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

Good morning. I said it before, if I sneeze in the middle, it's just allergy, don't worry about it. It's not COVID. Thanks very much for joining us. I'm pleased to host Anshul Sadana, COO of Arista. We've written a lot about Arista recently. We published a primer on switching and routing, and we spoke about basically drill down to the bottom of discussions of what drives growth in data centers, what drive growth in cloud to address growth in campus. It's a big 30-page report that goes over the drivers and the companies and of course, Arista is one of the highlights of the report. So we have a copy of this primer down in the main floor and if you don't see it there, and if you want to have a soft copy, just shoot me an e-mail and I'll send it to you. So first, thank you so much for joining us.

Thank you, Tal, for having me.

That's great. One of the highlights of the report was that the market is going through superior growth. You don't have to be a rocket scientist to see but certainly, the question is about sustainability. And I know that Arista did not see the decline, meaning you're a different player. But let's try to speak first about the market and not about Arista. What is driving the current increase in demand that we're seeing across the board? And in your view, is it sustainable or not?

Sure. Tal, there are 2 broad markets we play with, that's the cloud, which includes the cloud titans, that includes the Tier 2 cloud or the cloud specialty providers and then you have the enterprise. The cloud is growing. There's no doubt about it. I don't have to sort of give you any analysis there, you already have all the details. You've seen the cloud companies grow. And the cloud is in a long-term growth cycle, whether it is a 2% CAGR or 20% CAGR, I don't know, but it's going to grow. When we talk to our cloud customers, these are the visionaries that are thinking through for the next 5 years, 10 years, 20 years, they believe that they've only gotten about maybe 10% of the workloads that should be in the cloud now running in the cloud. So the cloud can grow 9x from where they are today. But no one knows the slope of that line because there's extra work needed, there's always competition, there's disruptive technologies. Things change. You can't really predict 20, 30 years out that easily. So I think the long-term trend is likely to continue. On the enterprise side, at least the markets we are in, we've seen good healthy demand, we haven't seen a slowdown. We've seen people continue to invest in technologies in automation, in sort of infrastructure that gives them a competitive advantage and data centers or IT infrastructure, in general, is a key pillar of that. So I don't think people will underinvest in this area. Now we're not a bellwether to declare a macro event. So if there is a macro event, we'll get affected by that as well but so far, we haven't seen it.

Yes. But I do see differences between vendors because your best segment last quarter you highlighted on the call was enterprise, the worst segment of Cisco was enterprise. What drives Arista to be so much more successful in this big enterprise segment versus competition?

I think you still have to go back to way basics of our technology. And we've built products, solutions tied up fundamentally better and as a result, we're able to gain share. As you know, an overall high-speed data center networking DNS switching, there are about 20% global market share, plus/minus here and there depending on the quarter and so on. Now we grew from 0 to get there. A lot of this comes from the cloud titans. But in the enterprise, there's more room to grow. So when companies face hurdles or challenges or they finally say, you know what it's time to wake up and look at a new architecture. In today's world, you don't have to be a rocket scientist to figure out how is Microsoft or Facebook or Google building their infrastructure and you look at the technology and say, you know what, actually, it's quite straightforward, I can do it too. Maybe I need a little bit more help on automation, and we'll talk about CloudVision a little bit more in the talk as well, that gets the enterprises going. Whereas with legacy technologies, they're not only seeing a real change in how they run the network. And this is a discussion I often have with all of our key enterprise customers. Do you know what's the most heavily used automation tool in networking in the enterprise segment? Anyone? It's called copy paste. Works really well. You copy a config, you paste it somewhere else. It has an error, you're going to replicate it throughout and you'll spend the next 6 months fixing it. And if you have to change one line of code, you'll need 9 months of compliance checks and so on. This is the 90s, we haven't grown up. So we've fundamentally changed how you think of networks, automation resiliency. Why is it that upgrading one switch in your enterprise requires an outage? It should not. Who runs their business that way? When was the last time you got an e-mail saying, dear customer, the cloud will be down from 2:00 a.m. to 4 a.m. Does that mean the cloud does not schedule outages or they don't do maintenance and upgrades and changes? They do all the time, but the architecture is more resilient. So you start from there. The architecture comes from how you design the network. It comes from how you build the software. It comes from how you operate and make changes on our digital basis, and we've fundamentally changed all 3 of things.

Got it. So I want to drill down maybe to the verticals and products, and I want to start with the cloud. Can you discuss, first of all, your position in the cloud and there are many aspects to your position in the cloud. Maybe first on a standalone basis, what makes you so successful in the cloud that we haven't seen other companies managing to replicate?

I think it's fundamentally execution. Execution is hard. In Silicon Valley, there's a common saying there are 1,000 ideas, everyone has them. It's very hard to actually execute. We've shown that over the last 10 years. We are roughly on our fourth or fifth generation product right now with our key customers. And every 3 years, like clockwork, they have the next gen ready for them with higher throughput, lower power, more efficiencies, more resiliency. And every feature that cloud customer has asked for or we have anticipated that they will need running networks at scale. And all of this has to be done with very, very high quality. In my previous life, we used to measure QA teams in terms of ratio of developers to test engineers. And that's largely how the software industry has been built. But at Arista, we have about 1,200 plus software engineers writing EOS code. Guess how many test engineers we have at the company today? The answer is 8 megawatts. It's not people. It's fully automated. So the product has a lot of test cases, about 150,000 test cases that run every day. And any time a customer ever runs into a bug or a fault, we'll add a test case to make sure this is never repeated again. But you can't just think of this and add it later. You have to build the software for testability. You have to have that architecture, you have to have modules that are completely independent of each other and segmented and isolated. And the power of EOS really starts to shine. The power of our hardware development team really starts to shine. And that is really why we have been ahead.

So on one hand, you talk about being extremely high end and doing automation, adding automation to the testing. On the other hand, we are seeing white boxing, we're seeing more commoditization in cloud. How the 2 worlds live together?

I wouldn't say more. I think the white box ecosystem has managed to maintain status quo. And to a great extent, we are part of that ecosystem ourselves. So there are really, broadly speaking, there are only 3 high-volume white box customers in the United States. Then there's thousands who try experiments, very low volume in labs and so on. Out of these 3, 2 of them have religion to build their own hardware. They run at scale. And no matter what you do, they want to build that particular layer of the network on their own, at least the hardware. But they started very early on when Arista wasn't a viable option. The others -- the other cloud titans, including one that builds and buys, and one that just largely buys but might run their own software, we found a way to co-develop with them. So you have to go back to what is it that the cloud companies want to do with white boxes or building their own product? And in most cases, the thesis is they want to own their own IP. They want to have their secret sauce. They want to control their own destiny. They want to get the market -- the advantage for time to market and so on, all the advantages you get by building a product. These companies who are very large customers for Arista have also studied what are the disadvantages of using your -- building your own white boxes. It's a huge advantage if you can gain time-to-market lead over your competition when you build your own products. It's a huge disadvantage if you're late. And if you build your own products, you only have one choice, and sometimes they are leaked. Sometimes they don't have enough supply. Sometimes, they don't get the most power-efficient solution because they have 3 hardware engineers thinking about in isolation rather than the entire industry. What we managed to do with our key customers is actually just sit down with them and have a very open discussion and the conclusion all of them have come to is by partnering with us, even if they want to build some on their own, they get to a much better solution. So as an example, with Facebook or Meta, we've announced 2 generations of products at OCP that were codeveloped by Arista and Facebook together. It's a very unique situation. When was the last time as a supplier, you go to your customer and say, dear customer, it's okay, if you want to build everything on your own, I'll still help you. But you have to think through the long-term implications, what happens 10, 20 years from now. And what we concluded is partnering with our customers is the best solution. Our customers have tried it. They love the outcome. They're able to build solutions faster than what they can do on their own. They get contributions from our entire hardware team and Andy Bechtolsheim that they otherwise won't be getting. As a result, they have a much more powerful solution. So this whole white box ecosystem, I think, is maintaining status quo. I don't think it's going to pivot significantly one way or the other. The hype might continue, but this cushion and our answers haven't changed since 2013, since our road show for the IPO. So I don't think we have to worry about it as much.

Do you think there is any chance white box switching, white box routing gets to the enterprise?

It's hard. There's been a lot of comparisons to Linux. What about open source? Why can't the entire industry get together and build it on their own? When you look at the computer of the server market, there's roughly 40 million or so servers out there, and you have a large ecosystem of app developers and so on, so looking at millions and millions of developers that contribute to that ecosystem. Switching on the other hand, has a much lower volume. So every 40 servers is roughly 1 switch. And the network technology and so on is also a little bit more complicated on the OS side with so many functions and features and security and so on that [indiscernible]. So it's hard for customers to put that much resource in there, which is why I think if you have companies that are willing to invest 10, 20, 30 software engineers to build a network stack of their own, maybe they can do it. They are compromised. And they'll say, I'll not just not use these advanced features, I just want to barebone network and so on. And some customers try that. They are generally start-ups or smaller companies, as they grow up, they to realize how much more is needed to run the business. If you look at any of the big banks, including yourself, the compliance side of the infrastructure is as big as network operations. Sometimes it's actually bigger, just the regulatory pressure and if there's a security event, which happens on a daily basis, just busy chasing that. And an enterprise trying to build a stack on their own and you have a vulnerability that you're required to fix in 6 hours, it's just not going to happen. So I don't think real enterprises will adopt it that easily.

I want to go just a few more questions on cloud, and then we'll move on. But we've seen Nokia recently making an announcement getting into a cloud also. Are you seeing more and more companies, as the market standardizes, let's say, are you seeing more and more companies getting in and taking away from the opportunity for you?

No. I think the Nokia announcement, if you go back a few years, they made a similar announcement with Apple. Nothing really happened. So this is more of a PR announcement in my view. And we are just not worried about it or concerned, I think it's noise.

Got it. I looked at the market share data for my primer that I published I look at market share data, and I was surprised to see that the 400-gig, you have 33% market share, Cisco was at 10% market share. And cloud companies are basically white box, which is cloud companies are basically half of the market. Talk to us through, first of all, talk about 400-gig in general? And what are the applications and why is it being deployed, but also your position in the 400-gig? Why we're -- at least historically, that's what the data shows you were very successful.

Yes. So the whole white box thing, there was an article about 6 months ago that said that 400-gig white boxes have 90% market share and 400-gig is different and white boxes will just take over the world. But you have to sort of -- if you read the book, I think it's called How to Lie with Statistics or something like that, you just have to slice through the data and realize that, yes, the 2 companies in the United States that build their own switches had all the 400-gig ports at that time and essentially that combined 90% market share. So the market will get bigger over time. So I think the white box ecosystem, as I mentioned, will maintain status quo. We're not seeing any major shift in this area. But when you come back to 400-gig, when you -- when we went from 1-gig to 10-gig, 10-gig to 40-gig, 40-gig to 100-gig, with each transition, the new technology made simple business sense, the new product, the new technology, the cheaper, it's faster, more power efficient, not just cheaper per-gigabit, but cheaper per port or per device. And as a result, everyone just upgraded. When you go to 400-gig, 400-gig or 32x400-gig switch, which is 12.8 terabits in throughput, cost about 2.5x the cost of a 32x100-gig switch. So if your network architecture and the applications require more than 100-gig of throughput per interface, then it makes sense to upgrade. If it doesn't, you don't have to, which is why the next generation with our customers is actually not just 400, it's 100-gig, 200-gig, 400-gig depending on the customer and the use case. But as you go higher up in the stack like the data center interconnect layers. The data center interconnect for many of these customers has long running fibers. You're running 144 fibers for about 100 kilometers or so to interconnect data centers. That's a very heavy capital investment. You can't just simply add more fiber just because you need more capacity. That's a multiyear project to get permits and dig up the whole place and what not and so on. So if you can add more bandwidth through those fiber runs, you get a significant advantage, which is why DCI is the primary layer where customers are using 400-gig. Now many customers are using it as top of rack, largely breaking it out as 2x200 or 4x100 instead of the 1x400, but the market share numbers will measure that 400, that's okay. Over time, over the next 3 years as compute gets more powerful, as you truly start to drive more than 100-gig IO per storage node, then the market will just keep on upgrading there. But we don't stop there, 800-gig is right around the corner. And there are multiple variations of this technology, which impacts the type of optics you use, the type of fiber runs and the capital infrastructure needed, whether it's 4x100 or 8x50. And when you get to 800-gig then 8x100 or 16x50, the various technologies customers are investing in. So market will sort of get fragmented through this upgrade cycle. But in the end, everyone will get to a next-gen product at either 100, 200, 400 or 800 with different [indiscernible] dual speeds.

Got it. My general question next is how cyclical is the cloud market, but I want to give a little bit of a background. Microsoft had starts and stops and starts and Facebook had starts and stops and starts. And the question is what we're seeing today might be -- and I'm asking it as a question, not as a statement. I'm just -- I'm trying to understand. What we're seeing today, which is very strong growth in the cloud space might be the result of an increase in traffic 2 years ago, a year ago because of Covid because our behavior changed, as consumers, as employees. The question is, how sustainable is the growth we're seeing today? We started with this question, but how sustainable -- how do you know that what we're seeing today is not just a build out that may last in the year, another 2 years, that is the response for the extra growth in traffic we've seen?

I think this is a really good question because there's so much hype in the market for the last 2 years. And the market was untethered and finally, there's a massive correction on stocks that benefited significantly from work-from-home behavior. And a lot of people ask us, did something similar to happen in the cloud. And if you go back, someone like Microsoft, they're actually published on their blogs that in one of the countries in Europe that to add instant capacity just to keep up with all of the video conferencing traffic and so on. But what people misunderstand is in that one country, they had to add 2 terabits of traffic at their edge. That's 2 pizza box switches from us. It's peanuts as compared to the whole business we do with these customers. So that doesn't really change the revenue outlook for us with the cloud. I think that cloud is in a healthy growth cycle. We've seen this from them from various angles, whether it's more enterprises moving to the cloud, there was been -- there was and maybe has been a shift in retail, going more to the cloud than before. And between consumer and small business, I think there's been different cycles and so on, but they're not the largest consumers of the cloud, at least not the public club, right? You might be using iCloud and so on as a consumer, but that doesn't change the build structure as much. So in the near term, I think demand is healthy. I'll come back to you, Microsoft Facebook starts and stops, but longer term, can we project how this continues? We don't know. These customers are somewhat volatile. Now we have more than one cloud titan, let's call it, 2.5 to 3, maybe if you add up all of them together. And as a result, there's a little bit of blending that happens. If one of them start and stops, the others keep going. It's not a big deal. Everyone stops at the same time, we'll feel it too. It's impossible for us to predict that. I think if I had to simplify this in my mind how -- because we have to plan for these customers, demand as well on the supply side. And I know you will not ask me anything about supply chain today. If you project a line out, I think there's good growth. And it's extremely hard to say if this is a 2-year cycle, 3-year cycle, 4-year cycle? Do we have a repeat of 2019? We just don't know. Longer term, it just continues. Now when you go back and why did things slowdown in 2019, 2020? There's nothing correlated between customers. These are completely independent items. With Microsoft, there was a start and stop it's just 1 quarter. If you go back to the February or May 2019 earnings call and we announced the start and stop, the full year spent by Microsoft didn't change much. So it's really a 1 quarter change in how they were doing financials for themselves and CapEx versus other buyers and so on. With Facebook, that has a longer change where they decided to sweat their compute assets more, for one more generation and that pushed out the network refresh and the upgrade of the next architecture as well. And in hindsight, if you ask them today, they probably wouldn't have pushed around as much, but no one anticipated COVID and the demand that came with it, right? So hindsight is really 20/20 in this case. But these customers also learn from that. And I think their current thinking is having a little bit of buffer in capacity is important for their business. They don't want to be as lean as they end up pre-COVID. So I think that might be a longer-term adjustment, but that's -- you're talking just 4%, 5% momentum around there. So impossible for me to predict whether there's a downside coming or when is it coming? I think longer term, we grow, but this is a volatile segment, and you have to be prepared for that volatility.

Got it. Data centers -- enterprise data centers. Let's talk about the drivers in the space because we have work-from-home, then coming back to the campus. There is also a change in the way applications are being deployed and applications are being consumed. So what drives the growth that you are seeing in datacenter -- in enterprise data centers? I don't think work-from-home actually changed much for us in enterprise data centers. If you look at our growth rate for enterprise, especially data centers, pre-covid, during Covid, even now, it's been healthy just throughout. There was almost no change. The only impact we saw with our customers was when they had to add more VPN concentrators or firewalls or load balances for all the work-from-home traffic coming into their infrastructure, they added more capacity, but that's generally you're adding a few gateway switches to connect these appliances to [indiscernible].

So I don't think enterprise data center has changed much. Most companies have realized that the investments in digital infrastructure is paying out and in fact, if they don't make the investments, they are a disadvantage relative to their competition. If you look at banking, you look at consumer-to-consumer payment systems, you look at just the way apps evolve and the consumer behavior evolves, it would be a huge mistake for anyone in the industry in their own vertical industry to make a mistake by missing out any of these trends, right? So you could be the 800-pound gorilla and you could be eaten up by a new disruptive technology. So you have to keep up. So people are investing. I don't think we've seen that change. It has almost nothing to do with COVID.

Yes. So maybe let's switch to campus. And I want to first understand your value proposition for campus. I fully understand what you're adding to cloud to enterprise data centers. In campus environment, what is the innovation? And what is the problem you're "solving"?

So when you look at enterprise infrastructure, you have users, you have apps and you have networks in between. Sometimes, the users are all just apps on compute, which is really the data center side of things. But when you have employees or devices or IoT sensors and so on, they all need to communicate, you build this network infrastructure. And then the question is how do you run it? And some of the challenges you have in the data center also apply to the campus. And in fact, for most companies, campus is not thought of to be a place where you put your best resources, the smartest architects, so thinking about how to prevent the next biggest cyber attack or thinking about how to sort of push the latency of stories down to improve performance for your company and so on. The enterprise networking teams actually struggle for campuses and campus are much more distributed because you'll have a couple of headquarter or large locations, and you have a lot of smaller offices, this discrete infrastructure that needs to be managed. And IT teams generally are stretched with the same problem. You need to patch for a security bug that came out, and now you have to patch it in 80 locations in the world and make sure everything comes back up correctly. You have an enterprise security team that's telling you, I need segmentation that goes all the way into my campus to every device. You have security teams saying monitor your IoT device is better because on the perimeter, you have firewalls on the endpoint, you have endpoint security, but in between, you actually have exposure. And if you look at the latest attacks, that's largely how people get in. They get in somewhere, they hibernate. They try to target an asset that they can compromise without getting noticed and without traversing firewalls or other devices because they will get caught. So you have plenty of things to do in that infrastructure and do it in an automated manner. All the benefits we apply to Arista, EOS and CloudVision work automatically in the campus. So to a great extent, this is a market our customers pulled us into and said, why can't I use the same product? It'll solve so many problems for me. And one by one, they pushed us and pushed us until we were convinced, and obviously, the business has been growing since then. So I think the products are very similar, at lower bandwidth. This is not a place where you worry about just bandwidth. But if you go back and are thinking, our timing truly was based on a next-gen speed and architecture that campus has needed as well because with WiFi 6, you can no longer run at 1-gig in the campus. You need minimum 2.5-gig to utilize maximum bandwidth to have all employees coming to the office for a video conference or even if you're coming from a meeting, you'll have other attendees from their homes or other offices. So for video conferencing you'd grow bandwidth quite a bit. And because of that customers think about change. When they think about change, there's good opportunity for us to go disrupt and that's largely what we're doing.

Who are the target customers for your campus product?

It's large enterprises and now expanding into, I would say, mid-range enterprise. But large enterprises is where we already have a lot of strength in the data center, good momentum. We have passed the tipping point where you're trying to prove to people and they require to have lots of preference customers and so on, right? Today, people don't even ask us those questions. They know the product will work. And as a result of that, I think large enterprises will continue to drive this.

And are they -- can you define them the meaning, are they large customers that are looking for a specific application for the campus? Is this something that you can just put through the channel or is there any unique characteristics of the customers that they're using certain applications or certain use cases, et cetera, that you address them in a more surgical way?

At the larger enterprises, the fulfillment might still be through the channel, but the discussions are a lot more detailed on security. There's a technology called EVPN, which is really an enhancement to the BGP protocol to segment users almost like a Layer 2 domain or legacy VLANs, but you're doing it using Layer 3 protocols that has a lot of synergies with new architectures that can scale in the data center as well. Those discussions could still have direct with the customer. Then there's hope for DevOps team that many large enterprises have, and we work with them on CloudVision. CloudVision can run on-prem or in the cloud. And in fact, as an example, we have a healthcare customer who really pushed me many, many years ago on, can you run this from the cloud, because they have a total of 12 -- this is -- a global 100 company, I don't know what their rank is, but they're a large pharma company -- international pharma company. And they have 12 data centers around the world, and they're a good Arista customer, but they also have about 125 offices around the world with remote hands third party is completely outsourced. The architects control the architecture and the outsourced company runs the day-to-day operation. We have to sell to the architects first, and they love monitoring and managing everything from the cloud. Just their ecosystem has changed. They actually don't need as many people on the ground that they used to before. And that was even a sort of a TCO selling factor in our discussion with them. But Tony was talking about automation earlier in the first half today, these customers are actually leveraging that already.

Got it. By the way, if there's any question from the audience, just raise your hand, we have a mic in the end, and I'll make sure that you get the mic. I want to ask about your routing and I'll ask it and you're very, very disruptive in pricing. And I remember I talked to Andy and he told me, no, even if we're extremely successful, you're not going to see it in the market share data because we're just at a different price level. So let's first start with the basics. What kind of innovation do you bring to the routing market? And what are the applications you're going after?

There's an unrelated question, a lot of people ask me, which is, which merchant silicon do you use? And why is Broadcom better or Marvell or Intel and so on? But a lot of people forget is that we're not just a consumer of silicon, we actually often go to these companies and drive the roadmap. So if you go backwards, switching and routing, the biggest differences have been the scale the number of routes and some of the features in the protocol stack. Switches used to have about 16,000 routes or MAC addresses, just plenty, 16,000 service connected to single switch. Routers used to have the full Internet routing table, which at that time was about 0.5 million routes, you need 2 copies of the table, so you need 1 million routes inside the hardware for a router. That was largely the difference. And then the routing companies added lots of features to a point where it's such a high wall that it's a huge barrier to entry for anyone else to come in. We work with the silicon players and said, can I get -- this is 2013, 256,000 routes, and then 1 million routes and then 2 million routes and then 3 million routes. And today, we support 3.5 million routes in the silicon with some unique Arista magic in EOS, the actual hardware may only support 2 million routes, but we managed to pack in 3.5 million in hardware. Now customers love that, because they no longer have to buy something that cost $20,000, $30,000 per port at 100-gig and they could just by an entire switch for that much. Now if you get completion and say, you know what, we should be charging $20,000 per port as well, then I don't think you get an opportunity to -- because we won't have all the legacy features. At the same time, we met the cloud companies on routing. The cloud companies have significant backbones. Their backbones carry as much traffic as the largest telcos in the world, maybe even more. If you look at their business models and transit services and business services that they're adding, they're actually trying to compete with the telcos in that space. They're able to do that with a little bit of SDN controller married to the programmability of our switches and routers. And as a result, we're truly able to disrupt that but that's great for the cloud. To take it to the rest of the market, we've been adding features over time as well. It's been 4, 5 years. We announced our route of product around 2018 timeframe, and they've grown well since then. It is harder even for us to measure our market share in routing because we just sell them as switches to especially the cloud customers. It's very, very hard to know how they'll use it, but we have good knowledge on where they're deployed, maybe not in real-time and we're quite happy with that outcome. I think the innovation in software, innovation in pushing merchant silicon beyond its comport zone and in fact, merchant silicon companies looking at us and saying, if Arista's asked for it then it makes sense, we should do it and make the investment because it will yield results. In other discussions, they might have ignored these and so on. And as a result, I think we're in a great place already.

And are you going after specific applications or specific use cases?

We are. I think if you look at the backbone use cases, absolutely. If you look at the peering edge, especially content companies, cloud companies or service providers and cloud companies meeting together at an exchange point like an Equinix. Those are the use cases that we do well in. If you want to build a legacy RSVP based MPLS backbone, I don't think we are ready for that.

So we only have a few minutes left and I want to ask you about supply chain because that's [Foreign Language]. I have a question that I don't know the answer. I don't think no one knows the answer I want to hear your view. Some companies are telling us we have a supply chain issue, margins get hit a little bit, but they guide up and the business continues to be strong. Arista is a great example. Other companies are telling us we have a supply chain issue, they guide down, there are issues, and it's like F5 and Cisco, examples there. How come you managed to see the same supply chain issues that your networking peers are seeing but you still managed to grow the business and have great outlook, et cetera? Is it because you are managing your supply chain better or is it because the other companies in your views are hiding real weakness under supply chain comments?

If I made the answer to the last question, I would be the biggest investor in the world.

Or a good analyst.

Or a good analyst. If you look at supply chain, supply chains have been disrupted throughout the world. There's not a single industry, not just technology, whether you go to construction, you go to food, you go to restaurants everything else been disrupted. So it's actually extremely hard to compare companies in the short term and say, how come you're having issues and you're not having issues because everyone's supply chain is actually slightly different. They're not as dependent of manufacturing in China, as an example. At the same time, we went through our own phases where we manufacture our products, we just shut down for 4 or 5 months in the first part of COVID in 2020. So we had to manage through that as well. Maybe what is different is we've been paranoid about supply since the start of COVID. And I myself have pivoted my role to focus a lot more on supply since the start of COVID versus before because you never know when you just shut down because of lack of supply. That doesn't mean we are brilliant. I think problems can occur to us as well. But what we can do is sort of try to buy ahead or try to buffer work with every supplier, see what the issues are and see if we can solve these problems. So if you look at going backwards to our purchase commitments starting Q1 of 2020, they've been growing significantly ahead of what we would need for our revenue. But that is all because of the paranoia because I would sleep better at night if I had inventory on hand, build switches versus not having anything. Again, this is not -- this does not mean we're in such a great place that we don't have to worry about it. I think problems can occur to anyone because right now, the China supply chain got disrupted, if some parallel supply chain gets disrupted same problems can occur. But because of the purchase commitments we've been making, finally, at some point, when supply does resume or gets better, I think we should be in a healthy place to meet demand.

Got it. Great. So the bad news is that we ran out of time. Thank you so much for your insights.

Absolutely. Thank you, Tal.

Thanks.