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

Perfect. Thank you very much, Zack. Great to have everybody joining us today. We have Mark Foss, who is the Head of -- Senior Vice President of Global Operations and Marketing at Arista as well as Liz Stine, who's from IR. We're going to do a fireside chat format. And if you want to ask questions, there is a box on the system that you can ask them to. We'd love to have as many questions coming from the audience instead of me providing them makes for -- much more interactive and fun opportunity. But with that, welcome, Mark. Welcome, Liz.

Thank you.

Thanks for having us.

So Mark, just for people who might not know you, why don't you give us a little background on yourself because I think there are some people who only know a couple of people over there and you're one of the more normal ones.

Yes. No. Well, I may not be very well known, but I am fairly normal. I run our global sales operations. I also run the Corporate Communications. I've been at Arista since we were 28 people back in '07. So I've kind of been here the whole time, ran international sales for a long time. And I've been in networking for the past almost quarter century, I've spent like 11 years at Cisco before that. So yes, I've been around a while.

Well, we won't hold the stint at Cisco against you at all otherwise we would be holding something against most of the people at the company at one point or another. So for purposes of a starting point, can you talk about the recent blowout quarter you guys just announced. That's -- that was a pretty amazing event. Go over the report results, the guidance with particular commentary around the expectations for CY '22 and longer-term guidance.

Sure. Yes. No, I mean we've had pretty good demand. Demand has been pretty strong across all of our customer segments. And I think that the fundamental concern as with everybody around the world is global supply chain. So for us, that's kind of been our primary concern. But we worked really hard early on to get our supply chain. We've been spending since summer of 2020 to kind of secure parts. So we felt secure enough about our supply chain situation to actually be able to beat the number and raise guidance into Q4. And I guess one of the benefits of having a long lead time situation is that customers give you more visibility into orders and what they're going to be doing longer term. So that was, I guess, one of the main drivers to give us the confidence to guide 30% growth for 2022. So that was -- that's -- that was kind of what the big news was on our Analyst Day.

So just to go back into the detail there. So you beat by a substantial amount on the quarter. You also provided guidance for the full year '21, I believe 25%. And then as you just noted, 30% growth in CY '22. Now how much of the growth in CY '22 is a function of people ordering, how to say this, people ordering with a larger duration to the time that they're anticipating that coming in. How much of it is driven by sharp acceleration in traffic volumes in the hotter network? And how much of it is driven by, Oh, my God, am I not going to be able to get the product, so we better get out there into the queue earlier?

Yes. I mean it's really driven by demand fundamentally. And I think customers are placing orders now for 2022, but they're saying like, "Hey, do not ship this part until Q1, do not ship this part until Q2, do not ship this part into Q3." So they're being prudent on how they plan for that. So it gives us visibility into the orders. But I don't think we're seeing customers stock or try to hoard product. They're legitimately trying to get -- they're telling us reasons why there's some urgent use cases to get the product in-house. So we haven't seen any customers hoarding product. They are planning to get product on time, but we're not seeing the product go unused, I should say that.

So is this demand being predominantly driven by enterprise, cloud service providers? It seems like it's probably heavily skewed to cloud given the scale of the some of the companies and particularly Facebook, increasing CapEx by $10 billion kind of caught everybody by surprise.

I'd say it's all of the above. I think in the most recent years, we've had more momentum in enterprise and financials that we have in cloud. But now I think there's potential of us coming into a new cycle where we're going to see some increased spending in cloud. So I guess there's a number of scenarios within 2022 on the mix of enterprise and financial and cloud that we may see to hit that 30% growth. We also said at Analyst Day that we're going to be doubling our campus revenue from a goal of going from $200 million in 2021 to $400 million in 2022. So that's also a piece of the growth as well.

Sure. So just going back to the equation there a little bit. If I were to look at that sharp increase in next year, is that a function of people not being able to get parts to some extent this year? I mean to be clear, CapEx for most companies, whether it'd be cloud companies or whether it'd be service providers, have been way below forecast in the third quarter. And they're forecasting big increases in the fourth quarter, but my guess is they're not going to be able to get enough product there as well. So is that part of the equation that you guys are gaining share because you can deliver?

I don't think we're necessarily gaining share because we can deliver. I mean there's some -- in tactical situations where someone has -- someone will have a lead time, which is a month less and they may have a more of propensity to go there. But on a strategic level, I don't think that's the case. I think that it's a combination of -- I think we are seeing increased demand. But at the same time, I think we're also gaining share. I mean, we've -- we said at Analyst Day that we're only about 25% penetrated in the Global 2000 and Fortune 500. So we have a lot of room to grow there. And then also we see that the cloud CapEx numbers are also going into new cycle and also increasing. So there's a lot of factors happening.

I'm a little puzzled by the cloud market because we had the Cisco say on their last Analyst Day, which they report on Wednesday, so I'll be interested what they say on Wednesday when they report. But on their last earnings call, they said that they had 160% growth in cloud, reaching 30% of their service provider business, which is -- that's a pretty big increase of 60%. And so -- then if you come into your numbers and you're seeing really strong demand in cloud. So it's a little puzzling that it's that strong. Is it a function of the 400 gig cycle driving upgrades? Is it a function of very significant increases in traffic growth because of the dispersion of the workforce, the use of technologies like this that really drive up traffic volumes? Is it a function of -- just something -- any other variable that you want to pick out? Why is that happening?

Yes. No, I think you hit on most of them. I think the long-term trend is workloads are moving into the cloud, right? With -- and that's still continuing. I think it's still expected to continue for the next few years before it hits a certain equilibrium. I don't think all workloads will ever go to the cloud because there's security issues, many of these financials want to have -- financials or other customers want to have their own data centers to be able to have complete control over their data. But I think it's a combination of a lot of factors. Yes, the 400 gig coming into effect. There's -- we also did do a small price increase to really keep the component cost neutral, which could also contribute slightly to it, but that was more of a -- just to keep our margins neutral because we had all of our...

We'll come back to pricing. [ We'll do it ] as a separate discussion.

Yes. Yes. But yes, it's just in general, it's just -- I think we're hitting a new cycle in terms of an upward cycle in terms of cloud demand and spend.

Yes. I wanted to go back to the guidance that you guys threw up. I think there was a slide there that had the '21 outlook, the '22 outlook. And then there was a third column, which was long-term growth rate of 15%. So when I look closely at that slide, it said 20% to 25% growth rate of 15%. But your first 2 years at 25% and 30%, that would imply a much slower growth rate beyond three years. Am I reading that slide correctly that you expect to slow down to, say, 10% or sub-10% in '23, '24 and '25 on average? Or is it that you just don't know? Or what -- how do I read that slide? Or alternatively, is it that you expect to continue to grow at 15% in '23, '24, '25, which is I think what most people saw when they read that.

I think we're just being prudent. I mean, when you start getting out '23, '24, 25 visibility gets less certain, if you will, and who knows you could have a macro trend hit. You can have the -- a cloud cycle, clouds can go in a downward cycle. We just don't know. So I think it can be a function of just visibility out there and just having prudence as a company.

Okay. So -- but it does imply a slowdown in the out years. And is that -- do you think that's kind of a reversion to the mean kind of mechanic?

Yes, it's -- yes, it's kind of our best estimate of a CAGR to get us to be a $5 billion company and insist what our projection was. And I guess we'll see how the latter years play out.

Let's talk about the enterprise segment for a second. Obviously, you've had really good success in data center and enterprise. And for good reason, the data center products that you offer are so smooth, easy to operate with all those cloud elements. I mean if I think about a, my vision of an enterprise data center, there's lots of people running around, and when I visualize a cloud data center, it's a 1.5 miles long building with nobody in it. And the enterprises want that efficiency. It's that simple, right, at the end of the day. So can you talk about how much better you are in that context at delivering the automation and basically reducing the labor cost of running a data center from an enterprise environment?

Yes. No, I think you're absolutely right. I think the difference between a cloud and enterprise is, the cloud, they have a DevOps team, which has basically built and innovated a lot of these automation features that allows the data centers to self-drive themselves. On the enterprise side, they don't have these resources. So what Arista did is we developed a turnkey solution called CloudVision which where we have taken many of the principles that we've seen in the cloud and put them in a turnkey solution for automation, provisioning, AI, and put these into the data center and now the enterprise -- these enterprise customers can now automate their data center just like a cloud can. And it's been a big game changer for them, to be quite honest. It's been really well received.

So with that experience, I would assume that part of the reason why you're moving to the campus because they're begging you to go there.

Absolutely.

And deliver that same intelligence, that same ease of operation, that same seamless low-touch management into the campus. It's my sense that when I look at the campus that there really hasn't been much automation and improvement in that. In fact, even if I look at the wireless edge piece the Cisco offers with the Meraki, it's really kind of a single-tenant legacy architecture. It's not a true cloud-native, multi-tenant capability built off of micro services and the like. So can you talk about how fertile of ground that is for penetration with those automation tools?

Yes. No, it's actually very, very much so. I think Anshul, our Chief Operating Officer, he always makes the joke that in the campus, their main form of automation is copy and paste. So that's kind of their biggest form of automation. And I think you've had an incumbent vendor that's been in there for 20, 25 years. And I think there's a lot of -- and you're oftentimes, they're using very old software, they live in fear of software upgrades, when you upgrade the legacy software architecture, you can break something else when you upgrade your software. And then -- and then with Arista, our latest version is the best, which is the way it should be. You shouldn't be living in fear of upgrading to the latest version of software. But CloudVision, again, is the same engine that's been very good for us in the data center. We simply just leverage that for campus. And it allows you to give visibility on your WiFi access points. It lets you automate and provision within the campus, and it's been a game changer for these campus customers. They've really had some incumbent vendor fatigue. They've been really looking for a viable second alternative, and now it's here.

So when I look at the $200 million mark that you hit, how much of that is coming from customers that are existing data center customers? And how much of that's actually new customers that you've managed to penetrate into the campus?

Probably roughly half and half, to be quite honest, it changes quarter-to-quarter. So it could be different in any given quarter. I thought it was going to be a lot higher for the initial -- for the existing customers initially, but we've been pleasantly surprised. We've actually been getting a lot of new logos from new campus customers that we had never even sold, we never even talked to for the data center, we never even had data centers. So that's been a good sign. So there's been a lot of momentum in that market, and that's one of the reasons why we're -- we feel good about giving another goal to double that market again for us.

So do you have experience of companies coming in from campus and then saying, "Hey, maybe we should be using you in the data center?

That, I'm sure it's happened. I'm sure it's happened. But there's -- I'd say that there's -- there's probably only -- there's probably less than 20,000 customers on the planet, which have data centers, but there's like almost 0.5 million customers on the planet that have campuses. There's a lot more campuses than our data centers out there.

Yes. As we move to a cloud direct world, the Zscaler type architecture, where you're going out into the cloud and providing security in the cloud, does that amplify your competitive position and improve the opportunity within campus?

Yes. I think security is really top of mind within campus for sure. And I think that was one of the driving forces behind our acquisition of Awake Security, which is...

No, I'm talking about security for you per se. I'm talking about they're changing in the architecture, so they're going to a cloud direct model.

Okay. So repeat the question then.

Yes. So the argument that Zscaler makes is that you shouldn't be going back into the enterprise data center to get it to the cloud. You should go direct out to the cloud and the securities we provided and they'll connect it to the applications. The question is, if we're moving towards a cloud direct model, does that change the demand or your penetration of that campus footprint? I would think it would actually amplify it.

Yes. It's -- we're seeing all types of things in the enterprise campus. There are situations where the campus is outsourced by a managed service provider, and that definitely happens, and you could have security implications there. But it's such a wide market that there's -- customers have many different areas of interest in architecture and how they do things. But there is definitely on the smaller -- on the mid-market side, there is definitely a trend to go into the managed service provider arena for campus for sure.

Well, we have a question that came in from the audience. So great opportunity to bring that forward. The question is, how are you able to grow while maintaining such high margins? I look at your competitors and your margins are very high. How are you doing that? How is this changing as you increase your footprint in campus?

Yes. I think we've spent -- I think in general, if you look at our operating margins, we spend a little bit more than the industry average on R&D. I think we're close to 20% there. And then we've spent less than the company or than the industry average on sales and marketing, I think we're around 8% of sales, I think, are on sales and marketing. And that's primarily been because we're focused on our core market, which is the data center market, and you have, you can have just a small number of salespeople and you're driving millions and millions, tens of millions of dollars in terms of data center spend from a very small spend on sales and marketing. I think as we move forward, we are going to be further investing in the channel. And I think that also continues to be a pretty good margin for us as well. I think the campus in general, because you don't have a cloud titan in there, which is generally the cloud titans get the best pricing on earth because they have buying power. So they will -- if they're strong in a given quarter, they can have more impact on the margins. But there's no real cloud titan for the campus, if you will. So we don't really have that effect for the campus. So as the campus continues to become a larger part of our business, I think that will just help margins. But we're also moving towards more software and software services. Our CloudVision is continuing to increase. We had a few acquisitions of Big Switch networks, Awake Security. I guess the whole software as a service model is also helping to contribute to margins as well.

One of the interesting elements of that point, and I think it's a really important piece of it is it doesn't look like your sales force is increasing hugely because the percentage going to sales and marketing isn't increasing dramatically. But that's actually not true because what you're capturing there is the mix. The sales force going to that enterprise piece has actually got a fairly high rate of growth in it. But because of the cloud piece doesn't have a whole lot of almost no real increase. I mean like, how many new cloud companies are you going to target. The spend is barely focused on a more narrow piece. And so it looks like it's a smaller piece. So what is the growth rate in the sales capacity targeting that portion of the market that is enterprise data center, enterprise campus?

Yes. We're -- I mean, our target is to really keep up with or hire faster than the company average, right? We just want to maintain that. So we have some pretty aggressive hiring goals, but we want to keep up with where the company is growing in terms of growth rate, but it's challenging. We don't want to hire any salesperson. We want to hire the best and it's been hard for us to grow fast -- hire fast within the company, let's put it that way.

Well, risk has always been a destination for people who are in this space. I would assume and I'm going to digress a little bit on the subject of attrition, wage hikes and the like. I would assume that you guys are seeing a fairly low attrition rate still despite the resignation era that we're out. Is that an accurate statement?

I honestly haven't looked at how we performed versus the -- I haven't looked at the latest stats on how we performed on attrition versus the industry average. But I think it's been fairly low. We obviously have attrition, we have turnover and we have performance issues you have to address. But I guess we also have a stock compensation for people. So when the stock goes up, generally, people want to stay around and have their shares vest. So that's always good for retention of good people.

We certainly have seen that happen recently.

Yes.

Yes. So the attrition rate is low, what about the wage inflation? Are you seeing wage inflation, particularly, I guess you've got a pretty large campus in California.

Yes. No, it's definitely been more difficult to -- especially when we're headquartered in Silicon Valley, you're competing with really the cloud titans for engineers, to be quite honest. So we've actually taken the approach to move our engineering to a more geographic model so that we can take advantage of talent in less expensive areas. So we have, I think, our second largest engineering is in India. Then we also have engineering in Canada, engineering in East Coast, U.S., Ireland. So we've really taken kind of a global approach now to engineering. So it's kind of more of a 24/7 software development, which has actually been great.

So when I think about the dispersion of employees, that also creates a big opportunity for you to expand into people who may not have been near one of your offices. Have you seen that also happening?

In terms of what, sorry?

Hiring people who are not local to an office?

Yes, yes, yes. I mean, generally, for engineering, if it's hardware engineering, you need to be near to office. Software engineering, you don't really have that limitation. So if there's good people, we have hired people that are not necessarily near a physical office. But I think the majority of our people actually are in the vicinity of these offices, but we have many engineering offices globally now, so.

So let's go back to the product lines for a second. I didn't mean to digress too much on that subject. But it seems pretty clear to me that you've made some pretty interesting announcements around capturing telemetry information, building out a data lake, getting into the security arena. Can you talk about the scale and scope of that strategy and how we should be thinking about it?

Yes, sure. I think you got to walk through kind of where we started first. Arista started with our software architecture, that's kind of fundamentally who we are as a company. In that software architecture, what we did was we separated state from the process. So we had a centralized database within the software, which maintain the state, which allowed for very, very high quality. I'd say probably in the mid-2014, 2015, we expanded that to be a network database. So then the state of the network exists on multiple servers, and then you have that really same architecture, but expanding it from a software -- just from the software architecture to the entire network. Now with the network data lake, we've taken that to another level to where you can now access data from other vendors. And if you're getting -- as you're getting into machine learning and AI, the more data you have, the smarter you can be. So having access to the data lakes of these other ecosystem vendors just makes your network that much smarter. And that's really the fundamental premise behind the network data lake.

So how does the -- these new capabilities expand your opportunity within your existing customer base? If I have a customer that's been a Arista customer in the enterprise for your core products, how much does this increment up that potential revenues to that particular customer?

Yes. It's still early days. I think it depends on who the ecosystem customers are -- that they have in there. If they have a VMware or Palo Alto or one of the other ones that we've listed on our ecosystem vendors. If they have these multiple vendors on there, over time, I think you'll see having more access to the data and features from all the companies that would enable you to make more use of the multiple data lakes in one customer. So it really depends on what ecosystem they have within their environment and what features end up coming on. But it's still early days. This is still early days.

I have another question came in from the audience. So Mark, you mentioned increase in software revenue is a way to boost margins. Is the NetDL announcement tied to this? How is the company positioning this? Are you going to monetize this or this more of a differentiation in the hardware platform, and any plan to incorporate it into wireless LAN?

No, it's still -- like I said, it's still early days. I think the existing software licenses are there to monetize it. There could be future software licenses that we do as products develop over time, but it will be monetized via the software, and it's just going to take time. It's -- we're -- still early days here.

Okay. So given the data lake element, does that change your cost structure at all? And how do I think about that build out as a driver of your business in terms of the infrastructure build that you might need to support it?

I don't think there's that much of a difference from the -- I mean, we are building a new data center in Santa Clara that's going to be used to expand our software testing. So right now, we have about 8 megawatts of data center space in Santa Clara, where all it does is it just runs 24/7, testing our software. We do, I think, 1.5 million test runs in a 24-hour period. So we basically have fully automated our testing. And this new data center that we're building is going to further that capacity as the software and the test cases get larger. So we're planning ahead in essence what we're doing.

So does your ability to build those data centers and then basically run tests on and to provide a natural infrastructure for carrying the data lake?

I don't -- that's a very good question. Yes, it's -- yes, that's -- you're at little out of my realm there, to be quite honest.

Okay. No problem. We'll go back to your realm. So I did want to talk about the wireless LAN side of things. The boys over at Juniper and Extreme, argue that their AI wireless edge gives them a real because it's cloud native and distributed it gives them a real advantage. You also are very much driving that same value proposition. How do the offerings compare or does it not matter because it's really only going against Cisco and with the Meraki because they're the dominant player?

Yes. No, I mean I think, yes, I think the 2 biggest players in the wireless space are Aruba and Cisco. But yes, no, machine learning, AI is very, very big for us. CloudVision, obviously, provides a lot of automation, and we have the access points can be managed under the CloudVision. And also the fact that we have -- we leverage the cloud for our WiFi infrastructure. So the services for our WiFi are all managed in the cloud. And you just have the access to all that compute power and you know anything is possible. And then we've also added -- on the security side, we've added Awake and the AVA, which is the Automated Virtual Assistant to also provide AI for network detection and response. So it's been -- it's actually -- the automation is just -- we're just scratching the surface for automation and machine learning. I think this is the trend over the next decade, this is going to further increase a lot.

But the 2 main competitors that you just cited, neither of them have seemed to move meaningfully in that direction. As I understand it, looking at just Cisco's security business as an example, less than 25% of their security platform is cloud native, which tells you how far behind they are on it. And Meraki, clearly is -- they're still their primary offering there, it's not, not either. So should we view this as such a big a Achilles' heel and that's huge...

Yes. I mean the market appears to be moving that way. The market appears to be moving towards cloud managed, and the market appears to be moving towards AI and machine learning. So that appears to be where the market is moving. So yes, you can make your own.

Yes. I wanted to go back to the pricing conversation a little bit. So when people hear that pricing is going up 10%, they assume that. Well, if the revenue is up 30%, then it's actually up 20% in volume. That's not the case. That's just not how it works. How much of the current business that you're looking at over the next 2, 3 quarters is already priced in at the prior rate? And how much of the increase in order of visibility is grandfathered preprice hike?

Yes. we didn't change any of our backlog, and we just did have the prices go into effect on November 1. So with lead times the way they are, it's going to be mid-2022 until you really see anything. And again, it's 10% on the hardware only so that we didn't increase prices on software, didn't increase places on service, didn't increase prices on optics, things like that. So it's definitely not going to be across the board. But it will probably be mid-2022 until you start seeing that fold in. But we did these price changes with the intention of being neutral, right? We saw our component vendors increase their prices, and we felt that this was the correct number to adjust to be margin neutral long term. And we'll see how that plays out, but that is -- that was the intention.

So going back to the competitive landscape a little bit. Cisco acquired Acacia, 400-gig ZR integration into their platform. It seems like they're trying to change the topology of their customers to a mesh routing platform architecture. Does that give them any edge versus you guys? Or most of your customers are capable of buying the optics and populate themselves and therefore, it release the relevant when they tie in software into it. How do we think about the implications of those moves?

I think it's the latter, to be quite honest. I think customers -- we haven't seen, at least the cloud customers bundle any purchases with optics. They generally buy them separately. The customers -- these cloud customers don't even buy the optics from us. So from us, there's been very little change on anything, but there hasn't been any bundling that we've seen.

Does ZR impact your business in any way?

Well, ZR for us was, it was kind of a prerequisite to have 400-gig ZR shipping so that the cloud customers could have a 400-gig data center interconnect link before they built out the spine. So that was really the issue around ZR and ZR is now shipping. So now it kind of paves the way to have customers, at least cloud customers, to start allocating some projects to 400-gig within the data center.

Right. So how do you think this translates to some of the older form factors? 100-gig, 200-gig, do those sustain, start to flatten out and almost all the growth is coming from 400-gig or are they continue to grow for a while? How do we think about the mechanics around the segmentation there?

I definitely think 100-gig will continue to grow for a few years. I think that some of the analysts -- the industry analysts are saying it's going to continue to grow through 2023, 2024. So 100-gig will continue to grow. The growth will slow down. 400-gig will probably grow the fastest. But it's not going to cross over with 100-gig to probably 2024 or -- 2020 -- yes, probably about 2024. 200-gig is probably more of a single -- it's probably more of a -- it will be volume, but it's going to be just a concentration of a few customers. I don't think 200-gig is widely used among a lot of customers like 400-gig is.

It's a half cycle.

Yes.

Within the context, as we look at the cloud customers, was demand diminished because they were waiting for the 400-gig stuff to mature? Because my sense is that they did not change their purchasing patterns at all relative to that cycle. But your comment a moment ago suggested there might have been some holding back of orders until that shift. Is that -- how do I think about that?

No, I don't think there was any holding back. I mean these guys will -- especially in this in today's environment, these guys will take what they can get. But no, I don't think so. I think like I said, it was a prerequisite to get to ZR 400-gig before they would even consider doing a project for the 400-gig. But this is just another upgrade or is just another -- we've had multiple product transitions within the 100-gig segment. And the way I see this is this is just another transition between segments. So you're moving some of the business from 100-gig to 400-gig, but it's the same -- we've seen these micro transitions between 100-gig for the past few years. So we don't see it being much different than that.

So there's a lot of people on my side of the street that basically argue that 400-gig is the opportunity for Cisco to get back into the market because they're going to beat up on this -- the space because they missed the 100-gig cycle, and you guys got a big footprint, but now they're going to come back in, implying that speeds and feeds matter. Where it strikes me that just missing the point all together that it's not about speeds and feeds at all. It's all about software, and the software really hasn't changed and that makes customers very sticky. Can you address that point?

Yes. No, I think you're absolutely right. I think it's the same customers that are buying 100-gig in volume are the same customers that are going to be buying 400-gig in volume short term. And you're absolutely right. These customers are buying the products based on the software and the software architecture. And that's -- if you look at qualifying a vendor or a product, over 90% of the qualification work goes into the software. The hardware is like less than 10% of it. So yes, it's going to be really -- it's just another transition that we've seen. We've seen all these micro transitions within 100-gig, you're going to see this other transition to go to 400-gig, same customers.

Yes. So speeds and feeds are still 2005.

It is.

I wanted to address another area before we run out of time, and we only got 2 minutes left. So I'm going to ask you to do it quickly. The white label story has always been there. It's been there since your IPO. Can you talk about what you're seeing in white label? And how you think of it impacts, possibly or negatively the trajectory going forward?

There's a keyword that I always use to describe white label and that's it's a contained market. There's only about 5 customers on earth that are -- that comprise about 95% of the volume there. And these are customers that have been doing this before Arista was even a company. So it's a contained market. It's growing but it's growing because you're seeing the growth rate of these companies that have always been on white box. So we haven't seen other customers come into the foray that want to go into volume on white box stuff. In order to make it economical, you got to have probably 1 million server installed base to make this to be an economical move. And how many customers on earth have 1 million servers installed base, almost nil.

Not very many.

Very few.

So again, going back to the heart of that question, is it improving or getting worse? I mean is there any change in the [indiscernible] early question.

The pendulum kind of swings within those customers every few years. They're thinking, okay, maybe I'm going to go further on white box. Some are saying that they're going to be potentially looking at moving to vendor for some projects. I think the pendulum could be moving a little bit towards them considering vendor for some projects, simply because the platform engineering cost of qualifying all of these different chips that is coming out has basically been very difficult for them to keep up. And for that reason, they're considering looking at vendors for other projects, but...

And the operator is going to get mad at me if I don't end this thing because we're running out of time. We've run out of time. I think, Mark, thank you so much, Liz, thanks for joining us and to all of the people who called in or dialed in, zoomed in. We appreciate the audience participation as well, and thanks for the questions.

Thank you.

With that, it's a wrap guys.

All right. Bye-bye.