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

Good morning. The contents of this call and the views expressed on it are solely for the use of Cowen and Company clients. Any publication, distribution, reproduction, posting, sharing or transmission of this information without the expressed permission of Cowen is expressly prohibited. The replay is not intended for representatives or of any media outlet. And as a reminder, we are not interested in receiving and you should not disclose any confidential information. In the event that you inadvertently disclose such information, please notify us as soon as possible. You may begin with the conference call.

Thank you. Thank you for joining us this afternoon or evening for those of you who brought for the inaugural addition of our networking Leaders, Legends and Visionaries Fireside Chat Series. First off, I hope all of you and your loved ones are safe and sound as we wait for a return to the 21st century. I'm excited to have with us today, Andy Bechtolsheim, and Jayshree Ullal. Andy is Co-Founder, Chairman and Chief Development Officer; and Jayshree is CEO of Arista. This is the first over 20 networking fireside chats that we have scheduled over the course of the next several months, and it is a great way to start the series. Andy and Jayshree checked all 4 boxes in the title to series, Leaders, Legends, Luminaries and Visionaries, and I do mean that. Their track records will replete with virtually uninterrupted success, long creating a risk. And, obviously, what they've accomplished at Arista speaks for itself. Regarding the structure of this call, I will chat with Andy for the first half, focus on technology and product trends, and then Jayshree in the second half focus on market trends and opportunities. And I do plan to save a good 10 to 20 minutes at the end of the call for questions that any of you may have.

So Andy, let's start off with a very big picture question. Dating back to the rise of SDN and [ NSP ] back in 2012, many investors have believed their networking as a sector has been undergoing commoditization. Of course, one could point to risk itself is undercutting that view. You deliver the speed of doing, ease of provisioning promised by ASEAN, but you're doing a traditional form factor. Question is why hasn't and why won't networking become commoditized? Or will it? Or perhaps what areas of networking has and are most susceptible to commoditization and what areas aren't?

Before I answer that, I just want to make sure I get our safe harbor statement in here, which is anything we talk about here may contain forward-looking statements that are based on our best beliefs and assumptions, et cetera, et cetera. You can find the full safe harbor statement in our earnings call slides that we published last week. So everything is covered by this kind of information. But going -- back to your question, what enables our very rapid growth over the last 10 years, if you recall, in the data center, 10 years ago, Cisco had around 80% market share. And today, down to 40s and change, whereas we went from 0 essentially to roughly 20%. And in the case of 100-gig Ethernet, we have shipped actually more 100-gig Ethernet ports than Cisco for the last 3 years. So we have been able to deliver a value proposition to customers that included a combination of the hardware cost performance based on merchant silicon and the combination of our EOS operating software, which provided and still provides the most resilient and high-quality network operating system environment in the industry as far as we can tell. Now so that's sort of explains -- that's a simple explanation of our success. There was no prior company that took on Cisco successfully in data switching. Now in the meanwhile, in the last 10 years, there was any number of attempts by various starter companies, typically, let's call them, the disaggregated network companies, such as Cumulus, Big Switch, Pluribus, Plexxi and so on. That, in total, raised more than $360 million of venture capital over the last many years that also wanted to go after this opportunity, that if you look at the combined revenue that these companies actually achieved in the last year, this -- and these are private companies, but based on our best information, it was around $60 million or 0.5% market share of the data center TAM. And 2 out of these 3 companies, Cumulus and Big Switch, have been acquired or in the process of getting acquired. So I think it's fair to say that the plans that these companies had just did not work out and they didn't actually succeed in either the cloud or in enterprise anywhere else for that matter because they didn't deliver a value proposition that solved the customers' problem. So -- and I'm sure there's yet more start-ups going from that trying to do the same thing. But history has not been kind to start-ups with, call it, limited resources trying to solve this problem. Our engineering team is now roughly 1,000 people. The vast majority, like over 90% are in software. The EOS operating system is now over 10 million lines of code, not counting open-source components. It takes an incredible amount of effort to deliver that kind of software environment in a box fashion to customers while keeping up with new product releases. Last year, we shipped over 35 new platforms with boxes, line cards, based on live different chip architectures. So we have been really, really busy doing this for the last many years now, and it takes a significant investment. We are spending about $5 million R&D per year to keep up with this. So in so many words, this is beyond start-ups in my opinion. This is far away [indiscernible] decision.

I was actually going to come back to this question, 2 questions down. But since you've raised it. So as you're pointing out, Cumulus, with the announcement last week, the acquisition by NVIDIA, and your own acquisition, Big Switch -- and I think, if I'm not mistaken, that Big Switch had stopped focusing on a stand-alone OS 2 or 3 years back. And really I'd argue, those, along with [ TTH ] disappearance, they really undercut the argument that SDN is going to have widespread adoption. But at the same time, I think the investment community continues to worry about Facebook's FBOSS and Microsoft's SONiC. And I'm hoping you could help us understand where those operating systems and the value that EOS delivers beyond those operating systems? Because I think there is this view out there that those operating systems are now sufficiently robust that EOS and other third-party merchant vendor operating systems don't offer enough in the way of value, in the way of differentiation. And so it displaces you. Can you help us understand in laymen's terms what EOS delivers that you can't get in SONiC or FBOSS or other open-source operating systems?

Great. So if you look at the actuals that have done their own software on their own white boxes, there are obviously places in their networks where they deploy that software on white boxes. But these networks have many, many layers. And the typical deployment of these homegrown operating systems is at the top of rack. Maybe it's a leaf layer. The more you get into the core of the network and in the higher levels of the network, the harder the problem is to deploy and to deliver high quality, resilient type of software. And again, it's a tremendous effort to keep up with both the new silicon that's coming out every 18 months. And with the installed base and new feature requirements and so on, I mean we are working really, really hard to do all this, and it's not a small effort. So again, if a cloud company wants to spend hundreds of million dollars in R&D to fund this kind of effort internally, and then some of them are, in fact, doing that, they can, but it's not something you take on lightly. And furthermore, it's not something where you really want to help other vendors or your competitors essentially. So you're really trying to just get your own network cable into volume -- into a high-quality setting.

And Andy, you and Jayshree also made a point historically that you work very closely with Facebook and Microsoft. It's not just simply you as an independent third party, but you're working very closely with those customers. I assume you have good insight to what they can and cannot do in terms of their own operating systems and where they're going. And this call is obviously focused on the longer term. Are you concerned, as you look at '21 and beyond, that they've spent many years, a lot of money, a lot of effort. Do they close the gaps sufficiently? Or do you think that gap continues to persist as you continue to drive EOS and your various platforms forward?

Well, we are very almost intimately engaged with all of these large customers. And we know what they're doing and we know what they're not doing. And we are focused on complementing their efforts, right? So yes, it is true that they're developing white boxes at certain levels of their network, but building your own routing-capable platform is extremely challenging. And it took us, I don't know, hundreds of millions of dollars in R&D to get these products. So it is -- let me say this differently. The -- any company has limited resources, even the largest cloud companies. So it kind of comes down to where do you want to spend your effort? And what problem are you trying to solve? Now at the beginning of time, when Arista started, our best friends at Google told us they wanted to build switches or have switches that were much bigger than what was available in the market at the time. And that's the reason they went off to build their own switching hardware. And our inspiration from their conversation was, well, there must be other customers that have similar problems. So we set out from the beginning to build much more scalable platforms than the traditional enterprise data center switches were designed for. And that resonated extremely well at Microsoft and Facebook and others. And this kind of explains -- our focus on the customer base explains our present market share there. And again, people have -- it sounds -- writing 10 million lines of codes that work well, it's a really, really tough undertaking. And you have to maintain it, and you have to evolve it, and you have to test it on every new switch and every new release. A company who's making revenue in the cloud with cloud services and other applications, normally, you would assign your best engineers to the revenue generation part of the business and not to a sort of theoretical cost reduction, where if you really do the math, the savings are not enough to justify the R&D spend you would have to put into the product to make it actually capable of doing what it's supposed to do. So it's a trade-off between how much money do you want to spend on your own R&D or internal cost reduction versus do you assign your best resources to essentially revenue generation?

So I just think you've answered this question, but I trust that the answer's going to be more of the same. But given that while you're the first to recognize the opportunity, you've now spent a lot of man-years, a lot of money, a lot of effort in EOS and your platforms, but, obviously, this market opportunity in terms of shifting workloads to the cloud, it's no longer unknown. Everybody's recognized it. How do you maintain or increase differentiation? Again, is it just more of the same? How do you drive cloud revenue growth now that you've got everybody targeting the opportunity as well as given your significant success, which has created significant market share? Are there key technologies or other factors that will differentiate you in one vendor from another? Or again, is it just a function of you've got so much invested and you just keep driving it forward incrementally?

It's not just the past investment, it's the time to market for new features. What really differentiates us is that we are -- we have been consistently the first to market with every new chip, every new chip from Broadcom and others. It's usually first from Arista with an operating system that actually works. And this explains by our unnatural high market share in 100 gigabit because we got these products out with a fully functioning -- fully -- high-quality, fully qualified software stack on day 1. What happens with traditional competitors is that you don't want to buy their first box because of the -- you get another branch of the operating system and unknown box and other issues, and you want to like wait 6 months or 9 months to let somebody else find the box. And then you start deploying it, but the chip only has a lifetime of 18 months, and then the next chip comes out, right? So the whole innovation cycle around merchant silicon requires you to be there on day 1 to take advantage of the life cycle of the product. And the customer wants the latest chips because they're always better than the previous product. So we are -- we have been shipping 400-gig before anyone else. We have been getting up to Tomahawk 3 before anyone else, et cetera, et cetera. So this is the underlying, call it, feature velocity and time to market velocity on this new silicon wave that is an ongoing -- some piece of -- based on Moore's Law and all the things that Broadcom is doing has been our success formula, quite frankly. And it just has been really, really hard for other competitors to get close to that or even for internal teams to emulate that. It takes an enormous amount of automatic testing infrastructure and processes around this problem of how do you get this new product out as quickly as possible with a fully qualified software release.

Okay. Let's talk about networking architecture technology. One point of view networking architecture appears to be a crossroads at the rise, obviously, of the cloud titan, traditional enterprise networks and private data centers and now hybrid multi-cloud environments. Of course, there's also traditional service providers. How do you see networking systems equipment suppliers successfully and profitably navigating these very different end markets with uniquely different requirements? Or maybe the requirements aren't as unique as we think. What, if any, is the opportunity to converge these discrete and different architectures?

Yes. So one way to think about this is there is a lot of brownfield out there and then there's greenfield. So in the greenfield case, let's just talk about like the routing here in a cloud data center, where we have been -- successfully been able to collapse the spine tier with the routing tier. And those greenfield connection goes to other cloud data centers of the same company and across the wider network all around the globe. That has worked really, really well for us, and there's a lot of routing bids going over. Now when it gets to the brownfield connections, connecting to legacy edge aggregation networks, depending on the protocols that people have deployed here, we may not have had all the protocols. And for that reason, the cloud company may have had to require a different kind of router that had those features. But what's happening there is that these legacy routers are so expensive on a cost per port basis or cost per bandwidth basis, they're just longer cost-effective. So we're seeing certain conversions there where end customers are giving up on complex features like RSVP. And by doing that, they can greatly reduce the cost or improve the bandwidth or reduce the latency to connect on these networks, and then we can accommodate those connections. So as the routing world gets simplified to more less complex protocols, it's a great opportunity for us. But it is true that if customers want to hang on to some legacy protocols, and this is particularly true in service providers that they had for the last 20 years, and we don't have it, we can't plug a switch in there.

Well, that's a good segue to the opportunity presented by the collapse of Layers 1 through 3, especially with the emergence of 400ZR. And I've been following this sector now for 25 years. I think we're going on about 20 years of discussion about Layers 1, 2, 3 being collapsed, and it really hasn't happened much, if at all, a little bit. But 400ZR clearly continues the game. And I know you have been very excited about this opportunity over the past year or so. It's still not quite here yet. Can you give us some insight as to how this market develops the opportunity for you over the course of '21 and beyond, and how it interplays with the dedicated optical systems vendors? I see the other day, you announced an interoperability agreement with Ciena. I think a lot of the investment community thinks that this is going to kill the dedicated optical systems players you may yet. But any thoughts you can share in terms of the revenue opportunity for you and how this plays out over time?

Yes. So first, on the time line, 400-gig ZR modules are in the lab. They're getting qualified. Some of them work quite well, but they're still not in production yet. We are telling customers that they can do field trials by the -- later this calendar year, like in the second half, maybe Q4. And there will be production in next calendar year, that volume production. So there's a number of use cases for 400-gig ZR, which is the first multi-vendor DWDM optic standard in history, meaning there's 5 different DSPs and 10 module vendors all competing on a cost performance kind of basis. So the cost performance of 400-gig ZR is, I would say, an order of magnitude better than the legacy optical systems. But having said that, there -- but if you look at the underlying use cases, and the one we are most excited about is, of course, the data center interconnect, primarily in a metro area, which is projected to drive the majority of the volume for 400-gig ZR. So this is essentially a greenfield opportunity, meaning the customers today like a Microsoft maybe be deploying COLORZ Inphi module, which is a 100-gig kind of technology, but they are itching to upgrade to 400-gig, which gives them 4x or 6x the bandwidth per fiber and much bigger pipes, et cetera, right? So the existing 100-gig DCI installed base will rapidly upgrade over the next, say, 2 years to 400-gig ZR. And the original OIS standard, which Microsoft was a big factor in pushing that is focused on this 80- to 120-kilometer reach, the vendors of the optics, by the time they tape out their 7-nanometer chip, which is a very expensive tapeout, they've put and enhanced toward air correction and other modes that enable them to go to 300-, 600-kilometer reach, maybe even a 1,000-kilometer reach depending on fiber quality. So this will also address other markets in cable and service provider infrastructure. And then there's other modes in the chip, 200-gig, [ 8 com ], et cetera, where they can go even further. So the point is that there's brownfield and greenfield. In the brownfield world, meaning the customer really has existing optical line systems and optical transponders and optical chassis, it may be -- and it will be much easier for that customer to stay with that vendor and add new line cards or new modules, including 400-gig ZR modules from that vendor to the existing system that is really relying on that optical management plan and all the features they're used to. Whereas in the metro DCI world, there never really was a separate optical management plan because it's a straightforward point-to-point network topology. And the large cloud customers have enough internal expertise to set up these networks and set up the link projects and do the traffic engineering on their own, that a traditional vendor like Ciena provides to other customers. So there's a big split between large cloud that has a lot of demand for these modules, yet is also sophisticated enough to do the lowest cost implementations, which is to put the optics directly into a switch routing port versus a traditional customer that already has the existing transponder and line systems, where they will add the same modules to an existing system. And that's just the easiest way for them to expand. And both models will coexist in the industry, but what is true is that the highest volume is going to be the greenfield data center DCI.

And Andy, from a revenue market opportunity, looking at -- I recognize next year is going to be the first year of shipments, these things always take time to ramp, but any thoughts you could share with us in terms of the market opportunity for Arista in '21 and beyond?

Yes. To be quite transparent here, our cloud customers tend to purchase their own optics. So part of the value proposition for them is they can buy the optics modules at cost -- wholesale costs from the optics, the various optic manufacturers. And so we will not get any revenue from the optics themselves, but we will, of course, sell them the routing ports that goes with that and enable them to deploy more bandwidth. So maybe a better way to think about this is that even a company like Microsoft and Google and Facebook, which are spending a fortune each on CapEx, they still have, of course, budget for each of these various functions. And what they really wanted all along, in the case of optical networking in the metro, was, say, 10x the bandwidth for the same costs, right? So they have a great desire to increase the amount of bandwidth between the various data centers, but they can't spend 10x the money doing that. That wouldn't be cost effective. So it is that ZR then enables them to do this. And this will lead to new deployments of more tightly connected data centers between the metro, in particular, to allow them to do soft data failover and other mission-critical applications, where you can essentially expand a single data center across the metro if you have enough bandwidth to work as one single larger data center. And that is very attractive for all kinds of applications.

Andy, can I push you to comment in terms of market -- go ahead.

One more thing. We -- obviously, we also sell the solutions to certain customers like small customers. They want to purchase it from us with -- and the support we will provide. We also introduced recently an optical line system in a single OSFP Form-factor Pluggable module that enables us to build a point-to-point link like 100 kilometers. And it's plug-and-play. You plug it in and it works. It's really quite nice, even I can do this. That's -- again, this is more for like building-to-building greenfield kind of applications. A customer that wants to connect their Manhattan facilities to a data center in New Jersey. These kind of things where customers have dark fiber and they want 0 latency, no overhead, no queuing kind of traffic engineering on these things. And there's no better solution than deploying 400-gig. Not going to fill up that link anytime soon. So there's other use cases, even in the 10 to 40-kilometer range. There will be special SKUs that they use, optics, not even coherent, that are even lower cost, right, to displace the traditional 100-gigabit links that are used for building-to-building today. So there will be at least 5 different SKUs of these 400-gig ZR modules from ZR- to ZR+ and building amplifiers and so on. And it really covers a very, very wide spectrum of opportunities. And again, we are more than happy to sell our switches and routers to any customer in that space that wants to deploy this at the lowest sort of most cost-effective fashion.

And Andy, I apologize if you said it, I might miss this, but can you give -- I know you've looked at the market numbers. Can you give us a sense for what the dollar opportunity is over '21 and beyond for Arista?

Well, I can't speak to an Arista-specific forecast, but we have shared certain estimates for the total opportunity for 400-gig ZR modules over the next 3 calendar years, meaning '21, '22, '23. And we have provided this to certain market analysts, and you can get their reports. But the number that I can quote you is 500,000 ZR port over 3 years and includes the ZR- kind of ports as well. And I do think that's really [ significant ] number the bulk of that will be in cloud.

Okay. I've got some other questions for you, but I'm not sure...

I'm not saying this is our market, right? But this is a large number of protected ports that are essentially all routing ports that are phasing to another location in the metro or further beyond.

Understood. Jayshree, are you on? All right. Andy, I'll keep it on with you.

Okay. So can I actually mention one other thing, which is 800-gig ZR. So there's no standard yet for 800-gig ZR, but it is the obvious next step of the 400-gig. And it will provide additional cost reduction and accelerate that could transition to IP over DWDM in the time frame. This is just getting started at the OIS as a standards effort, but there has been particular interest in expanding the use cases of these optics into the data center, even into the 2- to 10-kilometer range, which is more difficult to achieve at the 800-gig speed with traditional 4-channel optics. So it is conceivable that these optics will be cost-effective enough to actually have volume inside the data center, at least in these large cloud data centers, and that would be a very substantial volume opportunity, in addition to addressing the 100-kilometer metro range, which is, of course, the other key unit.

Okay. Andy, let me ask you...

So these things are about 3 years out. So...

I'm sorry. Say it again.

These things are about 3 years out. So this is not the day after tomorrow. Okay. Paul, your question?

Got it. In the enterprise campus, obviously a big opportunity for Arista over the coming years, how much room is there for differentiation? Cisco's been in this market now for 25-plus years. They've got umpteen products, a tremendous amount of invested, and they keep pushing the technology forward. How much room is there for you to distinguish your products, your architecture and thereby to gain meaningful share?

Which market segment was it you're saying seeing in switching or routing or --

Apologize. Enterprise campus switching and wireless LAN.

Oh, enterprise. I'm sorry. Yes, in Enterprise, of course, we're very familiar with the market. This was my -- I had a product line like this that's been at works at Cisco. And Jayshree, of course, had -- grew up in this market. The expectation we set for our first year of shipments starting last Q3, basically July until June this year, was for $100 million in revenue. And I think we commented on our earnings call that we are well on the way to achieve that number. It is true that with COVID, there's a little less urgency right now to look at campus buildouts or upgrades since people are working from home, but this may be a temporary kind of issue. There are technology transitions in the campus to -- primarily driven by the WiFi that needs 2.5, 5-gig or even 10-gig connections in the future compared to the traditional WiFi segment with gigabit Ethernet. So here there is an upgrade cycle here, but it's much slower -- it's a much longer, slower rolling upgrade cycle than you've seen in the data centers for every 18 months to 2 years and the chip comes out. So we do see customers that are familiar with our value proposition on EOS on the data center side that are very open to installing our campus solutions. And we're still at the very beginning of that journey, obviously, but it is a large so...

Let me pause -- let me see if Jayshree has joined us. Jayshree, are you on?

Yes, I have. Hello, everyone.

Oh, that would be the perfect question for you, Jayshree. The question was, what about our opportunity in campus enterprise?

I was put on hold, and I was listening to some big violins, so I didn't hear the question.

The question was, I recognize you've spoken about the $100 million target, which you're on track to hit in campus enterprise, but the focus being longer term, what's the opportunity for differentiation? And we all heard your comments on the quarter. I won't ask you to repeat them out about the challenges that COVID-19...

Yes, I know. I appreciate that.

Yes. Right. About the challenges COVID-19 has presented. But those apply to all companies, and those are near-term events. But longer term, Jayshree, how much opportunity is there for you to differentiate in campus enterprise and continue to drive that forward?

Sure. Well, first of all, thank you for the opportunity, Paul. As you know, you've been a long time friend and fan of Arista for well before we went public. So this must be at least your 10th year of association with us, if not longer. I think the...

Jayshree, I love you, but I have to stop you. I'm never a fan. I have great respect for you and Andy in your company, but I'm not paid to be a fan. So go on.

Okay. That's great. A friend maybe. The -- I think the real issue on the campus, if you step back, and I'm going to take you all back into the '90s a little bit, is the campus migrated from shared hubs to structured wiring and switching in the '90s and 2000 era. And we basically had 30 years of the same architecture, which was -- I think was a fantastic migration from shared hubs at 10-gigabit to switching at 1-gig and beyond. And this architecture was the advent of the worldwide web. And the familiar network architecture that most campus networks are built on, which is access allocation and core. And then there was a separate WiFi architecture that was built on controllers. So the reason I bring this up is to say, "Okay. What is the risk of differentiation?" And it's really threefold: First of all, we think there needs to be one seamless edge for wired and WiFi because there's an explosion of IoT devices, users. And no more can you say 1 user equals 1 PC or 1 iPad. You've got 1 user equals 10 different devices that we all carry. So dealing with this explosion of clients really requires you to have a single campus edge, no matter what your connection might be, wired, WiFi and, in the future, others like 5G as well or WiFi 6. So this is a huge differentiator for Arista because we're able to now bring a familiar data center architecture leaf-spine to the campus. By collapsing these 3 layers, access aggregation core into a flat fact leaf and spine, where the leaf is now no more just a server top of rack or storage, but really a WiFi leaf or a PoE leaf, connecting into the same fundamentals and principles of the spine, and you don't have to have a different campus spine or a different data center spine. At the same time, we're not disrupting their cabling or their IDF or their MDF infrastructures, and we're bringing this controller-less architecture. That's the first differentiation. The second one is software-driven. When we say cognitive, we mean everything from the client, to the WiFi, to a third radio scanning with the right inference engine in ML, to what's very topical today and being able to prioritize Zoom and Google Hangouts and Microsoft Teams over smaller pipes like WiFi, to a cognitive management claim, where we can really understand the traffic patterns and respond to them and automate your workflows because in the campus, you're going to have so many users. You need a day 0 automation, and you may need some real-time streaming and telemetry. And the final differentiation, I would say, is quality of experience, including segmentation and security. You can't bolt security into the campus. You really have to think about it all the way from the client to the cloud and working with partners like Palo Alto and Forescout, we will take their intelligence with network admission control or firewalls and bring them into the infrastructure seamlessly to track flows and also to prioritize them and provide the right levels of segmentation. So this may have been a long answer to a short question, but, hopefully, that gives you the idea. And the benefit of that is the single [ request ] and one single management claim that can work with all the multi-vendors.

That's a good answer. I appreciate that. Jayshree, if I could ask you, take a step back, can you give it a long-term focus? Looking beyond COVID-19, assuming as the world returns to some sense of the 21st century, the thought arises, not that you ever want to profit from tragedy, but as workloads likely get shifted to the cloud as a result of the pandemic that this actually long term could be good for Arista to the extent you can hold market share with your cloud signing customers. That's been such an obvious big part of the equation. So that said, looking beyond, as we emerge from the crisis, what are the market opportunities you are most excited about for the industry and for Arista?

Yes. No. Like I'm sure Andy described, one of the things that I'm most excited for Arista right from the beginning, and it continues, is we are disrupting in 3 ways: We're disrupting with our data plane with merchant silicon. And Arista's, I believe, in the best position with the drivers and abstraction interfaces we develop to apply that not in the data center but in the routing layer, in the campus layer and extend that to the branch and really help not only our cloud titans but really help our service providers and enterprises pitch the Internet. So that's one huge opportunity. I call this places in the network, which were silos that become places in the cloud. And a customer shouldn't care whether you're building a campus network data center or network or branch network, they should all be like consumption models of small, medium, large, the largest being public cloud providers. So I think the opportunity is well beyond the data center, where you can -- the data center becomes the starting point, but the ending point could be at the client with the campus or the cloud with the public operators, cloud titans, as we call it. But perhaps what excites me even more is the clarification of SDN as a definition. Many would still say SDN is a controller architecture or much more of a single controller. But Arista always thought about this as solving, again, 3 types of problems, cloud scale, how can we build hundreds of thousands of servers or billions of virtual machines connected to provide the right automation and right availability, which is why we -- that availability that at the box level, at the network-wide level, at the spine level and at the software level, I think, is a huge differentiator. No one's been able to do this, and Arista software stack is continuously improving to provide all types of availability. But the other thing I see is that the market will shift from SDN being a name to SDN making networks more operational and easy to use. So that goes all the way from automation to availabilities to allow the built-in, real-time streaming and telemetry and flow tracking we're doing. So doing them at 1-gigabit is one thing. Doing this at 100-gigabits and 400-gigabits take some real skill. And then finally, the right programmatic APIs. I think to talk to the cloud, each one of them has their own overlay architecture. So I call this the 5 As, where you -- Arista differentiation started with building a highly available network. And today, we're automating it. We're analyzing it, making it more programmatic with the right APIs. And eventually, we have to put the right agility, whether it's in a virtual form, a containerized form or a physical form, to talk to the cloud.

I appreciate that. And Jayshree, again, looking at calendar '21 and beyond, do you agree with the view that the current crisis will lead to increased demand for data center switching because there will be -- and I'm very mindful that the big surge we've seen to date is largely not entirely a result of video, real-time communications, whether it be Zoom, Webex, Microsoft Teams, et cetera. And that doesn't implicate the U.S. West traffic phenomenon that would, in turn, translate to incremental demand in terms of capacity for data center switching, whether from you or others. But thinking about longer term, exiting the crisis, do you think, in fact, that the demand environment is better than it would have been, but for the crisis?

I think the crisis is too connected to the demand environment than I agree with, whether it's now or 2021, and then a step back from the health care pandemic crisis, which needs to be solved with vaccines and the right therapeutics. And maybe there's an element of building the right AI clusters to do more forensics and data science on understanding this virus from what it does to the network. Because from my perspective, whether I work from home or whether I work from the office or whether I work from a branch, I'm still a user that's putting tremendous pressure on the network, independent of my location. And the type of pressure I'm putting is so slightly from video collaboration applications, but much more from large data sets and cloud-native applications and business-native applications. So I think we're connecting 2 dots that, in my view, are 2 different problems that -- they're a little bit related. But I think if I had to step back, I'd say, independent of where the user sits, there's no doubt to the fact that the cloud is here to stay. And it's a powerful data set that both public cloud providers and private and service providers, who want to build off of cloud principles, have to do. Now the question then becomes, who will do it? And I still believe it will be 6 of 1 and half dozen of the other. There's some applications and workloads that just naturally are friendly and belong. And you can move the workflows, move the workloads, move the video work streams, all of them in the cloud or in a co-location like Equinix. And the others that are just so mission-critical and security-dependent, they may stay on the premise. And I think the biggest challenge going into 2021, 2022 is this definition of a multi-cloud, hybrid cloud architecture that can seamlessly move across clients, public cloud and premise cloud for lack of a better term. And that's what I get excited about is the bright future, independent of whether the user's at home or in a branch or in a headquarter.

Okay. I've got several hours more with the questions, but let me pause and give an opportunity to the audience if anybody wants to ask a question. Operator, can you pool the audience?

[Operator Instructions] First question is from the line of Peter Law with Analog Management.

Your comments on 400ZR were very interesting. And just last night, Inphi, on their earnings call, talked about acceleration and pull-in of 400ZR opportunity and that their hyperscale customers who might align with your hyperscale customers are looking to accelerate that. Are you seeing that in your testing and your sampling of those products?

This is Andy here. So yes, we can't speak to their schedules, but it's fundamentally limited or driven by the component availability of the 400-gig ZR module. So Inphi is one of several vendors that is providing these modules, and they may have some shipping by the end of this year that are prequalified or maybe in volume next year. But that has been roughly the time frame. But obviously, the cloud customers would like to deploy these things as soon as they can. It's really limited by the remaining qualification time.

Okay. How is the qualification looking?

Yes. So just to add to what Andy said, Arista has had a portfolio of 400-gig switches. I think there's about 10 products we already have. So in my view, the cloud titans are more limited by the ecosystem of optics. And indeed, companies like Inphi and others are and should be doing trials and testing with the cloud titans because it's very important that to get a kind of general availability even in 2021, you have to prove that out. I think this year will be a year of trials and hopefully next year will be the year of deployment.

Any color on how the trials are looking?

I don't know if we can comment on Inphi. So...

We can't really comment on different vendors. And what I said earlier is we had received these modules. We are testing them in the lab, and there's modules from multiple vendors. And things are looking promising, but the thing to understand is for these things to get production released, they have go through a lot of reliability testing and qualification testing until they put the wafers into the fab to actually ramp the volume. So that just -- it really takes time. So even if it's, say, fully qualified by this summer, it would still take almost 2 quarters just going through the TSMC fab process to ramp the production of the silicon. So you see it's a component ramp issue.

[Operator Instructions]

Andy, Jayshree, while we're waiting if there's any other questions, one of the clients pinged me, and it's a good question. Mindful that service provider has been slower developing for you than you had first hoped, and I think there's good reasons for that, but one word that hasn't come up on this call today is 5G. I recognize that there are other issues going with service provider of 5G in terms of wireline impact, whether routers or optics, it's still somewhat early, but what are your thoughts with respect to befitting from 5G? You obviously have routing capability, albeit not full capability, relative to your peers, but I assume that will improve over time. What are you thinking about in terms of the impact longer term?

Yes. So I can kick it off. From our perspective, the last mile, whether it's 4G, 5G, cable, it's metro Ethernet, et cetera, Arista has not really played in, but, obviously, it has benefited from that because the more bandwidth you put into those users and cell towers, the more back end, backhaul aggregation you will need for service providers. So from our perspective, we see only benefit with an increase in 5G and/or even the increased use of 4G. I also think that the line between LAN and WAN is getting thinner and thinner. So use of 5G also connects greatly to WiFi and campus networks because you've got to have a seamless end-to-end connection. So we're big fans of 5G. I think it will take time, and really, driven by the folks who are manufacturing 4G and upgrading the towers, which like 400-gig is a process.

Operator?

There are currently no questions waiting from the phone lines at this time.

Let me ask you another question posted by a client. You mentioned the other day on the earnings call, I think you've put out a press release recently pointing out that you partnered with Microsoft and SONiC, enabling your hardware to be sold to third parties. Can you help us understand that? I could see why you would do that, but let me let you address the benefits to Arista from taking that approach.

Sure. First of all, as you probably know, Arista has a long history, since inception, both industry collaboration and open networking. It started with merchant silicon. It started with a lot of our programmatic APIs. And as you probably also know, we have worked with large titans like Facebook on integrating their EOS And SONiC on integrating with our EOS as well. So I view this as just another step. If you sort of step back and say, what are all the aspects of open networking? There's the industry standard chip for merchant silicon. Then there is the right attraction interfaces, and this is what Arista introduced, that next step in our development efforts, where the engineer-to-engineer collaboration and contribution really happens if you can speak to the silicon with either our EOS or other in office. And Arista has always worked in tandem with other network operating systems. We view our silicon -- our switch abstraction interface can work with multi-process ASICs and management software and work into SONiC as an open networking operating system. This doesn't, in any way, change our views or change our customers' views on the importance of the EOS, but it gives our customers flexibility and choices. So they can have a containerized EOS, so they can have a physical switch with EOS. Or they can have a bit of both and run both SONiC and EOS on our APIs or, in this case, the SAIs.

I've got 2 big picture questions I'm hoping to get in before we end the call. One is, clearly, optics, optical components, silicon photonics and switching silicon appear to be converging, and you've got various players that are moving towards putting them on the same die. And are there implications? You guys clearly made a choice from day 1. It seems to be a very good choice that you'll focus on the software and utilize the very fine silicon out there from Broadcom and others. Does it change going forward 2 to 3, 3 to 5 years from now as others come into the market with integrated optics and switching silicon solutions? Does that change the equation for you at all?

So if I could answer this. Silicon photonics is a technology that's used inside the optics module that is more cost-effective and more reliable than traditional ways of building these optics from individual components. So this is really not that visible from the outside, but it's critical to bring on the cost of the optics and to enable scaling there. We are working with all the known companies developing silicon photonics to understand how these things may evolve to become eventually co-packaged. And these are clearly a couple of years away from reaching production status, so this is not something that will affect either our product designs or our customer decisions in the short term. But again, if there is a viable solution that is available in the future, we would obviously deliver these solutions to the market.

Yes. And just to add to Andy's, I think the concept of co-packaging options with silicon is certainly technically feasible. The question you have to ask is, are you trying to do this for economic viability? Or if each component were economically viable, why would you want to do that? It may be flexible to have them as separate. So certainly, I think as Arista delivers more and more multi-terabit systems, and if you start looking at what you do with them, we can see the flexibility of having multi-terabit systems with adding any type of optics, which is one type of option. And then if you're truly trying to give one specific flavor, then you may want the co-packages. So we won't be taking a position on that, but I don't think it, in any way, change -- the co-packaged optics doesn't in any way change the future of individual optics and individual switches. I think that market is ripe and vibrant.

Yes. I will -- if you take -- one of the quotes from a friend of mine, who's in the optics business, he pointed out to me the other day that the idea of co-packaging optics has been around for a long time. And he added the problems haven't changed either. So this is just not a soft problem, if you know what I mean. I mean there's all kinds of yield issues in manufacturing and who takes the ownership to do this in a clean room setting, et cetera, that have just not worked out in our future, but it's not here today.

Jayshree and Andy, I started this conversation by pointing out that a lot of investors for many years now and still today have worried about commoditization networking broadly across product segments and across vendors and, Andy, you responded earlier. Now I want to end by closing the question, is networking even more critical today than it was 2, 5, 10 years ago and likely will it be even more critical in the future counter to the commoditization argument? I always worry that, as a sector analyst, I'm biased. But I certainly think it is becoming more critical as we seem to be recreating the old mainframe, but in the cloud and divorcing the constituent components. But I'd love to get your view on the issue.

Sure. So I think networks -- go ahead, Andy.

My view is that networking has never been more critical than today because failure is just not an option, right? So any business, larger or smaller cloud, has to be online all the time. You cannot get off-line because you're dead, right? So networking is the most critical piece in the whole IT offering. And this is very different from service or storage where an individual server can sail and the application will continue as long as there was a copied data somewhere else on the network. So the network has to survive. And that's, I think, our chief value proposition to customers that we do have the most resilient network operating system and offering.

I think Andy said it best. I think it's a general mix that even the large cloud or enterprise companies just want to deploy commodity boxes. I think they're looking for their main integration as mission-critical, got to work, integration and control of their environment. And the network couldn't be more critical today or in the future to make that happen.

Well, I've got several hours more of questions, but in the interest of keeping everybody on schedule, I greatly appreciate you and Andy taking time out of your schedules, out of running the business to share thoughts with us. And I hope this was helpful for all the clients on the phone. Again, we've got another 20 calls set up over the course of the next 4 months. There are 2 or 3 next week. And if I could give any help as always in your investment process, it would be my pleasure. Thanks again. Please stay safe, everyone. Have a wonderful weekend.

Thank you.

Okay. Thank you. Bye-bye.

That includes the Cowen Conference Call with Paul Silverstein. Be safe, be healthy, and enjoy the rest of your day.