Arista has the features. We have the routing switching features. But you absolutely need the ecosystem of optics. 400-gig ZR and ZR+, best case, from a time-to-market will be second half this year. So we are production networks of any kind will really be late 2020 or more likely 2021. So I think that's the #1 issue. The second issue is, I don't believe anyone will build 400-gig networks as little islands. There's always going to be a combination of 100 gig and 400 gig. And so the mix and match of that in a chassis or in spine and testing for that, testing for scale, making sure that all that works will be important. And Arista is actually well placed and well poised and positioned to do that since they're already sitting there with these platforms and our chassis do support a mix and match with that.

Jayshree, Steve over here. Still failing to go away. Campus, somewhere there, if you're going to shoot for the king, you better shoot to kill. And I'm a little bit...

I'm a vegetarian. I don't know the hunting analogy really well.

I never saw the show. So I'm a little [indiscernible]. I'm a little bit underwhelmed by what you're doing there, in the sense of...

Sorry to say.

Yes, you're going head-to-head with Cisco and a lot of other people have tried that, and it's typically been a graveyard for networking companies. And I look at the offer and I don't really see the same degree of innovation or compelling kind of magic sauce that allowed you to break in the data center. And since I know you're smart, I figure that means I'm missing something. So what am I missing?

No. No. I think you're smart, too. What we have to understand is the data center market favored performance. And in the data center market, we came at it very much from the cloud networking and cloud scale perspective. So performance and scale. And if we had tried to take that to the enterprise customer, you would have probably been underground at that time, too. So we had to find that customer that was disruptive and loved our merchant silicon, our software and needed the scale. And we found all 3 in the cloud titans, right? That's when our business really took off. There isn't a cloud titan in the campus world. There are no campus titans, right? So you're right to be underwhelmed that it will take us longer, and we have to do it one customer at a time, systematically. And if you look at our data center share, it's taken us 10 years to go from 0% to 18%. In that same time, Cisco has gone down from 80% to, pick a number, under 50%, 45% last time I checked. So you got to be patient that, that campus market doesn't have the same performance drivers. So what are the campus drivers? The campus drivers are really ease-of-use, automation, security, segmentation and quality. They really value the EOS. So just to give you the story on this, 3 years ago, we asked our customers, should we get into campus? And they said, "No. Don't. Stay focused on the data center." Last year, they told us, "We're too late, please get in." Because customers really want the quality experience that they had in the data center in the campus. So there isn't a disruption of performance and scale, but there is a disruption of wanting the cognitive, self-learning, wired, wireless, bring me the same architecture, flatten my layers of topology. Now remember, we have to go at this one customer at a time. So you will be underwhelmed if you're expecting us to have a rocket ship in the campus. Our first 4 quarters, we are expecting $100 million, which we think is very respectable to go from $0 to $100 million. And I think the team can double again the next year and double again in 2 years. So I expect this business to be a $500 million business in the next 3 or 4 years. And if we extend longer, it could be $1 billion business. But I think those are realistic, achievable goals, but not intended to topple a legacy player who's been selling campus for 25 years.

Tom?

If you could just elaborate on -- you mentioned the 6-month lag time between server sales into the data center and networking. So given that we saw such strong Intel data center numbers in 4Q and some positive indications from disk drives in 4Q, 1Q on cloud CapEx, should we expect 2Q or calendar 2Q to be when you start to see a recovery there? And then second question, just -- Cisco had -- Cisco launched some new data center products late last year. They had some of your key customers on stage. So just make us feel confident that there's not a Cisco share issue.

Yes. Yes. So I think 6 months, maybe it's longer than 6 months, I think, but to expect like our second half of this year to be tied to the server spend is a reasonable assumption. I don't think there's an immediate correlation to 6 months, but it's around there. In terms of Cisco and their fantastic marketing efforts, look, many of the customers they have put up on stage are many of our customers. And you talk to these customers, they all -- no one of them are single vendor. They're not going to buy all Arista, all Cisco. It's always going to be multi-vendor shop, right? And they're always going to be looking at best-of-breed. So we have never said that any of our cloud titans are exclusively our customers, and we don't expect that to be. But we continue to earn the merit of getting their fair share, and we feel very comfortable to suggest that while other competitors may have design wins, Arista's design wins continue to be strong and we don't believe we're losing share in any fashion.

Great. Well, Jayshree, thank you so much for being here.

Thank you all for being here. And stay healthy, be well.

Thanks, everyone.