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

Good afternoon. My name is Fahad Najam. I'm the senior equity research analyst covering comm equipment and optical components for MKM Partners. And today, it's my pleasure to welcome Anshul Sadana, Chief Operating Officer of Arista Networks; and along with Arista IR. Anshul, welcome, and thank you for joining us.

Thank you, Fahad. Pleasure to be here.

Anshul, I recently had the opportunity to listen to Mark Zuckerberg's keynote speech about metaverse, followed recently by Jensen at NVIDIA talking about omniverse. And I'm still trying to figure out what these things mean. But I know that networking is going to be a critical piece of all whatever Mark or Jensen are envisioning. So for the benefit of the audience, at a really high-level picture, can you connect the dots for us? How does Arista play to this opportunity? And what does that mean for Arista?

Absolutely. Fahad, when you look at AI and Andy touched on this on our Analyst Day as well, AI is huge because it's finally reached a point where you can utilize it to offload many, many different types of things in decision-making that is going to be very hard for humans to try and keep up with. So when you apply that to our customer base, you mentioned Mark Zuckerberg at Facebook as an example, they have tons of data. And in order for any consumer to see their Facebook feed or their Instagram feed, there's a lot of analytics going on in the back end. And the better that decision tree of what to show the user, the better the experience, the more revenue for all of these companies as well, which means AI is here to stay. It's already starting to get leverage, and I think it will continue to grow. But AI, unlike compute, pushes a lot more data into the network. And the next generation of network architecture or the speeds going from 400 gig to 800 gig or 1.6T are starting to get driven by the AI ecosystem less so by compute and storage. I think that is a great opportunity for us, and we're already participating with some of our customers, and I think we'll continue to grow with that as well.

Thank you. So you laid out an ambitious and very impressive outlook on your Analyst Day: Mid-teens growth from calendar '20 to 2025, almost 30% revenue growth for next year. But can I -- beyond the numbers, help us understand and appreciate where does Arista want to go into the future? What's in store for Arista? And what are you thinking about in terms of that vision that you have? How much is that going to be done organically versus M&A? So kind of big picture view connect the dots for us there too.

Yes. Fahad, we started with cloud networking focused on cloud customers and then enterprise data centers. Since then, we've expanded quite a bit. We expanded into campus. We expanded into routing. We expanded into security, observability. And we continue to do more in cloud connectivity as well. So today, our TAM is already about $35 billion or so, and we're just touching $3 billion. So there's plenty of room for us to execute and grow in the current TAM for years to come. And I think that will continue to be our primary focus because if you try to do too much beyond that and get distracted, it's also easy to miss and then misexecute, and we don't want to do that. But as the ecosystem is evolving and is doing well in campus, we've mentioned our focus on software and services Awake Security, the AVA AI engine are good examples of that. We're going to find many more opportunities in those adjacencies to keep on enhancing our solution and growing our TAM as well. So I think most of the growth will come from within the company, but we wouldn't discount M&A. If the right opportunity came, we'll do that too. But that's not the primary focus for growth. I think we have plenty of growth left on our core business through the odd peak existing TAM, including adjacencies.

Anshul, so I think that's an interesting point. You talk about AI and the enterprise. But I'm a former network engineer myself, having spent 10 years in the industry. AI requires intelligent infrastructure, right? You cannot run AI workloads just by themselves. You need intelligent infrastructure, smarter hardware. And that's something that, whether it's a WiFi access point or through your Mojo acquisition, but the supply chain and the whole ecosystem around intelligent infrastructure is still coming together. And so it's a huge area possible, but I was hoping maybe you can talk a little bit about the AI opportunity in the enterprise. And obviously, you're going to grow with your intelligent cognitive cloud software's features, but you also need the associated hardware. And that's something that you haven't done historically yourselves.

We mentioned this at the Analyst Day, the evolution of NetDL. So the EOS stack, our operating system has evolved, and now the state streaming has gotten to a point where you can build a network data lake. And you're absolutely correct, the smart infrastructure, not just on the GPU but in everything around it as well, and the network is getting way smarter too, understanding where does data flow, understanding how to deal with link failures or outages without disrupting the application is going to be a huge too for us as well as for our customers. So the Awake acquisition got us the AVA engine, which is a machine learning-based AI engine that deals with flows in security, but we're now able to extend and apply that into all of our datasets. And essentially, route that to a much smarter network. That is our vision for the next 5 to 10 years to evolve the network into being much smarter as well. Fahad, we talk about AI. We talk about the future of the industry. But you mentioned you've come from networking, and you know the answer very well. Network engineers spend most of their time on practical stuff like code upgrades and box scrubs and p-sorts and patching and so on. That just needs to go away. Your car and your phone can upgrade themselves. Why not the network? Why do network engineers have to spend so much time? When you look at CloudVision, when you look at AVA, things we're trying to build here, that's the direction this is headed. We're building a much more intelligent network for the cloud, for the enterprises, for the whole world, and evolve ecosystem into bettering networking as well.

All right, so let's shift focus to the near to mid-term. I get this question a lot from the guidance you guys gave on your third quarter earnings call, about at least 30% growth in fiscal '22 and mid-teens growth from over the 2020-2025 period, mid-teens CAGR growth. Obvious question is how could you really anticipate competing on the fiscal '22 targets? What's driving that picture, if you can? And what are the risks? And what are the upside potential to that outlook too, and obviously, the risk to the downside?

I was hoping 30% is great job. We're just satisfied with that. Look, growth is good. We did have a pause a couple of years ago, and the cloud is back, very strong, growing with 400 gig, data center interconnect, multiple use cases, many, many roles. And the enterprise business for data center in campus continues to be strong as well. But so are the other remaining verticals. So if you look at the providers, both the Tier 2 cloud and the service providers, that's growing as well. So I think overall, we're in a healthy place where the entire business is growing, led by enterprise and the cloud, but the others are strong, too. Fahad, when you look 12 months out with the cloud especially, we have good visibility largely because of long lead times on components. But with enterprises, you don't have profit visibility 12 months out. But I think we can project our base on the models we have. So I think 30% seems like a reasonable growth target, given the upside we've seen from the cloud. Obviously, if the cloud decides to spend faster and grow more or the enterprise did better or supply came in better than expected, then there can be upside. I think those would be the levers for upside. At the same time, 12 months out is still -- there's always some risk and any one customer changing their mind, and as you know, that's happened a few times in the past, can cause downside too. But for now, I think growth seems stable and healthy. And we're quite optimistic about the future, not just 2022 but also the next 3, 4 years after that.

So if you can rank order the greatest growth opportunities, '22 and onwards, what would they be in the cloud, enterprise, Tier 2 specialty cloud service providers?

Fahad, the enterprise, I think, will continue to be a strong growth vector for us for many, many years to come, both data center and especially campus, campus, where we're just getting started. But those I think will continue to be the strongest, more sustained, I would say, for years to come. The cloud is a volatile sector, as you know. We're in a growth phase now. At some point, the growth will peak out, and there'll be a slowdown before there's the next cycle. Can the 2 cycles overlap and there's no slowdown? We don't yet know, right? We'll find out 3 years from now or so. But the cloud in general has been volatile. We just expect some of that in general. So when you mix the 2 for the company to keep on growing, we have to do better in the enterprise. I think that will help evolve and provide stability and diversification to the business. And then everything else that is happening is strengthening it, whether it's the security focus we have or the cloud connectivity focus we have. The Any Cloud architecture, I think, will improve too. Does that answer your question? Is there anything else specific you're looking for?

No, no. Those are big buckets that I was looking for in terms of enterprise, the biggest driver of growth, followed by cloud. I guess maybe a follow-up to that is, in the cloud demand, you talk about enterprise being sustainable over the next 2, 3 years, even longer. But the strength in the cloud, is it just because of the perfect storm because of the component shortages from '21 meta and Azure growth plans creating a perfect storm in '22 for demand? Or is this very compressed? Or can this also be sustained over a 2- or a 3-year period?

Well, you know what, 2022 looks like it's going to be very hard to predict accurately what all of the future years look like. But someone asked me a few years ago, the cloud has grown very well since 2008, 2010. The public cloud is growing. Many workloads have already moved to the public cloud. How much more does the cloud need to grow? And there's no easy answer to that question, as you very well know. So I ask my cloud customers the same question. I got some different answers. But the way these companies are thinking is that they believe they'll need to keep on growing, and they'll keep on taking more workloads for the next 10, 20, 30 years, so these are very long-term sustained growths. The growth rate may slow down, but it won't be 20%, 30%, 40%. It will go down to single digits at some point. But these are TAMs that will keep on expanding. I think because of opportunities like these, I think we will continue to grow as well, along with that market. Same with enterprise. Enterprise is a very large TAM, largely underserved by reputed players competing against the incumbent. And today, we're gaining share. And I think there's still plenty of work left to do. So I think these 2 vectors, when you play it out, can get you a lot more. The next 3 years, I think some of it is modeled in post-2022. But we had different up-and-down scenarios we played out and sort of came up with a blended model to get to the mid-teens growth for the 2025 period. But there can be upside and downside in that, and I think really comes from the end customer demand largely from these 2 verticals, but also the providers I think are needed, right? We don't want any single vertical to slow down because then it becomes a baggage. Right now, all are growing very, very well. And the actual raw demand is good. This is not a perfect storm simply because of supply chain issues. If anything, customers are being more thoughtful in planning now than they were before, but the raw demand is very strong. And we can plan in the market, right, how much time do people spend on their phone, on their apps? How many workloads keep on going to the cloud? I think that will just keep on evolving.

Yes, that's a very powerful data point. We're all -- we're connected all the time, and with 5G, it's not just us with our smartphone, but our devices will be connected all the time. So the massive growth in data consumption is just going to continue. There's a question here from one of the investors. Can you share a bit more color as to campus opportunities? For example, how do customers see a better ROI by switching to Arista from the campus network providers? Any anecdotes you can share? Also, how you plan to go to market for the campus, including the local support? Do you work more frequently with local distributors? Loaded question.

Yes. Well, when you look at the campus market and the opportunity, there's plenty for us to do. But it's not just replacing boxes with new boxes. The architecture is changing. Segmentation and security are far more important on campus now than ever before. You talked about 5G and IoT. And in campus, we worry about whether you can deliver 30 volts or 60 volts of power to the access point of the phone with power over Ethernet. But if you have a badge reader or a camera, how do you make sure that's not compromised? And on a laptop, you can install an agent and have endpoint security. You can obviously do that on light switches or cameras. So now you have to think about security along with the network. And this is where the weak network detection and response solution comes in as part of our offering. So things are changing. So that's one dimension of that. How does a customer measure this [ TO ]? It's very simple. You run a network for 6 months with Arista devices, and customers swear by it, that this is the first time the network works as expected. They don't have too many outages. They don't have to waste so much time on worrying about config drift, worrying about security patches, worrying about scheduling downtime, worrying about oversubscription and why the packets were dropped. It's very, very easy to use. So very rough metrics from our customers. The time to deploy, time to monitor, the time to make the network work is 3 to 4x less than what they have been incurring with the incumbent. And as you scale campus markets to large global networks, large global companies, you have a small team. We have a pharma customer that has 11 network engineers and architects in the core team that manage all their campus infrastructure globally. They have 70 offices globally. And everything else is remote hands through some outsourced company that is helping them with the wiring and physical sort of touching up the switches, but the actual config changes and operating of the network is done by these 11 engineers. They look at CloudVision, said this is a game-changer. They can manage 70 offices all from one place with CloudVision as a service. This is very, very unusual, sort of unheard of for them. And they're able to scale significantly better. They're a great customer already. And we have many more in the works similarly. So I think the quality, the ease of use, and the level of automation that we bring in with CloudVision is unmatched and helps us succeed in campus.

I appreciate the answer. Going back to my train of questions. In terms of your fiscal '22 outlook, 30% growth, can that be sustainable -- maybe can fiscal '23 be growing 30% as well given the massive growth in demand and backlog that you're seeing in the visibility, 180 days, I've now heard now extended for some components of 200 days. So can this 30% growth be sustained into '23 as well?

I think that's a high number. Obviously, if there's upside in demand we'll see that. But at this stage, it's too early to project growth with that high rate post 2022. I think as the numbers get bigger, the growth rate will come down. As I mentioned, our customers have surprised us in the past. So if they surprise us positively, we will be happy with that, but it's too early to say.

Got it. On the component shortages, on the risk side, obviously, components are gating your ability to ship against this massive backlog. Can you provide us just the latest view on backlog. I had -- on component shortages, I had a company this morning talk about that the chip availability is actually improving, and they're becoming available for a price, of course. But that is unlike 90 days ago, but you just did not have chip supply. Now a piece is becoming available. So can you share with us what you're seeing on the component shortages side?

Yes. The component shortages are still there. I would say the market has stabilized a bit in the last few weeks. Even in calendar Q3, things were getting worse, but now things have stabilized. I wouldn't say they have improved significantly, but we're no longer getting lots of decommits every day. Everyone is managing to hold on to their dates. If they promised 52-week data, they're at least delivering in 52 weeks rather than having a delay beyond that. The broker buys and some availability through brokers are rare, right? Because the broker market is very, very expensive. They're talking around 60x. In one case, we saw 250x the cost of the chip if you wanted to buy it from brokers. And you can do that for 1% or 2% of your buys. That can't be your primary suppliers. You have to get it from your key suppliers. Overall, the semi space, so there's commodities, metals, connectors, heat sinks, PCBs and so on. There's smaller chips [ TI ] stood on long lead times but, again, now finally stable. You have FPGAs that are on long lead times. You have Broadcom chips on long lead times, CPUs as well. So all that market is still constrained in 2022. And what we are hearing is there isn't enough fab capacity in the world to try and keep up with so much demand this quickly. So by the time things stabilize or come back closer to normal, you're looking at mid-2023. It won't be a sudden improvement. Every quarter, you'll get a gradual improvement to get back to normal. But our current expectation is that the recovery is actually in 2023, not 2022. But we just stay with this stable long lead time structure for a while.

Okay. Well, that's actually good news that it is stabilizing at least.

The only risk or the very big risk here is that assumes nothing else goes wrong in the world for the next 18 months because the world is very fragile, one more natural disaster and things fall apart, right? So we have to get through this entire period of tight supply without any more hiccups.

So hopefully, no earthquakes and tsunamis and pandemics [indiscernible] for life. All right. So focusing on end markets a little bit. Obviously, cloud is your largest vertical. My estimate is it's over 40% of your revenue. Hopefully, we'll find out in your 10-K what it was in '21. But of that cloud revenue mix, I think Microsoft and Facebook account for most of it. Can you speak to the diversification in the cloud? Or like -- I think your non-Facebook and Microsoft revenue actually doubled last year based on the 10-K analysis. But can you share with us what you're seeing with the non-Microsoft, Facebook hyperscale penetration?

Sure. So Microsoft and Facebook continue to be very good customers of ours. Their spend is volatile, so when they spend more, we benefit. When their spend slows down, you will see the slowdown in our revenue as well. But we do well in the overall cloud titans market as far as the Tier 2 cloud, too. Now Fahad, I'm not sure of the numbers you mentioned. I think since they're a year older or older, I'll skip those for now. But in general, I think our cloud momentum is strong. So we're not expecting that to slow down. A big part of our growth in 2022 is coming from the cloud and better than what we -- even this year, we started with flat to down. And then it became flat last year, and then it became flat to positive growth and then single-digit growth and now the company is doing better. So when you look at cloud, you have to come up with a 3- to 5-year average rather than look at any one specific year simply because of the way their spend cycles work and their upgrade cycles work. But in general, we are happy, we are pleased that they will continue to do well. The other cloud guidance we've done well as well. It's a matter of their spend in this space, not a lack of execution from us. I think they are very pleased with our execution and are quite happy and are continuing to reward us. And I think that relationships -- those relationships will also get stronger and bigger over time. But they all take time and all work in the right direction, right? You've done the back of the envelope analysis, that there -- the other cloud titans are growing very well as well, and I think their companies will grow.

So as I look at the cloud opportunity, I think one thing that the component shortages I've highlighted is that the white box vendors were hit the most because of the lack of scale that they had relative to you and your peers, the more vertically integrated suppliers? And are you beginning to see cracks open up in the white box wall within the hyperscale?

I've actually mentioned this for a couple of -- at least a couple of quarters ago, maybe almost a year ago that we were already seeing some cloud customers open up a few layers of the network, not the entire but a few layers of the network where they have been using white boxes to be more open to buying from the industry. And that's more of an opportunity cost where if they can leverage the rest of the engineering team, get a better outcome, then why not, versus trying to build on their own and if they're late to market, they take a beating. If they are single-sourced on the internal solution, that's a risk as well. Now on top of that, the whole white box situation, where everyone was trying to do just-in-time inventory, they realized that you do need to plan better and buffer up. But it's not just the ability to take risk. It's the ability to invest so much time on planning how to take that risk. That goal comes into the picture. And we've been doing this for more than 10 years now. The cloud customers know that very well. We carry buffers of chips. We carry buffers of many key components. We have a larger business that we can scale as well. So based on all of that, I think the cloud companies are much more receptive to us. There may be short-term opportunities where they're saying, "Hey, we are short of white boxes. Do you have something?" But that's not the business we're trying to chase because those are short-term wins. We're looking for more sustained long-term wins. And the design wins we're going after, I think, are more sticky as well. It doesn't mean we don't get the benefit if we have products people will buy those too. But I think our business, we believe, is much more stable rather than the short-term change in mindset. But some customers will change their mindset and say, no matter what, if they were only on white boxes, that's too risky as well.

Are you beginning to have this conversation with this customer...

Correct. And I mentioned this on the Analyst Day, which is why it's not build versus buy. It is build and buy.

Got it. I guess further out, not imminent, but further out, maybe 4, 5 years out or maybe longer, is the shift towards more co-packaged optics? And I certainly think that the ODM suppliers of white boxes do not have the technology and the capability to deal with that co-packaged ecosystem. And I believe that is another mega cycle or insertion opportunity for Arista. But can you share your thoughts on what you think about that opportunity?

I'm not here to predict the timing of co-packaged optics. It's a highly debated topic as you very well know in the industry. I think the customers' preference in general is to stay with pluggables as long as they can and be ready with co-packaged, be fully, fully ready with co-packaged as soon as they absolutely need it. Everyone thought that cycle might be '22. It's turning out that it's not. It's pushed over '24, '25. And if the 100-gig ecosystem, 100-gig Sun ecosystem requires it, then we'll see that. If not, then they'll get pushed out to the 200-gig ecosystem, which is more 2027. But you are correct that the products are getting more complex to build. And even without CPO, co-packaged optics, if you look at just the thermals of a device today, we announced our 25.6 terabit products along with Facebook on stage at OCP. And the thermals are mindboggling, right? You're looking at 1,900 watt -- 1,800, 1,900 watt type of switches with the optics and all the airflow cooling that you have to plan. You play that out to 51 terabit, and you're looking at just the chip itself being really hot. And then you have the optics and the interconnects. And you're looking at, again, 1,900-, 2,000-watt systems. Those are not easy to build. They require a lot more complicated analysis than mechanical engineering. And then the manufacturing and the testability, all of that comes into play. I think customers do have that in their mind that their current supply chain ecosystem may not scale. And hence, they are better off leveraging our teams for these types of situations.

Maybe more nearest term, 800 gig and 1.6, we -- is probably a little bit further out. But Andy Bechtolsheim had this really awesome presentation at ECOG where he essentially laid out a case that maybe some of the hyperscalers will start upgrading to 800 gig by calendar '23. So it looks like there may not be a pregnant pause between upgrade cycles to your benefit. But just quickly, we're almost running out of time, but I just want your thoughts on that.

People think of 400 gig as an upgrade cycle and then 800. But actually, there's a lot of intermediate speeds and breakouts and every other transition that you need. But I think there's enough demand for bandwidth, especially driven by AI workloads that, for certain use cases, customers will go 800 gig as soon as it's available. So I think we'll participate healthily. Even in 2025, my prediction is we'll be selling 100 gig, 200 gig, 400 gig, 800 gig all at the same time to the same customers, but that expansion will continue, something which bodes very well. But you're correct. 1.6 also is around the corner, and Andy and the team have been working hard to make that a reality, at least by 2024/2025.

I have to keep the train running on time. We're almost out of time. If I had the luxury of time, I would have asked you about service provider, why that segment is struggling a little bit, but we'll use that question for next time. Thank you all. Have a good evening and rest of the week.

Thank you. Bye.