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

[indiscernible] Citi Conference. My name is Atif Malik. I cover U.S. semiconductors, semiconductor equipment and networking equipment stocks here at Citi. It's my pleasure to welcome Chantelle Breithaupt, CFO, Arista Networks as well as Ashwin Kohli, Chief Customer Officer. We also have a friendly neighborhood IR, Liz Stine, sitting at the back. Welcome, guys.

Thank you.

Thank you.

I'm going to kick it off with fireside questions first, and I'll give you an opportunity to ask your questions. [Operator Instructions] Chantelle, if I were to remind a year, I had Ita the stage in September of last year, and the big debate was InfiniBand versus Ethernet. Is Ethernet ever going to be competitive enough versus InfiniBand, and I feel like that bid has been settled now because you guys are slowly entering your outlook for this year as well as your competitors are also talking about very strong growth in Ethernet market. And third parties like Dell'Oro, they keep on raising their forecast for 2027 to $20 billion plus. So things are kind of moving in the right direction in terms of the adoption of Ethernet for AI back-end servers.

So on your last earnings call, you guys raised your guidance for the full year to 12% to 14% or greater than 14% growth now. What is driving the upside to prior expectations? And what does your outlook reflect with regards to your end market segments?

Yes, sure. Thank you. And nice to see everyone in the room and online if it's there too. So from the perspective, so we did enter fiscal year '24 with 10% to 12% guide, low double-digit guide with the expectation that there could be some moderated traditional classic cloud growth. We weren't sure coming into the year how AI would perform from a CapEx needs perspective for some of our larger customers. As we progress through the year, we saw a few things, and I've always a fan of optionality when we talk about guide, so we're not single-threaded to get to an outcome. And so we went from 10% to 12%, 12% to 14% and then at least 14% in our last earnings call. And the thought behind that, at least 14% revenue growth is a few things that we're seeing. If we take the traditional classic Cloud, we are seeing spend there, a little bit more than moderated. So we're not seeing all CapEx diverted to the AI conversation, which is great. We are continuing to see encouragement and kind of leaning into AI. It was -- it's gone from if AI something to how big is AI. So that kind of flipped in the last 9 months. And then great sort of growth in the sense of momentum in enterprise and also the traditional cloud services provider. So if you look at all of those things, they're all coming in a bit better than expected, and that's what's allowing us to get to that at least 14% growth. Part of that InfiniBand Ethernet, I think there will always be room for InfiniBand. But in the sense of having some validation that Ethernet is being seen by our customers as a choice to make, just gives us confidence generally in the Ethernet going forward for this year and the years to come. I don't know, Ashwin, if there's anything you want to add to that?

No, absolutely. Atif, once again, thanks for inviting us to the conference. I would say, last year, customers were asking me specifically and us at Arista, whether there was -- it should be an InfiniBand or Ethernet debate. I think that's been kind of put to rest this year where depending on the scale and the choice that the customers want and how efficient they want and what performance they want from the scale as well. From a technology side, they're really leaning towards Ethernet, not only for today but for the foreseeable future as well.

Great. And what assumptions for AI are included in your fiscal '24 outlook? And can you describe the path forward to your fiscal '25 target of $750 million sales for AI revenues?

Yes. And I know there are lots of AI numbers by different companies. So specifically for Arista, when we mentioned an AI revenue number, that is only related to the back-end cluster, just to be clear because there's no standard definition. So that's the Arista definition. And I would say, as we came into the year, again, back to what is AI going to be to the fact that AI is something, this $750 million revenue target for next year, it's pragmatic. We want to be sure everything comes together. The whole ecosystem has to come together, the cooling, the power, the cabling, the facility. And so all those things in the ecosystem in become together. And it's a glide path. It's not 0 in and it's $750 in '25. And somewhere in the quarters in between, we'll get to that $750 million number. A lot of it relies on when our customers will be finalized. And for them, it's a new thing. So we're trying to be timed with our customers. But between those quarters is where we'll see and we'll update every quarter on what we're seeing. And if the $750 million can be a different number going into next year, Jayshree and I will determine that going into FY '25.

Great. And maybe just one for Ashwin. Where is Arista is with this -- with regards to the ramp up its 800G products? And when do you expect 800G products to be a meaningful contributor to sales?

No, that's a great question. So back in June of this year, we actually launched several 800-gig platforms. And the goal over here, Atif, is to give our customers choice, right? And to give them not only choice but actually to give them product breadth as well. So what we launched back in June of this year. If I had to summarize that very, very simply put, customers have 3 types of choices. So you can buy either a chassis single box 576 ports of 800 gig or you can buy a 2RU switch, which is 64 ports of 800 gig. So that's a single box solution versus a single hop solution, which is, once again, we launched this in June of this year. That allows a customer to go scale to 30,000 accelerators in an architecture. Or the third choice is a multi-tier leaf and spine architecture that allows customers to go scale to hundreds of thousands of GPUs. So that's what we've launched this year. And what that does is it gives the customers a choice to go scale across different requirements, single box, single hop or multitier. But what's great about Arista is they can use the same operating system that we've been developing for the last 20 years across any solution we have across all 3 platforms.

Great. Just staying on the operating system, when I picked up the sector last year and kind of looked at you and to me, there is a similarity, what NVIDIA does on the CUDA side, which is their advantage. And you guys have a lot of software know-how that makes your pools very reliable and plug and play and ready to go. And can you help us understand the stickiness of your software ecosystem?

Yes. Yes. So if I had to summarize the differentiator Arista has in our software, I would put it down to just 3 simple words, it just works, right? And as we look at providing value to our customers, we want to make sure any use case that customers deploy in their environment, whether it is either AI or data center or routing or cloud, it should just work. And so the value that we provide in our EOS is, it's a single software that goes across multiple use cases, right? So that's the first value that I would articulate to everybody over here who's looking at Arista from a software perspective. The second thing is around how easy is it to go provision and infrastructure. How easy is it to go troubleshooting infrastructure. How easy is it to go provide visibility, security, automation, there's multiple factors over here as well. The value that comes to customers around the software, and this has been the challenge historically with many legacy vendors where networking is not an island, right? The whole point of my network is devices need to talk to each other. So traditionally, what's happened is customers would build these islands, and they would have different software, different tools to go run these islands. And the islands I'm referring to, one could be data center, one could be campus, routing, cloud. And so if you can imagine, right now if I'm any size customer, I have to have different tools, different platforms, different software, different teams in order to go run these different islands. What we've done very eloquently at Arista is to provide that consolidated view. So it's one system that does everything for all multiple use cases. That's how you go drive TCO for customers and then give them the ROI when they actually go do the life cycle management for many, many years when they deploy Arista.

If I can add too, those are the great customer benefits. I would also like to add the inside Arista sticky benefits, which, therefore, help the shareholders and investors is that when you have one elegant software that you're using, it's only one operating system we're developing and one that we're supporting. So all 4,200 people wake up every day with one operating system software to support, right? So it's very clear from a capital allocation. I'm investing in this one because it's the only one we use. So it's very clear from an investment perspective. And on the support side, and you have some great customer NPS scores to share. It's also very easy to support because it's one system operating software we're supporting. So it's also quite financially sticky from a running the business perspective. I think that's the thing we don't talk about enough but it makes it a very nice business model that way.

Absolutely right. And if I add to what Chantelle said, the goal is about allowing customers not to be locked into specific software technology, right? And so if we can provide API integrations to customers, whether they want to use other vendors or different technologies, that's the value we actually use it with software as well, right? Very key.

Great. Maybe just staying on the 800G to 1.60 ramp and even 3.2%. I mean you are a very close partner with Broadcom that is providing silicon to you guys. Can you talk about the alignment of your road map versus the GPU road map in terms of proliferation of the stacks that the GPU makers are talking about and their timing at 1.60. And if there's any risk to -- according to our work, we think the road maps are quite aligned right now. But what is the risk of kind of kind of falling behind the industry road map?

You want to take that, Chantelle.

You can take that.

Yes. So listen, there's always risk over here, right? And any time you're upgrading technology, you always want to make sure you're doing this methodically and thoughtfully as well, right? That's the first one that I would say. The great thing that Arista -- that we do is we want to make sure that not only are we looking at newer technology, we're actually partnering with our customers very, very closely as well, right? That -- I want to emphasize on that, that is the core of Arista culture. So rather than us working behind the scenes inside of vacuum and developing our own technology that we think customers might want the way Arista culture works is, we are sitting with partners and with our customers, with them trying to go figure out how do we go build the next set of road maps jointly together. I can't emphasize that enough, right? And so what that means is our engineering teams, the customers' engineering teams are sitting down and trying to go figure out what is the best software, all the best hardware for their use cases as well. That will drive innovation and technology, right? You can pull that up, Chantelle.

No, I think that's spot on.

And then just on the competition, it has talked about a few billion dollars in terms of their SpectrumX Ethernet ramp publicly, and we don't know exactly what's in it. But -- and obviously, growing from a very small base, how do you see the competitive landscape? Traditionally, you've had Ciscos and Junipers of the world, but oil a bit of a different kind of a competitor to you. So can you just help us understand how you see the competitive landscape moving forward?

Yes. I think -- I can start and then Ashwin can add a bit. I think that now it's a bit segmented, right? I think you alluded to, it used to be us and Cisco, and it's a changing landscape for us. So I think when it comes to ourselves and NVIDIA, we're very happy to partner with them on the mics and with customers to make sure customers are getting what they need. So very much enjoy the partnership and cooperation on that perspective but also very aware, we will compete on the networking perspective. And I think that from the size of the TAM numbers for AI, if they're even close to true, there's enough room for 2 vendors for a while. So I think we recognize that, and I think most customers would want multi-vendor optionality. So that makes sense, and we're very excited to get our fair share of that market. Whatever that TAM ends up being. There are many numbers out there on what the TAM is. And I think on the -- I would say the enterprise side, which is more where we're seeing the traditional Cisco, Juniper conversations. I think there, we're pleasantly surprised in the sense of the kind of the conversations we're having in regard to us being a pure-play networking company, our road map, our trend on execution, the one EOS operating system. So I think there, we're getting some tailwinds from some of the M&A activity that's happening around us that perhaps make the road maps a little less clear to the customers in the enterprise space. So we're excited about that, and we'll take the opportunities we can there and keep our heads down focused on making great products and expanding the portfolio for the campus side of the enterprise specifically.

Great. And then Ashwin, you guys have talked about being 4 out of the 5 hyperscalers you're engaged with. And is your view that those hyperscalers are going to make kind of the best of the breeds decision in terms of how they build their clusters?

Yes. Yes, absolutely, right. So I think you're talking about the 4 to 5 for the AI specifically around that. Yes, absolutely, right? So I don't think we've actually just gated only 4 to 5, right, Chantelle can correct me if I am wrong. There's a mix of other kind of Tier 2 cloud providers inside that as well. So yes, there's a competition -- there's a mix of both around that as well. Yes.

And then concentration risk historically has been a topic with investors with Meta and Microsoft being larger customers but you guys have diversified to an article, a Tier 2 and the other cloud service providers as well. So how do we see your diversification across the cloud to the AI titans?

Well, I think that -- so first of all, we're very grateful for those customers because Arista needed them in our journey. So always thankful for them. I think in the sense of the percentage of our overall business as the denominator gets bigger, it becomes less of a concentration unless they grow as fast as we grow as a company, then it remains the same. But I think from that perspective, we're pleased to keep working with the titans. Sometimes we get conversations on the Amazon and the Googles, and we'll continue to see if there's ever opportunity there. Always happy to help if that becomes an option. But I think we enjoy our relationships. We have great partnerships, great engineering discussions. So we'll continue that, whether it's traditional cloud, enterprise in the cloud customers or AI. But as we expand the other areas, grow the enterprise, grow the Tier 2, it just diversifies in geographic expansion in industry or vertical expansion, land and expand with our current accounts in the enterprise. There's a lot to go get to diversify. So we're excited about that but always appreciative of those customers.

All right. Let's talk about enterprise. I mean, it is not much of a debate that you guys are very strong on the cloud side and AI. But what are the key areas, I mean routing and other areas in enterprise that differentiates Arista and how much headroom is left to gain more market share in that market?

Well, I'll start with where we are from a market share perspective and then maybe you can speak to the product, Ashwin. So we estimate based on count of logos in the Fortune 500, Global 2000, we're maybe 25% penetrated on logo. So there's a lot to go look after, and we're very excited about that. That's part of what the investment we're doing in a direct sales team with [indiscernible] is the sales leader. So that's in the sense of where we're starting from. So lots to go get. And then if you want to speak to the portfolio, Ashwin?

Yes. I mean when it comes to enterprise customers, once again, it comes back down to what they're looking for is best-in-class, best-in-breed technology, right? That's really, really key. Most enterprise customers do not want to be locked into a specific vendor or a specific technology. They want something, which is open. They want something, once again, it just works, right? That's the kind of mentality enterprise has, depending on whether it's any vertical inside the enterprise. So our view at Arista is to basically provide a breadth of solutions for enterprise customers to tackle multiple use cases. And the multiple use cases are a data center. So let's say, for example, you're an enterprise customer looking to go deploy a new data center or upgrade an existing data center, maybe you have end-of-sale, end of life equipment up there. That's when my team gets engaged with the enterprise customers. You may have an enterprise customer saying, okay, I've done a data center but I might want to go to colo, which I can go connecting to the cloud. So we -- the teams helped us in the second use case. There could be a third use case, which is campus, right? That is -- there's a lot of large customers out there, specifically enterprise customers and say, well, I've got a number of large campuses around the world, I've gallon large banks branches around the world over here, how can Arista help me connect data center to campus and then can you provide a routing solution as well. And to be able to grow all these things together, they're looking at my team to be the trusted adviser to come in there and actually give them a whole solution in order to go drive value for them, right? That's really key for the enterprise customers.

Okay. And there has been some significant consolidation in the networking sector with Juniper and HP. Can you just talk about the impact of the consolidation on Arista, a potential impact?

I think, we all speak to potential because I think we're just getting started. I think -- so anecdotally, what we're seeing is back to -- let's take maybe a few years ago, there was a customer we were trying to get into as Arista was going through its growth journey. And there was an incumbent and it was thank you, but no thanks we've got our incumbent and we're good. Now we're seeing some we're getting called back actually getting the calls from those customers saying, hey, we'd be really interested in speaking with you because the incumbent, we're not as clear where the incumbent is going. You guys are just focused on networking and have the OS, and I've heard this. So that's really great. So I think it's affording us conversations as of to your question, that we may not have had earlier or we would have been trying to break into but they're coming to us earlier than us finding them necessarily. Sometimes it's just down to the refresh cycle. Now if there happens to be a refresh cycle with a new logo, we have an opportunity because now they're looking at the choices and now we see as a good choice for them to take a look at.

Yes. I mean I would say -- I'll add to what Chantelle said over here, I would argue, if you look at us compared to all our competitors out there, we're probably the last best-in-class pure networking vendor out there, right, that provides customers the breadth of solutions for any use case in the data center as well. So that as in -- that is definitely an eye-opener. And today, Arista has a brand recognition globally, right? So customers view us as the #1 player in 100 gig and above. They view us as a #1 player in high-speed data center switching. Once we -- once the team goes in there and actually provides the value from there, then, let's say, for example, like an executive at an enterprise customer is starting to think about, okay, should I move to my incumbent vendor or a vendor who is actually consolidated over there now may not be the right time to go with that vendor. Let me go look at the best-in-class choice over there, and then that's what the conversations pivot to Arista.

Great. Let me pause here and see if there are questions in the audience. If you have a question, please raise your hand.

I think earlier you mentioned 3 types of solutions for AI, I think boxed kind of single hub and then leaf spine. Can you may be pull the curtain and discuss engagement across those 3 where are you maybe seeing more engagement? And then the second part of the question is kind of -- I think we have seen the debate around open approach versus the more kind of just a vertical approach and on a more -- on open approach, one of the benefits is you have more vendors and cost. And on the kind of more vertical approach, it maybe have faster speed to market -- market speed. So I guess can you discuss on, particularly your open approach, is there any risk of you being ready but then missing a couple of pieces here in there? How do you make sure that you are kind of on the speed portion of getting to market you are out there with your...

Yes. Okay. So I'll take the first question first, right? So I think your question is who is using those different types of use cases. So just to summarize, the use cases could be single box, single hop or multi-tier solutions out there. And I'll be honest with you, depending on the customer, it's very, very different, right? And so originally, that if a customer is just trying out the use cases, then they may start with a single box solution. They just want to try out, okay, is my AI application going to work? So it could be a single box solution from there. Then they will say, if you find my application works over here. They've done a little bit of training. Let me go expand that use case. And then the goal would be is, okay, do I want to go deploy a few hundred GPUs or tens of thousands of GPUs or 100,000-plus GPUs? That is a big differentiator between the single hop and the multi-hop solution over there. So the factors that you want to think about or people think about over there, and this is what Chantelle talked about before is power, space, cooling cabling, right, those physical environments always come into play when you're actually designing a single-box, single-hop or multi-tier from there, right? So hopefully, that answered your first question. The second question is how -- I think your question was how was Arista open providing the value of openness over here for my integration to customers? I would say, listen, we're entering a new phase. This is a shift in the market over here. In a sector, it's a new use case. It's going to transition over many, many years. The last thing we want to do is be either ahead of the curve or to behind the curve. So Arista's view is best-in-class know what we want, right, know what we do, we're really, really well and open with -- openly partner with as many vendors out there, right? And so we've got great partnerships with NVIDIA on the accelerators, right? But if I'm a customer, I just want to be locked into a single accelerator. I may want to try multiple choices out there. Arista's goal is to go help them, go do that with not only a single vendor or multiple vendors from an accelerator perspective. A customer may say, okay, around, let's say, for example, connectivity to another data center or routing. Our goal is to write open APIs and Ken Duda, our founder at Arista has talked many, many times about NetDL, and the goal is to try to go figure out how can we take the knowledge that we collect from a clusters, either data center campus, AI routing, and can we open that very, very openly to APIs that we can interface with other vendors, so that -- that drives value for customers so that they can take the spend that they've done in the infrastructure and drive a TCO and ROI from there. And many times, customers are only focused on day 0 of the deployment, but it's always a life cycle of how that infrastructure is deployed in many, many years. That's what most customers need to think about, right? And so that openness over years 2, 3, 4 and 5 become very, very critical as well as the customers use and consume that technology.

You had a very large increase in your aggregate deferred revenue balance in Q2. And it's an unusual time of the year for you to have one of those. And I think what I've heard you say is it's largely like normal course type of stuff like don't get too excited or like read into this. So can you just remind us like why did that happen? And why isn't it numerically significant as we think about the next 12 to 18 months?

Well, I think there are a few things. So the one and not to be facetious, it will eventually come back to the P&L, right? So a lot of the question is, what's the timing. And the timing is different for every use case of customer. So that's in the sense of it will come back over time. That's what the deferred revenue does. I think that what I'm trying to caution is people tie it to one specific AI outcome, deferred. Deferred has been there before AI, and it will be there after AI, if there's an after AI phase. And so I think it's a combination of things. I think for the size of it, in my prepared remarks, I said as we grow as a company, that balance will continue to grow both in variability and size just by the nature of the size of the company and the size of the customers or the volume of customers. So for me, I think it's the same as before, but just as a bigger example of how big the company is getting or the specific use cases that are being used and have to show out to be in the deferred to come back to the P&L. Yes.

Chantelle, what is your appetite for M&A? And can you just remind us what your top capital allocation priorities are?

Yes, sure. I'll start with the capital allocation model, then I'll go to M&A because you'll see why in a second. So I think from a capital allocation perspective, I would say in the sense of not much has changed from Ita retiring to me coming in, and I can understand why Ita had this view working with the company. So I would say, from my view, it is organic investment. We talked about all the enterprise investment, the go-to-market, R&D is always an investment for innovation. There is the -- in the sense of investing, we're still getting quite a good interest income from investments. We're talking about share repurchasing. And those are kind of the top 3. The next one would be M&A, but tuck-in M&A, certain things that fill in the part of the portfolio that we're looking for. But the one thing from an Arista perspective, which I think we need to try to ensure we always keep to some degree is the fact that all 4,200 people wake up and do networking versus trying to grow horizontally where it becomes confusing where our investment and time and energy should be spent. So we won't be doing large M&A just for the sake of growth. We'll do M&A in the sense of it bolsters what we do today and what differentiates us from our competitors, which is an elegant business model, very clear, allows us the margins we do. Because the other thing that gives us this really great margin profile, which could be impacted by M&A and besides our road map becoming confusing, which we don't want is that because of the leverage of the cloud sales model and the elegant service and support that Ashwin and his team run, we're not fixing the product. We're not supporting the product after it's installed, and that gives us a lot of margin latitude. And if you start to do M&A and that starts to impair that, that could be growth at margin, not so smart. And so I think from that perspective, we just want to be really clear and curated on M&A. And so that's -- those are my thoughts on M&A. It would have to be very specific and support what gives us our differentiation today.

Just a follow-up question on the deferred revenue increase. Is that broad-based? Or is that related to just one customer?

It's not one customer.

And a bigger picture question on the AI side of things again. When we think about these GPU clusters scaling, how should we think about the networking value portion of the pie? I think Broadcom has said that this mix is moving in favor of networking as opposed to the actual sort of physical GPU. If you could comment on that. And then perhaps sort of talk about hundreds of thousands of GPU clusters. Are there limits to this? Can you discuss what your sort of customer road maps are implying and over what time frame you're expecting?

Yes. I think I'll take the first question.

Yes, you take the first one.

I think it's always enjoyable for us when other people put out big numbers and then we have to explain them. So I think that from the perspective of, you mentioned Broadcom to your question, but there are other numbers out there. It just comes down to what's included in the number. What does it actually mean at the end of the day for Arista. So taking the Broadcom example you brought up, completely happy for hop turn in the Broadcom. But there's many things in there. If you could segment in OEM Arista, I think China is a market, et cetera, or what they put into that number, nothing's really changed in the relative outcome in the sense of that relationship. So I think that I would leave that there. Sometimes we get compared to other companies that just have different things that are included. So I would just -- it comes down to the devils in the details of what that number is actually representing. We work very well with Broadcom, a great partner of ours. So nothing's changed in how that relationship will work. And if you want to talk about maybe the 1 million cluster.

Yes. I mean if you take a step back, last year, people who are doing trials on their AI models. And so I would have said that would have gone from almost hundreds of GPUs, right? That was the early transition -- early bake off and then you would have had pilots where you're going from hundreds of GPUs to maybe thousands, if not 10,000-plus GPUs, right? So that was the next logical phase. And now customers have actually worked with us that we want to go from tens of thousands of GPUs to up to 100,000 GPUs, right? So that's where we scale and that's what we've launched out in June of this year, products that are aligned around that. The next aspirational goal has been -- we don't know as yet. There's nobody who has come in and said anything around go from 100,000 to 1 million GPUs as well, right? And so in order to go do that outside the networking, there's going to be a lot more limiting factors around there. How do you go, how is power space, cooling, everything I said before. So I think there will be a lot of physical limitations before you get there, but there is no reason why when we sit down with customers and we can jointly partner together, figure out the use case, there's nothing that actually stops Arista from scaling to from 100,000 to a few hundred thousand up to 1 million GPUs, right? That's only time and it's training and it's collaboration, working with their customers.

Just looking at the size of the opportunity that you've outlined for your AI versus the biggest competitor out there and then the number that they are kind of reporting. Given the constraint that they are facing, I certainly understand the physical limitation that you have outlined. But given that they're doing, going through a product change, upgrade to the next generation and given how constrained they are, does that present an opportunity for you to actually gain some share at the expense of them?

Do you want to take that, Chantelle, first?

Well, I think the -- so I go back to in the sense of what numbers are being compared for our $750 million for next year, it is only the back-end cluster. We're being very clear, curated to just -- because there could be AI spend on the front end that, that looks the same to us so we can't really differentiate that. So I think that just in the sense of apples-to-apples, it may not be that compared to other numbers. But in the sense of the competitive possibilities, market share, maybe Ashwin, can comment on it.

Yes, I mean I would break that down probably 2 ways of thinking about this, right, in terms of how do you go partner around that itself, right? So scale is going to be a big factor, right? Performance, speed, all those things are going to be a big factor, try to figure out what is the competitive nature around there? And can we go break into other sides. I'll keep coming back down to what do customers want, right? So they want choice, right? They want breadth of portfolio. And I know it sound repetitive over here but that's been our experience with customers, right? And so generally, they want to work with best-in-class, right, work with the brand recognized name that's delivered value, not for the last 5 years, 10 years but last 15, 20 years in this space, try to go figure out, okay, depending on their use cases, how can we go drive value for multiple use cases, right? And so we could have very easily back in June of this year, delivered just one single box and said, "Oh, my God, we've got 1, 2RU does 64 by 800 gig. We did not do that, right? We came out with our first announcement that multiple use cases of 800 gig. So that sets the mark for any customer to be able to use the next-gen technology to do -- to work around that as well, right? And then listen, I always view competition is healthy. We welcome competition every single time as well. It makes everybody better. And so the market is big enough to go for us to go partner with incumbent vendors or -- and that is a value to go in to operate with them as well because once again, whenever customers do call my teams, they want the best world-class support out there for those teams and that support, which may not get recognized day 1 when you're looking at just GPUs really works when you're supporting that day 2, day 3 and an operational infrastructure as well, right? That comes in the competitive space, right? So it's all good news from where we sit at Arista.

Great. We're almost out of time. Chantelle and Ashwin, thank you for coming to the Citi Conference.

Thank you so much. Have a great day.

Thank you so much. Thanks for having us.