Well, thank you, everybody, for joining us. My name is Tom Blakey. I'm the infrastructure, technology, and software analyst here at KeyBanc. We're grateful to have John McCool here from Arista. He's the EVP of Engineering and Operations and also the Chief Platform Officer. We have Liz Stine from Investor Relations here. If you want to bother her, you're in the front row. You know, thank you very much for joining us, John and Liz. I just want to start off, because I think you know, from Chief Platform Officer, maybe just a minute, what does that entail?
I'm going to have to stand, Tom, because you're standing, buddy.
At Arista, what are your responsibilities?
Chief Platform Officer, I have really two domains. One is hardware development, so all our new products, and then manufacturing and supply chain. Effectively, I get to see products from inception and concept and then all the way through production and build-out.
Excellent. just want to jump in. We're asking all of our co-companies at the conference here to try to get a, some sort of semblance of the macro. We know how things ended at the end of last year, macro softness and things kind of, spending has gotten a little softer. We're speaking specifically, I guess, to enterprise for you, not related to your cloud business, but as we, go into the second half here, just generally speaking, what is, you know, what is the, the business kind of update in terms of, you know, from, from the beginning of this year to the second half of this year?
Sure. you know, in terms of Arista, I'm not sure we would be typically a bellwether in enterprise. We got a lot of share that, that we have to gain and an opportunity I think that's a little independent from, from the macro. We did talk about in our last call, the success in, in the enterprise. That's momentum that we've built out through the last couple of years in 2 dimensions. one of is, is new use cases and new products, coming out from just data center, moving into routing and campus. Then we also talked about, you know, cloud was big news for us. Early in the supply chain challenges, I think cloud recognized that there would be impact, and really prepared for the big build-out.
It took enterprise a little bit longer, and now you're starting to see enterprise catch up in terms of our, our shipments and deliveries.
It's a good segue to my next point, point of questions here about enterprise. You know, Arista has been very successful. The high-performance data center has meaningful share there with some, you know, Gartner data, for example, has 45% market share at the high end. You mentioned the new products that you've launched in the enterprise. Now, companies have this ability to standardize on Arista. You, you had experience at Cisco. What, what are the dynamics that, you know, are, you know, kind of different throughout the rest of the stack that maybe Arista could, you know, the ups and downs, the risks and the benefits of what, how Arista is going to try to emulate this success in the data center and throughout the rest of the enterprise?
Enterprise, the whole, the whole picture.
Yep.
Yeah, so, so for folks not familiar with Arista, you know, we started out with a data center play. We actually started out in the financial vertical with a low latency switch. We, we got a strong footing in certain verticals, financial and media entertainment, around the data center solution. We started to build out new technologies or use cases around the solution, specifically routing. We went from, you know, connecting your servers and building out a distribution network to being able to interconnect site to site and redundancy. We made an acquisition on Wi-Fi and built organically our own campus switching products, which would connect end users into the network. With Awake, we started to get into network security and detection.
Mm-hmm.
The last piece that we've built out, I think is pretty important, is WAN transit, so being able to connect remote branches back into the network. We feel pretty good today that we have all the, you know, essential building blocks for building an enterprise network end to end. If you're, you know, looking at an alternate to your incumbent, you really want to see somebody that can challenge across all those use cases, and I think we're, we're in a good position today.
Is there are there any, additionally, along the lines of you finally having this end-to-end solution, technologically? Walk us through, like, not so much the go-to-market, but this is a market where you're taking share, taking dollars away from something that's already installed. It's not like a software company that's starting a new TAM, if you will.
Right.
What, what are the technological advantages of Arista to kind of, you know, basically rip, rip and replace some of the infrastructure that's in place, with the, you know, the incumbent?
Yeah, I think our, our approach is very different. We get this question a lot, and people want to point to a differentiation in all those roles. Like, okay, your campus solution, how's that better? Your WAN transit solution. We take the approach that we build a better enterprise network. Our operating system fundamentals are stronger from a software approach, which leads to better quality. Our sales teams will go out and say to operators, "You know, don't you want to spend your night and weekends with your kids and your partners?" Because they spend a lot of time just reacting to moves and changes or upgrades to the software that lead to instability. All they care about is just running the network and making sure it doesn't go down.
There's a lot of issues with vulnerabilities in, like, the Linux stack that forces upgrades. Upgrades are ripe for, missteps, either manual configuration errors, or you get a software regression. In other words, something that used to work in your network was broken through that upgrade, and it, it was not uncommon in networking before Arista. We're selling, you know, peace of mind, ability to sleep, take your weekends off, and just run a better network. The campus comes along with that or the routing piece. From an investment in your people, you just have to learn Arista EOS and CloudVision, and then you can deploy in all these use cases rather than learning a different management stack for each piece.
It's more the cloud management?
Yeah, it's around the, kind of, the integration of the whole stack across all those use cases.
That's all integrated today.
Yes.
You have all pieces, but everything's integrated.
Everything's integrated today.
Basically.
Right.
You, you mentioned part of the, the Chief Platform Officer is responsible for supply chain. The campus piece that you've been breaking out, I think, since 2021, maybe, had some supply chain issues in 2022. Can you maybe just update us there?
Yeah. It, it wasn't unique to enterprise. I mean, it was kind of a generic, all the supply chain challenges, but campus being the new piece, in terms of just kind of hitting some of the revenue targets, the interest was there, but our ability to deliver was impacted by some of the supply chain shortages, and you're starting to see some of that catch up now.
From that hierarchical view of back- building backlog we've seen, and we'll get to it in terms of cloud titans in a, maybe downtick in one of your large M&M customers, you took up guidance for 2023 in terms of 30% growth, implying through the modeling, that enterprise would, would, would, depending on where you were in terms of your models, inflect, into 2023, in the second half, maybe. Is, is some of that backlog related, just satisfying that backlog, or is there some other element to the, the, the strength or even increased strength in enterprise there in 2023?
Combination. I mean, we've kind of built up that strength going forward. We've seen larger deals, both in, in terms of customers that were already Arista customers, just building out more, as well as acquisition of new customers.
Let me just pause there for a second. Is there any questions in the audience before we switch over to cloud?
What you're saying is the impact of the supply chain issues that caused tough comparison is kind of waning a little bit. You caught up a lot of what?
We're starting to see availability of components. While the lead times of components are still very long, the predictability of delivery has been there the last 2 quarters. When we order something, it comes in when it's supposed to. We're not seeing pushouts in terms of. The lead times on semiconductors are still very long, I mean, close to a year. The capacity's improved, predictability's improved, but we're still probably seeing 2x the lead time at the component level than we saw pre-COVID.
Switching to cloud titans and that, maybe just exploring, not necessarily, specific, obviously, numbers in terms of the slowdown that was talked about on the one most recent call for one of the one of your large cloud titan customers. That was really important for growth in 2022, it was driving growth. Walk us through, like, the, whether it be wallet share dynamics, like, why are they slowing? I'm trying to get to the will this spend come back? I'd like to just maybe even before we get to that answer, what drove, because I think it would help answer that question, what drove that big uptick in spend at this customer in 2020?
Sure. I mean, if, if we look back in 2019 at our business, we actually talked about customer waiting for next generation server upgrades and kind of putting a pause. Going into 2020 with COVID, there, there was a little bit of reaction to that pause, as well as increased demand of these networks that really, you know, turned on the, the cloud business for us. There's some smoothing effect there, and, you know, we saw tremendous growth in 2022 in, in the cloud business. I think we're kind of getting back to a more, probably normal cadence.
Again, and again, just using that as a backdrop, going forward into 23 and beyond, trying to understand, you know, that sounds a little bit more cyclical. I'm trying to ferret out, is there a was there a technological need at that particular customer that was looking to shift more dollars to, say, an Arista as they're building out, whatever they're building out?
Yeah, I mean, we have good relations with, with all these cloud customers that have been multigenerational. There, there's a lot of integration and engineering, engineering work that we do with the cloud titans. Kind of looking forward just in the cloud space, I think there was kind of some decision making around what, what's going to happen with, with AI as a next generation use case, that people are starting to understand and digest when they start to look at 2024 and 2025 and beyond.
Just trying to wrap it, conclude that they're digesting, some things that they built prior and then shifting focus to a different area.
Trying to understand that area. It's pretty emerging, right? I think a lot of focus is really how can AI help my business? How will the applications integrate? What's the value proposition around AI to my end customers? That's kind of going on at the top of the stack or maybe even at the business level. You have teams working on infrastructure, thinking about, okay, if this really comes to fruition, how many GPUs do I need? How do I deploy them? How do I deploy it at networking scale? Thinking about how that can be more consistent to what they've typically done in the data center. You know, I would say in the last, you know, six months, this dynamic and discussion around how AI evolves has really taken front and center.
you know.
Can you talk about how the cloud, that AI in addition to what's going on in the cloud, just in general, about how that's affecting?
... the demand for higher speed.
Yes.
We have 400 now, gigs and 800, whatever we had in that.
Right.
How they go, what the read about?
Just to kind of repeat the question for folks that are on the call, how does AI affect the speeds and the speed transitions, and maybe how does it map in, if you will? AI is definitely, you know, a bandwidth consumer and really wants to see low latency. You're spending a lot of money on GPUs and the endpoints, and to sit, sit there idle, like a factory sitting idle, you don't want to waste that. There definitely is a propensity to the higher speed grades and a strong interest in 800G to move forward. People are using 400G today in, in trial activity.
There's also an interest in making the network better for AI and how you might enhance Ethernet moving forward to either get better load balancing or better visibility into what's happening through the AI piece. There's not a binary, it has to be 400G-800G, but it's definitely, you know, the people who are doing AI today are definitely wanting to use the highest performance available and definitely pushing towards 800G.
Is the industry ready? For the links and everything.
Yeah. The question is: Is the industry ready on, on speed of the links? I think people would like 800 gigs sooner, but, you know, 800 gigs in development today, and that's going to, you know, probably be the primary use case and driver to 800 gig.
What's the critical path? What, what's the critical area?
It's just work. I mean, it, it, it's the whole ecosystem. It's the semiconductor chips, it's the optics. I think a lot of that work was done for 400G. There were some new breakthroughs in coding schemes, et cetera. I, I think that'll follow its natural path, but it's just not, it's just not there today.
Vishant?
It's, I read about that with the Ethernet Consortium, that you guys are, it seems like it was kind of a reactor to, NVIDIA's role on the whole AI thing. Welcome to some other stuff together. They, before they run too far ahead. I'm wondering what you think about it?
Maybe I'll back up. Is it okay if I talk about InfiniBand and Ethernet? I think it's on your list, so it sounds like a good segue into the whole...
The whole unified Ethernet.
Yeah. So maybe I'll start with just some perspective on Ethernet versus InfiniBand. InfiniBand was also developed by a consortium, and the use case was high-performance computing. InfiniBand has some really good things. At the time, Ethernet was 1 gig. InfiniBand drove the technology for 10 gig. That was later adopted by Ethernet. It also pushed 40 gig, which was adopted by Ethernet. At some point, these became more lockstep because the industry investment and the underlying technology is the same. You now see 200 gig, 400 gig, 800 gig on both the InfiniBand and Ethernet roadmap. The second thing that InfiniBand drove was a technology called RDMA, to allow the applications to basically communicate without copying data. It's very efficient. Well, Ethernet co-opted RDMA with a technology called RDMA over Converged Ethernet, RoCE.
So you see this kind of development of both. Actually, you know, Ethernet today can be implemented in a lossless fashion with flow control similar to InfiniBand, but there are still differences, right? Because of how they, they, they grew up. Ethernet, I think, has an opportunity, as well as InfiniBand, to improve for this AI workload. Neither was built, purpose-built for AI, so the consortium will look at things as actually improving the flow control mechanisms to make it more granular, to add visibility, and to do some things around load balancing and other ideas that the consortium might bring to bear. I think it's the trajectory of Ethernet has been more of potentially competitors getting together to develop an open standard that can be adopted by many, and people have liked that. It actually leads to competition and more investment.
With InfiniBand, at this point, being more captive with NVIDIA. That's kinda the dynamic, is a little bit around the consumption model today, right? I don't think this is something unusual that Ethernet is doing just in reaction to InfiniBand. It's really the AI workload, and we see interest from the largest customers moving to Ethernet because of that multi-vendor component. They're interested in other GPUs. They might be developing their own, so they're interested in the open ecosystem.
In some way, wait for it?
I think pieces of technology can be adopted in a pre-standards mode, and I think typically, the Ethernet consortiums have gone to the standards committee for ratification, so it'll be a combination of both. I think a lot of people have the perception that Ethernet can't do AI today. That's not exactly true. It just can't do it as well as it will next year or the year after that.
Questions. I'd like to go back to this wallet share notion, that maybe I'm not fleshing out well, but there was a reason why these, these, these providers, like you said, might have paused spending in 2019 for, for myriad reasons, and then there was COVID. There were, there was a reason why, architecturally, when they built these roadmaps that are, you know, five, 10 years long, they picked Arista. There was a certain percentage of their spend that they were, they were affording to you, and then this changes as we go forward. Collectively, the hyperscalers, especially your customers, are, are spending more in 2024. I want to know, you know, when you look at that roadmap, what does that look like for Arista relative to what you're discussing with them?
Is it a notion of, not asking for numbers, but, you know, I would love the technological reasons for why more wallet share spend from these guys? Like, when you sit down strategically with them or if you know what they're working on, I mean, you know, not everything can go to the GPUs.
Right. I mean, you know, I think Arista was fortunate to enter the, the hyperscale market before they were hyper. They were, they were. It was an underserved, classic underserved market, and they were buying products designed for enterprise or service providers and redeploying them. A couple of them weren't happy with that and went out and built their own technologies. When Arista kind of emerged to focus on that market segment, you know, Microsoft and Meta, in, in particular, were kind of poised to grow at the same time, and we've been able to, to develop that over multi-generations. They were very concerned about operating networks with not as many people as users, millions and billions of users with a very small team, relative.
That's still very much of a focus, is simplicity of deployment, stability of the operating system, real block-and-tackle type of things. They've all diversified into more use cases. Edge use cases, Data Center Interconnect was something that was not on the original roadmap that we deployed. AI is probably the next new use case that, that we see coming to these data centers. We'll have to adapt. We're going to have to execute, you know, on those. Each, each one of these transitions or additional use cases, is in a little bit of an inflection, but I think we're well ahead of it.
Any other questions from the audience? I don't know where this fits in terms of your responsibilities at the firm, but Arista's been, is a good percentage of share of their revenue tied to service, software and services, and a good portion of that is software. I was just wondering if maybe you could comment in terms of that increased functionality and increased asks around this highly programmable, open cloud-based system that's Arista. you know, you could translate itself longer term for you all. This is probably more enterprise than cloud.
Definitely, yeah.
In terms of, ancillary software services.
No, great question. I mean, I, I think that we've seen in the cloud, you have customers that have the wherewithal to build their own management stacks and how they operate, also how they do network visibility, being able to collect a lot of data from the network and analyze it. Enterprise customers don't have the, the software teams to actually build that in-house. We're filling that gap with CloudVision and implementing many of that capability that is typical in the cloud and CloudVision, and we've extended that CloudVision into, you know, other SaaS-based technologies. Our Wi-Fi deployment is a cloud-based operating system. It's a ratable revenue model. We have now CloudVision as a Service, which we host rather than have the customers have to deploy on-prem.
You start to see these kind of elements come into our enterprise mix, and I think you can expect that to continue.
Especially with the mix shift to, you know, to enterprise.
Absolutely.
Going forward. Just, if I may, I know this, we don't have either with us who's, who's retiring soon, but, from a go-to-market perspective, just I think investors are always interested in terms of, you know, what's needed and, or, you know, we can just even maybe start with what's the go-to-market kind of motion...
Right.
for Arista to attract, to attack this gigantic TAM, where they have relatively very low share in many aspects of enterprise network equipment and how it could impact the model.
Yeah. I think, I think a couple of things that we've done differently, as folks have moved into the enterprise, and challenged the incumbent, they typically come from a mid-market setting, and then have a value proposition that it's good enough. You don't need all the features and functionality. This is cheaper. It's 80% the, of the cost, and actually all that complexity is something that you don't need. Our approach has been to actually go after the most sophisticated customers first and build downwards. Financial vertical first, very challenging vertical, and then build out from there and bring all the architectural pieces that you have to build an enterprise network.
The account teams get very focused on the Fortune 2000, how we can grow our wallet share in existing accounts, how we can penetrate some new accounts that have a high network spend. We have, you know, partners and resellers that come in to deploy those networks. Oftentimes, they're able to find other opportunities in adjacent regions that we, we probably wouldn't have gone after ourselves, and that's how we build out the go-to-market model. There's always an opportunity to expand in a region. If we find a new opportunity and an anchor to start deploying, you know, a sales engineer and an NSE to go after that and start to build out a region. You'll see us do those kind of things to build out the go-to-market market.
in the near term, just focusing.
Focusing on that model. I think we get a lot of questions about the channel, and at some point in the future, as we start to get satisfied with our share in, sort of that Fortune 2000, I think that is an opportunity that would require more investment in the channel, but that's not here today.
Well, there's one more time in the audience. One more question? Well, thank John very much for his time.
Thanks, Tom.
Thank you for coming.
Appreciate it. Thank you.