That was a great video. Welcome everybody, to the Arista Networks Fireside Chat at the Goldman Sachs Communicopia and Technology Conference. I have the privilege of introducing Todd Nightingale, who's the President and COO, and Martin Hull, who's the VP and General Manager of Cloud and AI Platforms. My name is Mike Ying, and I cover Arista and networking equipment here at Goldman Sachs. We have about 35 minutes for today's presentation. First and foremost, thank you so much to both of you for being here today. It's really a pleasure. Hi, Mike. Awesome. Hello. Great. Todd, September marks two months since you joined Arista Networks as COO and President. Now that you've been able to spend a little bit of time here, what opportunities at Arista Networks excite you the most? Where do you anticipate you'll be spending most of your time? Maybe you can talk a little bit about the video that we just saw as well. Yeah, sure. Yeah, super excited to be here. Thank you. This is my first conference, so I'm super, super interested in all the discussions we've had today. I come from a networking background. I ran the Meraki business at Cisco and the Enterprise and Cloud Business there as well, Campus, Data Center, Cloud. I guess for me, there's really, you know, two enormous opportunities in front of us. I'm sure with the video and all of the excitement today, we'll be talking a lot about AI, which is the first really big one, and the whole industry is talking about it. For us, there's also an enormous opportunity in the campus. Arista Networks, for years, has delivered truly best-in-class data center networking. Our DNA is delivering the most reliable networks in the world. It's 2025, and the time has come to bring that, like, true reliability to every network. It doesn't have to be your, you know most mission-critical data center in the world. Every network is mission-critical now, whether it's at a hospital, a government, a school, an enterprise. These are all mission-critical networks, including the one we're broadcasting on today. It's time to bring that Arista value proposition to the rest of the network. That is incredibly exciting to me. The second one would be reaching more customers. With our traditional focus in cloud and data center, we've been able to become market leaders and really focus most of our go-to-market effort on really the Global 2000. There's so much more to come when it comes to campus and the rest of that space. We're just scratching the surface of that opportunity, yet we have so much expertise, so much IP to leverage there. I'm super excited about that. Of course, I'll be focusing a large part of my time also on the manufacturing and supply chain because the demand, as the demand has been surging for so long and we see it to continue, we're focused on making sure we can deliver. That's what's exciting. That's a great overview. Maybe we can dig into those things a little bit more. Last quarter, Arista Networks raised its revenue guidance to 25% year on year. You really saw broad momentum across the entire business: AI, classic, enterprise, both on the data center and the campus side. I was wondering if you could just talk about these drivers in a little bit more detail. What's informing your confidence for the rest of the year? Where maybe there possibly be opportunities for upside as well? Maybe I'll start on the campus, and Martin will talk a little more about the AI. We've been targeting a $750 million. Campus number we raised that to $800 million. That's exciting momentum for us, not only because we're moving faster and we're reaching more of the market, but because we completed the VeloCloud acquisition. That completes the campus portfolio for us, meaning there isn't really a campus network that we can't compete for now. Wi-Fi, switching, large site, traditional routing, and now SD-WAN backed with a NAC, Network Security Solution, on the back end. With a very complete offering, we can really go after that. Raising that number and now looking forward to how we continue to drive growth, not just on the revenue side, but in customer acquisition, new logos, driving more revenue diversification, that's exciting. From an enterprise point of view, reaching new TAM, yeah, it's awesome. Yeah, and on the AI and cloud side, you know, we came into this year talking about a $750 million AI backend number. We also talked about a $750 million incremental frontend, and that is the pull-through from the additional business there. We started at the beginning of the year talking about five customers. We paired that back to four. We've reiterated that $750+ $750. In the last earnings, we talked about having 25- 30 incremental customers who are AI customers, large enterprises, cloud specialty providers, AI providers, NeoCloud, Sovereign Wealth, without disclosing names, although you might have seen some of them on the video. That gives us the confidence that the growth through this year continues, which is why we then increased the guidance for the rest of the year. We're increasingly confident about how that AI business is playing out between the largest customers and then the next set of large customers after that. Great. You know Arista is often called and viewed as a best-of-breed solution, certainly in the high-end data center networking market. EOS is commonly cited as something that's been a point of competitive differentiation for Arista. I was wondering if you could talk a little bit about how Arista EOS differentiates itself versus ODMs and white box solutions, and how does this also extend from the data center into the campus as well? Yeah, I mean, we are one of the only, if not the only, pure-play networking company now. The Arista EOS Extensible Operating System is the crown jewels. It's the heart of everything we do from a software perspective that runs on our switches, our routers, our campus products. Unlike some other companies out there, when we talk about EOS, it's the single image that runs across the data center to the AI, to the campus, to the branch. It's the same source code. It runs fundamentally the same way on everything we ship. That differentiates us, but only if it's a quality product. Having the same software everywhere doesn't help you if it's buggy or unreliable. One of the fundamentals about everything we do is quality first, and then we can add features and worry about shipping time. It has to be quality first. If you get the chance, watch the video from Kenneth Duda talking about how quality is in everything he does, from engineering down, the auto test infrastructure that we have that makes sure that we don't only test new versions of software. We test every single release of software every single day in our auto test environment. Because of that, we don't make mistakes. People don't get tired and go home. We have compute clusters running our own software doing chaos tests. That software quality then makes its way into the shipping product. I can't tell you how many times I've had a conversation with a customer and they say, "Yeah, yeah, I've heard that before from everybody else. Everybody else says their quality is the best." Hopefully, we then win that customer. I go back a few years later and say, "OK, now I believe you." It's become that reputation in the industry that our quality is the highest. That comes from software. It's EOS. Did that address the question or compared to ODMs? Yeah. Without software, we don't have anything. I'm the hardware guy, right? It's software first. The other one thing I'd add here is these two things feed off each other. There's exactly one version of EOS. That means that all of the investment that has ever been made in that test suite has been compounding upon itself for years and years. From the outside in, the validation, the verification of this stuff, it's truly amazing. That quality focus baked into the DNA of Arista Networks is real. On the hardware side of the equation, you know, what is Arista Networks doing on the hardware side or the silicon side that might contribute to your market share leadership? You're working with Broadcom as one of your merchant silicon providers. What are the advantages of doing that? Yeah, as a company, we're fully committed to merchant silicon. We're not half committed. We're fully committed to merchant silicon. Within that, we have a diverse portfolio of products using different types of silicon fit for purpose. An enterprise product, a hyperscaler scale product, we have cloud-grade routing, all using merchant silicon. A range of products based on that. We differentiate by making the right choices. We differentiate by having a detailed, in-depth architecture-level conversation with our customers about what's the right form factor, what's the right timing, what's the right combination of features that solve their problems, rather than saying, "We have a product. Whatever your problem is, our product will solve it." We have that diverse range of products. The other way of coming at that is that we do innovate. It's merchant silicon. Anybody can go to Broadcom and get the same chips. They can put them on a PCB. They can wrap it in sheet metal and ship it. What makes Arista Networks products different is the depth of knowledge we have about the silicon. We don't just take the off-the-shelf silicon with the off-the-shelf SDK. We write the drivers, the forwarding agents that integrate the silicon into our software, the abstraction layer, the programming layer. We can get functionality out of the silicon that our competitors using exactly the same chips aren't able to achieve. They can't unleash the features. They can't hit the scale. We have capabilities and innovations in our products that other people just don't have, literally the same silicon. We've got examples of that. We had internet scale routing with some telemetry capabilities and counters returning features on that weren't supposed to be there. That depth of knowledge built up over almost 20 years of experience in the silicon now stands to our ability to continue to execute as the new silicon comes to market. We can get it into a product and get it shipping at scale with reliable features. Shifting gears a little bit and focusing more on the campus side of things and enterprise within campus, there's obviously been a tremendous amount of momentum. Todd, as you mentioned, the guidance raised for this year. Why does Arista Networks have a right to win in campus? Does best-in-breed matter as much in campus where you have peers that have just an incredible amount of investment in go-to-market? I think completing the campus solution here with VeloCloud bringing SD-WAN into the portfolio gives us a full architecture. A lot of times when people talk about best-in-breed in campus networking, they would say, like, a wireless-only solution or a routing-only solution is best-in-breed. If you can deploy to the whole campus, you become an architectural win. That campus part of the market, I think we're maybe maturing beyond best-in-breed and into a complete, a really complete solution as that type of buyer would see Arista. I think that's the real power of that VeloCloud acquisition. It's like the last pin to fall. The other piece of the opportunity here is the expansion of the go-to-market. I don't mean to minimize the technology, but reaching those customers matters. With the VeloCloud acquisition, we brought in a real expertise in approaching and building businesses with managed service providers that we really barely started scratching the surface with before. We're already seeing those managed service provider players starting to find interest in the rest of the Arista portfolio and the rest of our go-to-market being able to bring that VeloCloud, that SD-WAN solution across. It's this kind of momentum and this heartbeat that I think we'll be focused on for years here, which is building out the rest of the pieces of the enterprise go-to-market to reach the rest of the TAM beyond that Global 2000, beyond the direct sales motion we've been so successful at in the last decade. Could you expand on that just a little bit more in terms of what does a full suite solution for enterprise go-to-market look like? How does SD-WAN from VeloCloud kind of fill in that gap for you all? Yeah. As far as campus networking goes, the buyer is largely going to be looking for a Wi-Fi solution, a switching solution that's going to connect all of the clients, both the wired and the wireless. That includes all the IoT devices, all the screens, et cetera, the wireless access points, and then some kind of routing. The biggest sites in the world, big campuses are going to use traditional high-capacity routing. We've been shipping that at Arista Networks for years and years. Small branch sites are likely to use technology, SD-WAN routing, which is designed to operate over broadband links, multiple broadband links, 5G, et cetera, and gives a lot more flexibility in how it's deployed. By being able to offer all of those things, all those different networking components, you wind up with a full solution. Finally, you can approach enterprises with a total solution, even if they have 1,000 tiny branch sites or 2,000, 2,500 little stores or even large stores across the world. That SD-WAN component is an important linchpin, and it's especially important in retail, hospitality, the rest of the professional services, distributed professional services verticals. We kind of earn our way into a seat at the table there. That's great. Could you talk a little bit about the go-to-market for enterprise customers? Arista's obviously had a tremendous track record with very large customers, the cloud titans. It does feel like the campus go-to-market had been less invested in, and for good reason. Where are we today? How are you thinking about having enough support from a go-to-market perspective, whether that be in your direct sales or relying on the channel to succeed? Yeah, I mean, the way this market has operated, and we believe it will continue to operate is, you know, those large cloud titans, we are deeply engaged with them directly. There's nothing that's going to change, not just at the kind of go-to-market level, at engineer-to-engineering level. In the Fortune 100 it can look similar, but it's deep, deep direct engagement. As you start walking down to, you know, mid-size and smaller organizations, the direct sales motion doesn't necessarily go away, certainly not right away, but it becomes more working in partnership with a systems integrator, with a service provider or managed service provider, or some kind of channel, indirect channel play. As you get all the way down to the market in the small business, it's really fully channel led. We have had so much success in the, you know, Global 1000, Global 2000, and that's been primarily a direct sales motion for us. We've had a lot of channel interaction even up there. We see a ton of those customers using systems integrators to fulfill for logistics, et cetera. We already have those relationships. Now, as we've started to really put together a channel program, even really just for the last 12 months, we're starting to see real traction. I'm excited to really kind of invest in that, put fuel on that fire so that we can start working our way down market. When it comes to the data center TAM, a ton of it is up market. When it comes to the campus, it's much more spread out. There's a ton more opportunity in the next 5,000 and the next 20,000 customers. Great. How does security and SASE fit into all of this? Arista's obviously announced partnerships with Palo Alto Networks and Zscaler. Why do that instead of have an in-house solution? Would you like to do that at some point? Yeah, I think Martin said it right. You know, we're a pure-play networking shop, and we take that seriously. We are going to deliver the best possible network solution. That includes the network security. We have a NAC solution, micro-segmentation, a whole host of sort of network security features and functionality, firewall, et cetera. That's critical to be built and integrated into the network. When it comes to the cloud security proxy, the SASE cloud solutions in the world, we lean into choice. We want customers to be able to choose frequently the security buyer, not the networking buyer. We want our customers to be able to plug in whether it's Zscaler or Palo Alto Networks or Netskope or something else and have the strongest possible integration into our network and certainly not try to lock them in to one particular solution that certainly won't be right for everyone. For us, being sort of that best-in-class networking solution means providing choice. Right. Maybe if I could squeeze one more in on campus before going back to data center, I was just wondering if you could shed some light into what's happening as it relates to campus networking refresh, you know, Wi-Fi 7, AI in campus networking. How meaningful are those as drivers over the next couple of years? I think any refresh triggers a campus buyer to look at their architecture decision. That's really where our opportunity comes, any time a campus operator is going to take a look and decide if they're going to continue with their existing vendor or make a change. For us, whether it's Wi-Fi 7 or a look at increased client density, especially with the explosion of IoT devices, potentially increased bandwidth with AI on the campus, we'll see how that evolves. Any time that operator takes a moment to rethink if they're going to continue or open up to competition, that is an opportunity for us. We are a share gainer in this space, so at-bats is kind of our currency. I would say Wi-Fi 7 is a real opportunity there. We're starting to see some people who are picking their head up and taking a look before they decide what vendor to use. We'll leverage that across the network, across the entire campus architecture. Great. Martin, shifting back to you, I was wondering if you could talk a little bit more about the major AI clusters and customers that you're working with. You reiterated the at least $750 million of backend switching revenue. How are those four major AI clusters progressing along? We continue to see increasing CapEx forecasts, at least from consensus across several of your major customers, Meta, Microsoft, Oracle. How does that inform how you look at demand over the next several years? Yeah, so there's two time frames to that, right? There's a Chantelle CFO perspective, right? This year, $750 million, reiterate it, stay on track. As I say, earlier in the year, we were talking about four, maybe five customers, the pace they were doing from pilot to trials to production. We've incrementally added to that number of customers. When we think about the front-end network and the back-end network, the product that we ship is the same. It's 800-gig networking products. As we get into 2026, I think we're probably going to move away from trying to differentiate front-end and back-end. It's getting harder, and I don't know that it's necessarily serving any value. We look at the year-over-year growth in that. You know, increasingly, CapEx budget increase is obviously a good thing. We have to continue to engage with those customers on the 12, 18, 24-month horizon, anticipating where they will be a year to two years from now, not just continuing to sell the same product to the same customers. We have to evolve the portfolio. We're in an 800-gig growth phase now. 800 gig has gone from nothing to a lot, and there's industry reports out there now about how fast the 800-gig adoption has grown. That's going to continue next year. We're not that far away from seeing the introduction of the next speed, 1.6T. That will be a new round of product development. We have to get those early products into those same customers' hands and start the next rounds of pilots and trials and then production. Scale-up Ethernet will come along shortly after that, maybe. It's not that far away from doing 3.2T. These speed changes, the silicon generations are coming at us faster and faster on each generation. That's largely because the appetite for GPUs, accelerators, compute is growing faster than we can keep up with just purely shipping the same boxes. We're seeing an acceleration in the technology cycles. We're seeing an acceleration in the customer demand, and that is then underpinned by the acceleration in CapEx. Chantelle's worried about this year, not worried, but she's focused on the next six months. We're looking out 18 months to two years from now. We're thinking about 800 gig going to 1.6T, going to 3.2T, and that's where we're kind of looking at the horizon of where we think we're going to be in two years' time. One thread I just wanted to pull on from comments that you made earlier is this acknowledgment that there were 25 to 30 enterprise and NeoCloud customers. That was incremental. I think it was 15 at the beginning of the year. How are those NeoCloud and enterprise customers that are doing more in AI different than the cloud titans? Are they consuming the same way? Are they more likely to use a branded solution versus a white box solution within branded? Why Arista versus a more integrated solution that comes with compute? Yeah, so they differ from the hyperscaler. The hyperscaler tends to buy and build, repeat function quarter over quarter, year over year. It's just a continued rollout. Some of the specialty providers are going to make a purchase. There might be a phase two. There might be a phase three. It's not necessarily going to be constant build like a hyperscaler would. If we think about some of the NeoClouds, maybe they're one and done because that's what their capacity is. If we look at the early entry large enterprises who are deploying AI on-prem, it's going to be a project-based technology. That's that 25 or 30 individually. They're not massive, but you start to add them up. They start to become meaningful at that point. It's why we call them out. In terms of their technology decision criteria, the NeoClouds, the cloud specialty providers have got a whole bunch of smart people who maybe have worked at other places previously. They've got the same level of technology awareness, but they maybe don't have the three years, five years to investigate, get it right. They might be making quicker buying decisions. They're going to do that based on what's available now, how quickly can I move to take advantage of this opportunity. We're going to see that kind of play into the decision criteria on branded versus not branded, Ethernet versus something else. They're going to go with what they understand, what they know, and what's available, what's proven. That's where our reputation plays into that. We're recognized as being the networking vendor of choice for AI networks, front-end and back-end. It gets us a seat at the table, gets us a voice in the room, doesn't guarantee success, but certainly we get invited. We have to execute and then win that customer opportunity. That's great. When you think about an AI cloud customer or an enterprise customer that may be doing AI on-prem and they choose a branded solution, could you just help place Arista versus a Cisco or a Juniper or a Spectrum? If you'd like, just talk about what differentiates Arista versus other branded solutions. We start with not assuming what we know the answer to be. We start with saying, what are the requirements? What are your needs? What are your application choices? How fast is this going to grow? This is networking 101 almost, discovery. We see with some of the other networking vendors out there that they pre-assume what the answer is and try and sell what their tech product is. We start with a deep engagement, understanding the requirements and trying to fit together requirements and capabilities. That's the right thing for the customer. The other aspect of that one is, again, we have a reputation. When we go into labs and trials, that quality starts to show through. We have examples within that 25 to 30 of customers that didn't necessarily choose Arista Networks first, but they chose Arista Networks the next time around. Why is that? Their experience the first time around maybe wasn't as complete as it could have been. They were minded to make a change. They looked to Arista Networks for that next choice. Those kind of aspects come into that. I don't want to name any particular competitors, but you kind of get a feel for kind of what's out there. OK, great. There was some recent press about Meta using Arista switches in a data center interconnect use case. I think it was in Ohio. Could you just maybe talk about the DCI opportunity overall? Is this a newer opportunity that has emerged? Is the go-to-market, the competition different than what may be more considered traditional backend? Yeah, so I'm saying to multiple people through the course of the day, data center interconnect isn't a new technology as such. We've been doing data center interconnect since it was 100 gig with WDM, building out a full mesh between multiple sites in metro areas. It quickly evolved to 400 gig, and now we're at 800 gig. What's changed is the customer's requirement to build multiple backend networks and then interconnect those backend networks across the data center and interconnect. The fundamental building blocks haven't changed. It's still high-speed networks, deep buffer solutions to make sure you can handle the dynamic traffic patterns. Now we're just starting to see that, oh, that's a backend DCI. Is it really different to the frontend DCI? Not necessarily, but maybe the bandwidth requirements have gone up. We've got the tools. We've got the building blocks. For us, it's data center interconnect. It's that recognition that I need to join together multiple clusters because my buildings are fixed size, fixed power, fixed cooling. I need more than that. I need to join together four buildings, six buildings, eight buildings. I better build myself a DCI mesh between those sites. It's not surprising to us. It's maybe more surprising to other people outside the industry. The public service announcement that I have for everybody is that Arista Networks is hosting an analyst day this Thursday at their headquarters. We can go straight from the conference to Arista's HQ. Could you maybe just share a little bit of a preview of what investors should expect to hear from the event? I guess relatedly, what part of the Arista story are you most excited about in the mid to long term? Yeah, you know, we're not supposed to share too much, like leave some suspense. I think it's going to be an amazing combination of discussing the business, hearing right from Jayshree and our CFO, Chantelle, on how we see the business dynamics changing. I don't want to spoil anything on that front. On the technology side, there's going to be some really interesting discussion here, especially on how the very high performance demands of the customer base, the power limitations, and where that sort of intersection is and what the next evolution is going to be. I think it really is worth a shot. I hope you all come. It's going to be great. We'll try to make it worth your while. Great. Kind of the medium to long term part of the story that you're most excited about? I mean, it's hard. Separate from the analyst day? Except for the analyst day, it's hard not to be excited about the AI momentum across the market. We're hearing that buzz top to bottom. The challenges people are seeing are, I think, small compared to the opportunity that this really has to change the world. Don't get me wrong, I'm a campus networking guy, but there's something changing about the way we build networks. This will be going on for a decade. It's awesome. Todd, Martin, it's been such a privilege and pleasure to have you on stage here. Thank you for being with us here at the conference. We really appreciate it. Thank you, Mike.