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

All right. I think we're going to get started. Welcome, everybody. Great to have you here at the conference. Hope everybody is having a good time, fully caffeinated. My name is Rod Hall. I'm the communications technology, infrastructure technology analysts, among other things at Goldman Sachs. And I've got the great pleasure of having Jayshree Ullal here with me, the CEO of Arista Networks. So Jayshree, welcome.

Thank you, Rod.

Great to have you here.

Great to be here.

I remember the days when you were at the booth that trade shows before Arista was public talking about the switches and the technology. So...

I'd love to do that again, but the pandemic has changed that. We have more virtual booths now.

Things have changed a little bit. That's true. I thought maybe before we jump into questions on supply and a lot of the things that we've been talking about with lots of different companies, I thought maybe you could give us a little bit of an overview from your perspective on where Arista is today, where you're taking the company and just kind of how you see the world from the perspective of the CEO seated at Arista.

Yes, absolutely. Like many start-ups, I think Arista is still considered itself at the heart and core of who we are, a highly innovative company. Of course, you're not going to hear any CEO come here and say, we're not innovative. But when you look at our innovation, I would classify us as sort of migrating from Arista 1.0 to 2.0. 1.0 where we were much more of a data center company and 2.0, where we are moving to more centers of data and excellence in networking where the data may reside anyway, anywhere in any way in the sense it can be in a data center, it can be the campus or branch or you have to deal with observability and security of that data. So very much still networking driven but more data-driven networking. So I think Arista 2.0 allows us to participate in a lot larger TAM and also makes us more relevant to our customers.

That's great. Thank you. So I'm going to try to get the boring things out of the way first, and then we'll get to the more exciting things later, if that's okay. So let's talk about supply. We've talked to everybody at the conference at least that I've interviewed about supplies and where things stand, how it's affecting Arista in this case. I wonder if you could just catch us up on where do you see the supply constraints and maybe when things start to ease? Or are they already starting to ease from your point of view?

Yes. Well, if you all step back and look, we've been in a supply chain crisis for at least the last 18 months. And while you hear rumors of things easing up, at least from an Arista perspective, one of the advantages of having a supply chain crisis. If I can look at the glass as half full for a minute, is we get greater visibility to the demand. So that's been helpful. One of the downsides is, of course, we're unable to fulfill the demand. And we don't see the supply chain easing for at least another year. So we see this going into 2023. And the reason is simple, although my team, our CFO, Ita, who's with me here today; and COO, Anshul, have been very good about planning ahead and planning multiple years. The reality is we need all components to build a system, and we're just short of a few components, but the few components are kind of like the golden screw problem if you don't have the whole hardware, then you can't ship the whole system. So we're waiting for these last few components, and it's been a long wait, and oddly enough, these are the more simpler components with simpler processes but are very, very scarce and there's a huge shortage in them. So once we improve the lead times on those components, which again, I think will take us into next year. We hope there will be some more relief in the supply chain. The cost of the supply chain is also exorbitant nowadays. We find ourselves paying an order of magnitude move for some of these simple components that 100 to 500 in prices, there's the price, the cost of logistics, freight, the rising inflation, everything is costing us more. So we're dealing with both the lack of supply as well as the cost of supply being much higher. So pretty tough. But again, I couldn't think of a better team at Arista that's navigating through these different issues extremely well, but it's not been easy. It's pretty challenging.

Well, given that, we've seen some of the consumer electronics stuff like PCs are very weak now. So you would think that, that would free up some capacity at some point. Are you feeling like if anything, this expectation that goes for the next year is too conservative? Or do you see other things out there, maybe we don't see that caused you to think, well, no, it's going to be going on through the end of '23 or at least to...

Yes, on one hand, you're definitely right. We see the slowing down of the consumer in PC sector. On the other hand, we have not seen a recession in the data center and cloud sector. And the other thing to keep in mind is in the last decade or so, you hardly saw any chips in the automotive industry. And today, a single car can have 8,000 chips. So you are dealing with additional high-tech sectors, including the automotive that are competing many times for the same type of chips, industrial, automotive, data center are all in high demand and are more than making up for the weaknesses in other sectors.

Right. All right. Let's get a little bit more exciting, not all the way there, but let's talk about hyperscale a little bit and your cloud titans customers. They slowed down a little bit in terms of growth in '19 -- from 2019 to 2021. Things seem to be picking up again now. Can you talk a little bit about the dynamics of that market and how you see their capital spending patterns may be evolving over the next couple of years?

Yes. So I think, first of all, the cloud titans is one of Arista's most exciting set of customers. Of course, we're diversifying to others like Tier 2 cloud service providers and enterprise, but Arista's cloud networking and all of the pioneering missions and our success and the reason for us going public with some of these extremely high-profile customers chose Arista and chose to center their cloud titan investments and infrastructure on us. As you know, we've stated publicly that Microsoft and Meta are likely to be 10% concentration customer. So those are 2 very good examples among many others. Now the thing to remember about these type of customers is the CapEx investments of these customers is significant. They can be measured as not just in millions, but in billions of dollars. The networking is a small piece of it, but an extremely relevant piece of it. And while it's volatile, you have to connect your storage, your compute, your AI workloads, your metaverse, your virtual reality, your gaming, your simple client server, and you're building the kind of scale that you simply don't see in other industries where you can measure the capacity and petabits, you can measure the servers in hundreds of thousands. You measure the storage in ways that you just don't see that kind of scale in a typical enterprise environment. So Arista has benefited from that. From 2014 to '18, we had significant growth, often in the 30s and 40s percentile. And then we slowed down a little bit in '19. And now we -- as you know, we had said we will grow at 30% and all indications in 2022 are we will grow north of 30%, maybe even 35% -- north of 35%. So we're having a banner year. We had a really great year last year as well, and we expect to have good years going forward. But it is a volatile sector and some years will be extremely good and others less so. But if you average it out, I believe we can still experience double-digit growth in this sector for many years to come.

Yes, I remember going into that 30% indication that you gave, we all thought 15% would be great. So that was a...

Yes, you thought...

Positively shocking there. I -- it reminds me of -- I've got a -- you gave me a plaque that says the greatest turnaround of an analyst as a joke a little bit. But that reminds me of when I -- we had seen commoditization of switching and hyperscalers at the top of rack. And I thought all this is going to move up into the aggregation layer. And you said not so fast, Rod, I don't think it is. And you turned out to be right, which is why you're the CEO of Arista, and I'm still an analyst.

There's -- many times, I'm wrong, too, but the number of rights have to be greater. That's all.

It seems like that, that theme continues to play out. We're now starting to see proprietary storage companies, like Pure selling into these hyperscalers, and I wonder if that -- I wonder what you think about that theme? In other words, did the hyperscalers start to realize that really spending this R&D on this technology, they're not getting the most bang for their buck, and it's much better to let experts like you deal with that. How do you feel that trend is going with hyperscalers?

I think you summarized it very well, Rod, which is every company, including Arista, has to decide what's core and context, right? For example, in Arista, we have made a very conscious decision to invest in our software and that's core to us. And what's context is we make sure we work with the Broadcoms and Intels of the world to -- with their merchant silicon. So you can have lots of money, but you still have to figure out how to use it and how to prioritize it and make the best of it. And I think that's exactly true for the hyperscalers or cloud titans, as we call them themselves. No matter how much money they have, their mission is to build some of the greatest world class cloud data centers, right? And that requires power, cooling, investment in scale compute storage. And at the same time, they can't do it all themselves. They have to spend their money on CapEx and leverage us for some of the OpEx, maybe savings they can get exactly the same for storage. But it's never that binary of either ore. One of the greatest successes of our large customers is that we don't just supply to them, but the word partnership really comes to mind because a lot of our activities are co-development. It's engineers talking to engineers. And so as they want to put SONiC or FBOSS on their platform, in some ancillary use cases, we're right there with them doing that on our platform.

Right, right. And a lot of those attempts to put their own operating systems have ended up in deciding to just go with your operating systems...

It can go one way or the other, but the important thing is the co-development to give them choices because remember, these guys have to -- have multiple sources, multiple vendors and multiple options. And we may be the preferred option, but we're -- we never be the only option.

So one of the things I know it's a big opportunity for you is this RSC installation at Meta, huge AI installation that seems to be growing by the minute. But I'm curious what you think about AI workloads, generally, they require much higher speed ports. How big a driver is that going to be for hyperscale demand as we look forward the next couple of years?

I know it's fashionable to talk about AI and artificial intelligence. But for the longest time, actually, we couldn't figure out what AI was useful for. We thought it was more of a lab exercise. But I think we finally figured out 2 very important use cases. One is in security and root cause analysis for some of our AVA, Autonomous Virtual Assist, which is separate than the question you asked, and the other is how to enable these AI workloads on our network as a use case. So what is AI? It's probably not exactly all the things I'm used to generationally like AR, VR, gaming, virtual reality, real reality, whatever it is. However, there's a suite of applications that are non-enterprise applications that is going to feed the next generation of network and put tremendous pressure on them. So a typical AI workload, if you look at it, it goes through a very intense computation matrix, where you're constantly computing at cycles of computation across hundreds of thousands of CPUs, GPUs, DPUs, CPUs, whatever you want to call these processes, right? And this cycle continues. And then once you get to the local data, you start again. And so the intensity of this is not only is it compute-intensive, it's extremely network intensive. Now when you look at what Arista has done, we haven't been here before. We've largely built the front end of the network for this. The back end of all this compute intensive has historically been PCI Express or InfiniBand. And now with these kind of AI workloads, we're really putting pressure on the network, not just for scale and bandwidth, but to have the right storage and buffer depth capacity to have the efficient spring, to have predictable latency, to have the ability to handle small packets, large flows, elephant flows as we call them. All of this, just translating it back into engineering is a tremendous amount of innovation required on traditional ethernet to make it run as a back-end network. And that's what's exciting about our work with the Cloud Titans, and we believe this will have use cases much like we pioneered cloud networking in our first innings back at IPO. We will lead our first innings on the AI spine. And I think we've got another decade ahead here to figure out all the apps and all the different types of networks we have to build for...

What I found interesting is RSC at Meta started off as an AI platform really allowed for that. And it seems like has evolved into also doing computation for VR, AR environments now. So do you think -- is that -- do you think that will be unique to Meta? Or do you think others will also be computing these kinds of environments on similar infrastructure?

I think everyone will have their specialized applications, but there will be a huge set of horizontal compute-intensive applications. They may not all be AR or VR, some may be gaming, so maybe the distributed computational for crypto. So there may be different applications, but the use case will really be horizontal for the network.

Right. We hear other competitors, Cisco, et cetera, talking about gaining market share at hyperscale. It's hard to pin down exactly where that share is. Are we talking about campus switches, we talking about data center switches? How do you feel about your share position now? And how do you measure it? When you look at market share, how do you think about it?

Yes, look, I think Cisco is a great company. I was there 15 years, and they're certainly the dominant market share leader in campus. I think specific to the data center, the stats are proving that when you have one dominant company like that with 80% share, which they did have 10 years ago, it's almost unnatural, and the customers seek alternatives. So Arista's share has grown from 0 to -- we've been in the teams for a long time in high-performance switching. And just on Friday, we received data that we've gone from being a teenager to an adult, and we're now in our 20s. So we still got room to grow. But as our competition has come down from the 80s to the 30s and 40s, we've gone to the 0 to 20s, and we've got a lot of upside to do better. In addition to that, we're doing particularly well in the high -- the highest performance, the 100-gig, 200-gig, 400-gig where historically an incumbent who gets share in switching always gets share in high-speed switching, too. But Arista has been the #1 market leader now for 3 years in a row in 100-gig, 200 and 400-gig. And the reports last year showed that we're almost double the share of our peers in the industry. So I feel pretty good about our share, and the TAM is growing, and I think we can continue to gain share independent of our peers.

Let's talk about 400-gig a little bit because that comes up a lot with investors. How big of a cycle is 400-gig in your opinion, 100-gig was a big cycle. A lot of people, I think, mistakenly thought 400-gig would be similar, but it's a different type of cycle. So...

It's different, yes. Well, there are similarities to the 10-gig cycle when that upgrade happened, the 100-gig cycle and the 400. So I'll go back a little bit in history to sort of come back and answer your question on 400. If you look at the 10-gig cycle, back then the servers were barely running at 10 -- 1 gig. So 10-gig actually took a long time. It was a very long tail of 10 years before you actually got 10 gig at a cost-effective price point to be a server attach. It started initially as a backbone in the 2000 era. And I would say as a server attach, it really happened in 2012, right? The 100 gig happened faster because of economics, the dollar per gigabit was much, much cheaper. And then the optics and the connections to the server were much cheaper. So I would say that 10-, 12-year cycle that took -- the 10-gig phenomena actually compressed to more like 7 years, especially with the hyperscalers, right? Now obviously, there's a long tail that continues into the Tier 2 cloud providers and enterprises. So the 100-gig cycle is very much still alive and well and continuing with both the Cloud Titans and our customers. And it started back in 2018. So we still got a nice 10-year headroom for that. In addition now, you've seen the interleaving of 200 gig and 400 gig happening and some of the optics vendors are coming out with 800 gig as well. So what you see is a very nice interleaving of 100, 200 and 400 where the 200 and 400 are happening more as an aggregation to the spine and the 100 is happening more as a server attach. And so you're seeing that's really the best possible era in that 100, 200 and 400 are all interleaving to give us a large opportunity for high performance as a category.

Yes, I'm always amazed. We talk about these geeky things as we've been -- both been around networking for so long, but they're powering these unbelievably powerful amazing computing infrastructure that's allowing everyone to sit out there with their laptops and their phones and so on and do all the stuff.

Yes, it is pretty amazing. The number of devices, users, applications that are putting pressure on the network, I -- 20 years ago, we used to say there's no load on the network, not the case now.

Right. So one of the things that we've seen, Google has just written a white paper, published it using direct detect optical switching in the aggregation layer of some of their data centers. How big a threat do you think optical is for Arista? Does it come in and begin to displace some of the router switch, switch routers that you're selling? Or is it something that works alongside them? Can you just talk a little bit about that sort of an architecture and how you think about that strategically?

Yes. When you mentioned the paper, I didn't get a chance to read it before the call we have here. But look, optical has always been a companion to routing and switching for so many years, right? So we don't view optical as a threat. We sell optics with our switches. We sell integrated optics at the data center interconnect layer with encryption capabilities. And we also partner with optical vendors. The reality is, depending on the customer and how much access you have to fiber, you can get very creative on how you combine L1, which is what a lot of the optics is, with L2 and L3 and build a really nice IP and optical network. And that interleaving of [ lambdas ] with L2 MAC addresses and L3 routing addresses is fundamental. But I would say it's going to be an IP plus optical era for a long, long time to come. Not only L1 or not only a L2, L3 switching, Anytime we do switching or routing, we've got to connect to something. And generally, we are connecting through optics, whether it's long haul or metro or short-haul. And maybe only in the intra data center do we see more of the copper, but almost everywhere else it's some form of optics. Particularly as the optics and the SerDes technology gets more and more advanced, like 800-gig and terabit, I think the importance of bringing them closer together, getting the right power capabilities, getting the right distance is going to be super important. So I think you'll continue to see them work in tandem and integrate as much as we can as reliably as we can. And sometimes, it will be through us and other times it will be through pure optics.

I've seen a lot of times these arguments made that some new technologies coming along opticals, a classic example. But it turns out that the automation layer technology, the software, all those integrations are the thing that really is the value in the product and...

Exactly. I mean in optical, you can move packets really fast at the optical layer. But then eventually, you need to convert it into electrical layers and then like you rightly point out, do the right automation, availability, agility, analytics, AI-driven workloads. And those all have to be intelligent software on that.

So there's been a lot of talk about silicon diversity, one of my pet theories for why Cisco has made some headways so they have their own silicon. A lot of people maybe liking the idea of diversifying away from Broadcom, which is dominant. What do you think about that? How is the risk to think about the silicon situation that's out there? And do you think this is something that -- I know there's a demand for it, is it really going to play out where we have silicon diversity in these data centers, do you think?

So I'll let you in on a little secret. When Arista started the company, we tried to say we're just going to build software and not worry about the merchant silicon. And we went to all our customers and said, what do you do with your software? We said, we put it on your merchant silicon of choice and run it. And they go, we don't know how to do that. Just like we -- all of you with laptops here, you don't build your own PC. There's some advantage to getting a fully made PC and loading your applications. It's really difficult to go to a customer and say, "Here's your software running on whatever silicon you want." They don't -- the network is too mission critical to do that. So Arista took on the responsibility of working with open merchant silicon vendors. And certainly, we have our share of vendors. We've worked with Marvell with Intel, with Broadcom. Broadcom has executed -- or executed, I would say, everybody in having the best portfolio of switch silicon. If somebody else were to come with better switch silicon, we would be open to using that. And if Cisco would ever build an open merchant silicon, we would be open to looking and reviewing that, too, has to be competitive, has to be customer-driven. But it's really hard to do open merchant silicon. And I think the reason Broadcom dominates is because they've executed flawlessly, not just with one family of chips, but a generation of families of chips, both for Tomahawk, Trident and Jericho. So [indiscernible] keep doing a good job and please keep supplying more chips as well.

More chips made would be handy. So your prediction or at least your view is that there's some probability for this to occur, but it's a fairly low probability that we'd see a lot more diversity in chips maybe in 5 years' time and...

Anytime I look for one, for example, we've invested in the Barefoot silicon Tofino as it's called, and now Intel has acquired it, so we do use the Intel Barefoot and it's very, very programmable. We can do really nice things with P4. And so its advantage is programmability, its disadvantage is fiber capacity. So what we find is, as we augment, we really build the portfolio, but they're not all the same chips. They really are different strokes for different folks.

So let's talk about enterprise a little bit, just kind of how -- Arista has been making a lot of headway there. My perception, I think its reality, is that, that has always -- mostly started with the data center and then you've become more interested in campus switching as you succeeded in the data center. What's driving your share gain there? You're clearly gaining share. You've been growing very rapidly in enterprise. What's driving that? Who are you displacing? What kind of trends do you see in the next couple of years in enterprise?

Yes. No, that's a very good question because, as you know -- first of all, let me just step back and say, the enterprise is not one kind -- one customer, right? I categorize them in at least 3 types. There's the early adopters, there's the fast followers and then there's risk averse. So it would be hard to do really well with the risk averse for a company like Arista that has bleeding edge, forward-looking innovative technology. We have -- but what's drawing people to Arista is we do have an unbelievable quality. We have an EOS, Extensible Operating System stack that doesn't do what anyone else does, right, meaning it is truly differentiated. It's the power of one OS that's data-driven, has a publish-subscribe state model and can store all of their information and give them a level of availability, automation and analytics. And so what happened was a lot of our customers started -- enterprise customers started using our Arista EOS in the data center. And as they started using it and they were buying campus products or routing products from other vendors. And they said, we want that in our campus and routing too. So we said, well, you've got other vendors, why do you want us. And so one of our customer events, I asked them, "how many of you want us to do campus." And 70% of them said, we don't want you to do campus. We said, why, and they said, because you're doing such a great job for us in the data center, don't change. The next year, they said to us, 70% said, we want you to do campus. And the third year, they told us we're too late, hurry-up and do campus. So there's been massive shift in the frustration that's felt with their incumbents and the desire to use an Arista high-quality, predictive, prescriptive, proactive software experience all now managed with CloudVision that gives them much more workflow and template-based approach that really, this has been a customer-driven exercise. Obviously, we're succeeding more in the data center, but we're also having good success in the campus. And they tend to be more of the early adopters and fast followers than they are the risk-averse today.

Do you feel like you're gaining -- usually, these things reach a critical mass point where adoption just accelerates word-of-mouth and so on, really starts to -- are you there yet, you think or you still...

I think I've very much there in the data center. I still think we've got work to do in the campus. And we've got especially work to do on the campus with the supply chain because we're getting the demand, but we're unable to fulfill all of that.

Do you see campus as a nice to have that follows on the data center, you're happy to provide it. Like you said, a lot of people said they didn't want it. You're happy to provide it, people want it. You have some good products. And it's nice to be able to manage data center and campus products at the same time with the same staff...

Absolutely. It's a little more than nice to have. I think the pandemic has changed it's from a nice to have to an imperative. And the reason I say that is because I mean how many of you are in your offices all the time. You're very untethered, good for you. And how many of you are not, the rest of you, right? So what's happening now in the new pandemic -- post-pandemic world is we have to keep track on the campus of all their attributes, their identities, their address, their location, their mobility. And yet we have to build a network that has the security, the network data lake, the data capabilities, the availability. So we're under pressure to do both where we had the same leaf-spine architecture we built in the data center. Now we have to disaggregate and make sure the leaf can be a wired leaf or wireless leaf, an IoT leaf, an NDR leaf, an observability leaf and untether their location, an IoT device. And so all of that's really changing the way the campus is built. In fact, it's built like a data center, you use the same spine, except you have different types of endpoints now. And then you need the same analytics and telemetry. Take a health care. When I went to my doctors, they didn't know or didn't have access to tools. Today, my appointments are video appointments. Today, my devices and how I connect to my health care is much more mainstream. So even environments that were not dependent on high tech and data now need mission-critical network. And so I would say the campus now is resembling the data center and becoming a center of data and a center of repository for all connectivity. And the mobility of that connectivity is very crucial.

Great. Okay. We're down to 10 minutes. I want to see if anybody in the audience has any questions. So anybody out there have a question for Jayshree. If you can't think of one now and you have one, just raise your hand in a few minutes, and we'll get a mic to you. Oh, there's here one upfront. Let's wait for the mic. Maybe say who you are and what company you're with.

Daniel Fu, Wafra. You mentioned on the supply chain, there's only a few parts that you're still lacking and they're actually smaller price points. There are certain categories that you see common to these kind of small price point things that you're missing? Like I guess...

What's the question -- you mentioned there was only a few set of components. Is there any category to it, is that the question?

Yes. Is there any commonality or category too...

Yes, the one commonality is they all run on older processes, so they are not your 7 to 10 nanometer, but typically 100-nanometer or bigger simpler processes, I would say. So technically, the fab capacity and the availability of that should be easier. But right now, there's a shortage even there. And there are also very simple logic components, like power controllers, ICs, ADD converters, logic devices that are not mainstream connectivity, but then you still need them to run -- to build your entire system. So it's down to 10 to 20 components that are very simple, that are causing us the greatest shortages.

Okay. Any other questions out there? All right. Let's continue with enterprise a little bit. So we have a proprietary indicator for enterprise demand that it points to considerable weakness, things weakening further, enterprise capital spending you've got a good amount of exposure there now, however, a low market share position. So I wonder, a, are you hearing anything from the people you talk to in enterprise that would suggest that they're reducing budgets or thinking about reducing budgets next year? And then, b, how would you expect Arista to perform in that kind of environment if we do get into it?

Before I answer the question, is your indicator for a year or 2 years or how long...

It's been about 4 months in advance. And so we have 2 indicators, real-time spending indicator that started falling off in May and dropped in May and June and then another indicator that predicts that by 3 or 4 months. So usually, it's about 4 to 6 months out from...

It looks like you're an early indicator, maybe I'll be a lagging indicator. Right now, we are not seeing any of that at the moment. And again, to the point you make, we don't claim to be a macro indicator in general. Exactly for the reason you said, we've got a $35 billion TAM. Arista will be $3.8 billion, $3.9 billion company this year or thereabouts. So we've got a lot of headroom to execute and a lot of frustration and fatigue built into the system that wants an alternative. So if these indicators change in the right direction, maybe we'll completely skip it and not see it. But if it gets worse for a prolonged amount of time, I'm sure we will also be a victim of that. But at the moment, things are looking fine. And we're definitely being vigilant and keeping an eye on that.

So you and I have talked privately about networking as a service as a technology trend. And I wonder, there are some elements that Arista is pursuing, some things that you're pursuing that are sort of similar to that, but these new networking as a service companies, all private, that are emerging are pure software layer companies. What do you think about that trend? Is it something Arista would be interested in? Any thoughts on that?

Absolutely. I think the innovation to do networking as a service is very much a software-driven activity. Of course, software has to run on something. And so somebody has to supply the hardware and Arista has certainly built that hardware system business relevance and experience. But if you step back and look at what customers are asking from networking as a service. Whether you run the network on the cloud or they punted to a cloud or they run it on the premise, what they're really asking is the ability to enable things as a service on an as-needed basis, right? That's really it. So forget a moment whether it's sitting on the cloud or premise. What do they need on an as-needed purpose. So first, they need to run some fundamental networking capabilities. So they may move some of their e-commerce workloads to the cloud. They may keep their mission-critical applications on a premise, and they want to make sure the network can toggle between a multi-cloud experience and the premise. So that's one. The second thing they need to know is that they can make the network highly available that you as a CIO criticized for the workload failing on the cloud or the workload failing here. So you have to make sure if anything like that happens, that you have a dual path, you have active-active, you have redundancy. And these are all risk attributes we've been doing for quite some time. Once you build that -- and then you have to have a data-driven foundation where you not only keep track of connectivity, but you may want to prioritize certain applications and workloads and give them the right prescriptive quality of experience. Now once you build that then -- and particularly as you look at this across a multi-cloud, you need some other things. You need Network-as-a-Service to mean automation. How do you automate and how do you do a point and click so that everything isn't a manual configuration. How do you provide streaming telemetry so that you can get visibility for all your workloads on the premise around the cloud. These are all things Arista has been doing, and we would be -- with CloudVision, we would naturally be able to do this network wide and do that today because we have that visibility. So Network-as-a-Service to us is an evolution of our management platform, CloudVision, where we can provide data-driven workloads. We can automate them. We can make them available. We can create the right telemetry experience. And then on top of that, you can add the right security, observability and identity to it. So to me, Network-as-a-Service is no more a box-based approach, but a service-based approach to really provide the service chaining of all of the software on different boxes or on the cloud itself.

Yes. Because what some of these people -- one scenario that I think is interesting is you've got campus or worker locations wherever those might be, and you're tunneling traffic into a cloud service. And you end up with cloud services that are 3-tier applications in cloud services, and they're talking to each other. And all of a sudden, you're AWS, you've got a database, you've got a couple of apps using the database, you can't troubleshoot that network because it's a very weird cloud network none of your tools...

That's a great use case. And in fact, we do that a lot with our Cloud Titans. That's why MACsec, which is encryption and keying at the layer 2 level data center to data center, IPsec, which is the tunnel sec you're bringing up, where you can go across AWS, Azure and connect the VPCs and VNets together with IPsec because no more is the traffic going branch to headquarters, it's going branch to multi-cloud. They're completely skipping the concept of a headquarters, right? So the IPsec becomes important. And then in our own vernacular, we are developing something called VXLANsec. You'll actually see a big launch of Arista introducing this tomorrow. I'll give you a little preview. My guys might be a little nervous in my doing so, but this is a great way to describe Network-as-a-Service because whether you're a mobile edge, you're metro or you're multi-cloud, you should be able to provide any sec and tunnel any of these secs, whether it's IPsec, MACsec or VXLANsec or GRE tunneling and make that completely transparent to the user, and this is a great example of Network-as-a-Service.

So you reach out from the hyperscale data center, but you can also start to maybe provide some services out to the enterprises who want to reach back into those things and some visibility and so on.

Exactly. So you go branch to multi-cloud and you make sure you secure all of those tunnels. And between the clouds, if you're between multiple clouds or between the cloud and the [ front-end ].

So we've got a couple of minutes left. There's 2 things I've never asked you on stage or any one of these things. So I'm going to throw these at you. We talked about...

Ita, be ready to jump in.

Ita, get ready. One is you've run Arista with an uncommon amount of frugality. You have a culture that is extremely frugal and that's different from most people in Silicon Valley. Most of the companies I talked to -- I'll quote somebody I won't name who said to me -- I said, "Well, why don't you make money? Why do you keep spending all of the gross margin you're making on sales and marketing and so on." And this person said to me, well, as long as money is free, I'll just keep spending it. And it turned out, it didn't end well that story. But I'm curious about that subject and whatever you've got to say about it. And I also -- the other thing I've never asked you about is being such a successful woman in Silicon Valley. And your experiences there...

A few more out there, my team...

And a few more here. So comment on any of that you'd like. We've got a couple of minutes here.

Yes, exactly. I think right from the beginning, and it starts with the founders, the CEO, our Chairman, our CFO. We all treat the Arista money like it's our money, and we spend it as a function of the priorities, right? Our commitment to the priority has always been engineering. We have never ever compromised on our R&D budget because the innovation engine is alive and well and Arista customers appreciate us deeply for it. And we've been anywhere from 15% to north of 20% as a functional revenue. But I think the other big piece of Arista, and you call it frugality, I just call it common sense, is achieve profitable growth. If you just achieve growth and spend money, then you really haven't provided all of [ you as ] shareholders, the value, right? And then you really haven't done the right thing in terms of making long-term R&D investments. So we're very much a company built by engineers for engineers, and you'll see us put in that kind of R&D support capability. Specifically speaking on sales and marketing, I think you can -- we're not the company who you'll see doing big brand advertising. But I also think we're selling to intelligent customers who are not looking for brand marketing, but are looking for very tailored customer support, systems engineering, network design. So they want our sales team to be consultants. In fact, my best sales leaders are the ones who I can't tell if they work for me or work for the customer, right? So that kind of prioritized focus on verticals and tailored sales and marketing doesn't require the broad brush that maybe if we went into the mid-market or SMB market or consumer market you'd require. So I think by being tailored by being prioritized and then also by philosophically prioritizing growth and profitability, we've grown to be the company we are. It's not going to be the same in every market, some of the new markets, we will invest more, right, as they require the early incubation and investment. So it's not the same thing for every market, but I think philosophically, we're very much there, like you described.

It's come into fashion now.

Right. And as far as being women in this industry, it's great to see more of them and for all the guys out there. I love the guys too. We work with both of them. Having 2 daughters, I can just say that encouraging women in tech starts from a young age, like middle school, definitely encourage them to be exposed to as much math and science. And they may reject it, mine certainly did initially. But then the biggest thing that my parents did was continue to instill the confidence in pursuing the sciences and pursuing tech. And in my case, I was particularly bad at non-tech. So it was a little easier decision.

Great. All right. Well, great, Jayshree. We're out of time. So thank you very much. Good to have you.

Thank you. Thank you, Rod. Good day.