I guess I can kick it off. Okay. Great. Good, good morning, everyone. I'm Samik Chatterjee, and thank you everyone for coming to our Hardware and Semis Corporate Access Forum this year. This year we have Juniper kicking it off. Rami, thank you for taking the time to come here. Right. Obviously, I think all of you know Rami, as the CEO of Juniper, so I don't need to introduce Rami as such. Obviously, we have a 50-minute fireside chat and want you all to be getting your questions in as well. I'll kick it off, go through a few questions on our side and then start opening it up, but please feel free to jump in with your questions, and we definitely want to have an interactive session as much as possible. Rami, thank you for the time again. Let me start off. I do want to start off with longer term trends. We can always spend a lot of time talking about the cycle- Mm-hmm. Near term, biggest trends across 3 customer verticals you have exposure to, how do you think Juniper is positioned to leverage? What are those biggest trends you're seeing? Does it change Juniper's growth opportunities in those verticals? Okay, thanks, Samik, thanks, everyone, for joining today. Appreciate you hosting us. Before I start, I just want to remind you that I might be making forward-looking statements in my comments today. I just direct you to our website for our risk factors. Trends. Let me start in the enterprise. It's easier to talk about this by market verticals. In the enterprise, I think every CIO conversation I have these days is around some sort of digital transformation, a recognition that the network can't just be best effort anymore. If there's anything that the COVID has taught us over the last few years, is that network is critical to pretty much any digital application, any digital transformation project that's being executed in a large enterprise. That is driving a need to invest and upgrade IT infrastructure to deal with applications that sit in the cloud, in private cloud, in public cloud, and so forth. Another trend that I definitely am seeing in the enterprise is, several years ago, I think there was almost like an over-rotation to public cloud. Everybody moving to public cloud was going to be a thing. 100% of workloads will eventually move to public cloud. I think there is much. There is a much more balanced, thoughtful view of applications and workloads being in both private and public cloud. This, of course, is leading to a need for easy automation of private and hybrid cloud networking. In the cloud space, AI, which I'm sure you're going to be asking me about, is definitely top of mind for most cloud providers, not just, hyperscale cloud providers, but you name it, what we call Cloud Majors, top 10. The long tail of cloud providers are all thinking about it. It is not consuming every cycle from these cloud providers. They definitely are also thinking about 400G upgrades, 800G upgrades, but AI, I would say, is a key long-term trend. In the service provider space, 400G convergence of networking across different use cases, 5G and private 5G in particular, I would say, is a key top-of-mind trend. The summary I'll have in terms of, because your question was around sort of the Juniper opportunity, I think we're very well positioned for all of the above. This is, in my view, very good news for anybody running a networking company, especially anybody that has the kind of technology, the engagement, the solutions that we have at Juniper. Okay. Okay. Let's start digging into some of that, including starting with where you started, enterprise. I think the primary question everyone keeps asking is how to think about sustainability of the double-digit growth that you're seeing in this business. What are the drivers that would drive enterprise customers to transform their networks, still continue to transform it even after they've done it post-pandemic, to some extent? Yeah. Again, sort of talk about where all Juniper can participate in that transformation. I am very bullish about our enterprise business and our prospects, and our ability to continue to grow this business and to grow it meaningfully. We have done well despite headwinds that have existed, headwinds such as digestion that is happening right now. We continue to grow not just revenue, but orders. In the onset of the pandemic, when some of our peers were experiencing declines, we were seeing continued growth. Also during the pandemic, essentially, any investment in campus, carpeted enterprise-type opportunities went away, and we continued to do very well. I think this is a function of the strength of our portfolio, the innovation, the fact that we're addressing pain points that IT teams feel today. One is around digital transformation, as I mentioned. The other one is around just OpEx crunch. You know, budgets are not exploding in a way that IT professionals would like them to. They are looking for solutions that can make them far more cost efficient. If we're offering a solution that can essentially eliminate trouble tickets, reduce the complexity of running the day-to-day, you know, operations of a network, it's music to their ears. Our solutions are really resonating with the trends that are happening today, and I expect to continue to happen over the next few years. For that reason, you know, I think we're going to continue to do well. We can, we can achieve continued long-term, sustainable growth in the segment of our business or the vertical of our business, where it's the largest and the fastest growing right now. Okay. Okay. Moving to Mist specifically, I think most investors perceive it to be early phase of growth for Mist, Wi-Fi access points in particular. How does the differentiation hold when you start to move beyond Wi-Fi access points into other campus products? How should we think about like, investors clearly think of it as early phase, and that's driving some of the growth that you're seeing, but as you start to reach more mature scale, the growth starts to slow down. How are you thinking about sort of the runway to Mist growth for Mist and Mistified products? Yeah. Our Mistified revenue in the Q2 period grew 100% year over year, right? This is, any, anything that we sell that sort of is under the Mist AI operations cloud-delivered umbrella, doing phenomenally well. I expect it to continue to do phenomenally well. This is not just about Wi-Fi, so I'm really glad you're asking this question. This is sometimes a little bit of a confusion. It is true, when we acquired Mist, it was a Wi-Fi AIO ps company. When we acquired them, day one, we saw the potential to make this into an end-to-end enterprise architecture that includes all aspects of the enterprise connectivity problem, all the way from wherever the client is through the wide area network, the local, LAN switching network, the local Wi-Fi network, elements of security to the cloud. That thesis is proving out each and every day. For every $1 of wireless, there are actually $2-$3 of wired switching opportunity for us to pursue. That's not the ratio of our business right now. However, the wired switching component of the Mistified business is growing very rapidly, and over time, it will start to align more closely to that $1 of Wi-Fi to $2-$3 of wired. We also had in Q2, a record SD-WAN quarter. That's primarily driven by the fact that we have now technically linked our SD-WAN solution through the 128 acquisition with the Mist AIO ps umbrella. We introduced a NAC solution, which is essentially a security solution that's next generation, cloud-based, again, highly integrated. It becomes essentially a feature of the Mist enterprise architecture. Anywhere we sprinkle that Mist pixie dust on our solutions, we see growth. I expect that for that to continue to happen. You talked about SD-WAN, and I was actually going to ask you specific to that, but maybe I, I'll broaden my question. Investors have seen Juniper make acquisitions to bolster the enterprise vertical, so the portfolio that you take to that vertical. Yeah. One way, technology being one. When you think about sort of... Obviously, you don't provide as much of a regular update on what those businesses are doing within the portfolio, but how have you seen the synergies come through in sort of helping you with enterprise share gains more and more as how should we think about the success of those acquisitions that went into the enterprise portfolio? Yeah. It's a great question. When we are looking for acquisitions, as you know, before we made 128 Technology acquisition, Apstra, for the data center, WiteSand for NAC, I'm looking for a few key attributes. One is, yes, it has to be highly relevant to the enterprise, although it's very much applicable to other verticals, highly relevant to the enterprise is definitely an important factor. Software, to the extent possible, like entirely or mostly software, because we're looking at a software control point sales motion to differentiate from our peers in the industry. A high degree of telemetry. Mist loves data. One of the things that we love about 128 Technology is, it really gives us granular data about the experience of the end user, and that data can be aggregated and fed up into the Mist cloud for AI operations to learn, develop insights, and ultimately to conduct closed-loop automations. That's the elements of the solution right now that are working incredibly well for us. As I mentioned, our wired switching business has essentially been, I don't know, prior to Mist, flatlined for years, is now taking off. It's taking off because of Mist. Our record SD-WAN performance in Q2 is because of Mist. The interest right now in NAC, our Network Access Control product, which is really just introduced into the market, is because of Mist. everything we had hoped for, as a result of this AIO ps end-to-end solution synergies, has not just met, exceeded my expectations. Okay. Yeah. Staying to the long term, before we bring it back to more near-term questions on the enterprise, when you think medium term, how does your enterprise business look, let's say, five years from today? What, what mix should we expect between Wi-Fi, campus switches, SD-WAN, or do you see other sort of legs for that campus, vertical? Yeah, I sort of touched on this. I'll just reiterate. From an opportunity standpoint, for every $1 of Wi-Fi, $2-$3 of wired and also some attach of SD-WAN is the opportunity. Although we haven't broken out the precise ratio of wireless to wired, maybe at some point we can do that, I can tell you that it is not one to two to three. The wired component is still way less than the opportunity that is available to us. However, the wired component is also very rapidly growing, and so this to me is, again, music to my ears. What we do is we track the percentage of wired switching that has a Mist attach control point sold with a Mist licenses. That's growing extremely fast. Again, it just, I think, speaks to the fact that the differentiation that we're offering is not so much in the hardware of our switching portfolio, which honestly is like hardware starting to look very similar between different vendors. It's the software control point. Same thing I, can be said about NAC. It's very small now, but most Wi-Fi opportunities are sold, in conjunction with a NAC solution. Until now, we've only had the ability to position third-party NAC. Now we have an end-to-end modern NAC solution that we can offer that can be, again, another kicker to that solution. I mean, I don't wanna disclose too much here, but just know that we're investing and innovating to continue to add more and more capabilities to this solution, so that we just have more features, more widgets, more capabilities that we can cross-sell and upsell our customers into with that solution. Yeah. Can I ask a question, Rami? When you sell Wi-Fi products, you were clear two years ago that you felt like a full switching, right? As you add more feature functionality, you're suggesting that first you start with a Wi-Fi first approach, and then the cadence of when they might replace the switching might take time, right? Yep. People with that, right? You're putting in this land and expand with Wi-Fi, and over time, you hope to sell the switching later, or is it more of a bundle of approach where you're selling all at once? It's a great question. I'm not sure if people can pick up the question. It's basically a question around sort of how do we go and pursue cross-sell from wireless to wired? Is it sort of an all at once type of sales proposition, or do you start with one and go to the other? In many cases, we start with one and go to the other. However, I am specifically tracking what we call full stack opportunities, meaning that I am selling one shop from the beginning, a solution that includes a number of these end-to-end components, because I believe that over time, the number of CIOs and IT professionals that want to purchase full stack and deal with fewer vendors in that end-to-end path will only increase. In many cases, we'll start with Wi-Fi and then move to wired. In some cases, we're now seeing start with wired and move to Wi-Fi, right? In Q2, I think I mentioned on our earnings call, we had a record number of full stack deals, meaning these are customers and deals that have purchased a number of these components all at once, which is wonderful because that's when our differentiation truly shines, right? When you can control and have visibility and assure that end-to-end all the way from client to cloud, it's a great value proposition for customers. [audio distortion] Absolutely. [audio distortion] Absolutely, yes. [audio distortion] I have two on that. One is, you know, obviously, the, the Mist acquisition, the execution of the post-acquisition, been a tremendous success. Clearly, you're sitting here and telling us a lot of drag along to it. It's not a subscale product, so it's, it's really meaningful on a quarterly basis. Yet, if I think about the forward, you know, we're, we're staring down the barrel of year-over-year decline. So the two-part question is, one, how should we think about the getting the angle of Mist and the Mist ecosystem working through your business versus everything else- Mm-hmm. [audio distortion] Yep. How you see that out? The other thing is, you know, we're sitting here talking about the enterprise opportunity. How do you take this and then work that through, if at all, the carrier opportunity? Because obviously, that's a bucket that traditionally is, you know, did not be so different growth trajectory and is currently hurting you right now, more so than. That plus hyperscaler hurting you. You know, how do you think through not just the enterprise opportunity in this where, again, tremendous success- Yep. Working that through the rest, so it can offset some of the pressures that come and go with these businesses volumes? Yeah, sure. First, it is already offsetting in some respects. It's helping. Our enterprise business is our largest, now over 45% of our revenues in Q2, and our fastest growing. If I play this out, this year, enterprise orders and revenue are gonna grow. Next year, enterprise orders and revenue are gonna grow. This despite the fact that there is some digestion that's happening in the enterprise vertical, it's just that the momentum, the strength of our business is so strong, I think we can overcome them. It's a. You're right, it's a bit of a different story in SP and cloud. In service provider, orders are gonna decline this year. Orders will recover next year, right? The demand is gonna be there for SP relative to this year. Same thing can be said for cloud. Orders were declined this year because of the backlog dynamics, the backlog draw. Orders will recover next year. The thing about SP and cloud is next year is it's a revenue compare challenge. This year, there was a lot of backlog draw in cloud and revenue. We're not gonna have as much of that next year, so it becomes a difficult revenue compare. That's not a reflection of demand, that's just a reflection of the year-over-year revenue compares. Enterprise will absolutely help us offset some of these revenue compare challenges for next year. That's kind of the way I would look at it. Now, as I get into the out years beyond 2024, I think there's going to be much more normalization between revenue and orders across SP and cloud. Couple that with the continued strength that I expect to see in the enterprise, I'm very confident in long-term, sustainable revenue growth and operating margin expansion. As you think about the longer term trajectory of both, both wired and wireless, but for a long time, these were deflationary market. I mean, tech has been a deflationary market in terms of, you know, cost per bit falling over time. The past couple of years have been quite inflationary from a component standpoint, all the way up to devices, both wired and wireless. When, when you think about the next two-five years, what's your expectation for market pricing within wireless LAN? Within that, how you think that the points of differentiation from an automation standpoint kind of help get the sale over the finish line, even in instances where, you know, larger competitors might be more price aggressive, [audio distortion] Sure. I think pricing has always been a factor of competition in the enterprise space with Wi-Fi, with really Wi-Fi wired, all the elements of the different solution. I don't necessarily expect that to change, either get worse or better in the next few years. Honestly, I don't see any trends now that would indicate any meaningful change. However, you touched on something that I think is really important, which is that today, IT teams recognize that the biggest spend is in OpEx, and if you can help them optimize that, they are willing to pay more for the equipment and for software services. Time and time again, I think we've proven that. Like, I know there are a lot of people that have jumped onto the AI bandwagon and that claim AI, but if you don't architect a solution from the beginning, ground up, to consume data and at scale and do something useful with that data in a closed loop matter, it's very difficult to retrofit a legacy solution to do that. Which is why the results with our customers, like, speak for themselves. We're reducing trouble tickets at companies around the world by way over 90%. I mean, think about what that means to an IT person that spends all of his or her time just keeping the lights on, the proverbial lights on in networking. They are now free to do far more important things for their organizations, not to mention the fact that they become heroes of their organization. I actually think the solution capabilities here and the differentiation allows us, in fact, to alleviate some of the pressures on equipment and software pricing in the solution that we sell up front. I'm assuming that especially in the maybe last six-1 2 months, as to your point, enterprise budgets have been tighter than, than not. Yeah. [audio distortion] on the enterprise side have really pushed that point home with customers around how much you guys can save from an OpEx standpoint. When your sales teams frame that, like, what's the typical payback period or TCO comparison that you guys are quoting now on the wireless-? Yeah. [audio distortion] You're absolutely right. TCO, and the TCO argument is very much part, part of the sales proposition, and it's very much real. Like, the payback period can vary. Maybe it can be two years, three years, depending on the deployment. There are other things that I think our sales team has done remarkably well. One is to focus on, in, on vertical segments of the enterprise market, where there's more of, let's say, a resilience to what's happening in the world. When carpeted enterprises weren't a thing, we focused on healthcare and retail and government. We started to make progress in government certifications and started to win government deals as a result of that. These were all areas that were more immune to some of the dynamics that were happening in carpeted enterprise. In carpeted enterprise, I know people are thinking, "Well, now, you know, nobody's going to the office anymore." Actually, what we're finding is that people are returning to the office. In some cases, the offices are smaller, but even that in and of itself is a good trend, because if you're moving to a smaller office, you need to invest in IT, you need to invest in Wi-Fi, you need to invest in wired switching, and that is presenting opportunities for us. So- I mean, this is a bigger picture question. For the last three years, you made the point, you know, next year's revenue and demand are not that [audio distortion]. Mm-hmm. Last year has been a very unusual period for the industry, where orders are great, revenues are awful. Yeah. Orders are awful, revenues are great. I mean, how are you judging the underlying spend positions? Because they're disconnected from the usual metrics. Yeah. we look at. The reason I ask that is enterprise, for example, listen to the software guys who are more concurrent with funding demands. They all talk broadly about slowing enterprise down in everybody. Mm-hmm. Yet, you could look at networking, cloud, and service provider, I mean, weaker, but enterprise has been the one area that's strong. Yeah. For you, I understand the share shifts, but I don't understand how do you look at what's happening, underlying demand versus whatever? Certainly, I mean, we track deals, the win rates, I mean, obviously, order patterns very, very closely. We've been clear that even in the enterprise, there are headwinds, like deals are taking more time, there's more scrutiny over budget, et cetera. I truly think we would be growing enterprise even faster right now if it were not for those headwinds. I think we're doing well in spite of challenges in the enterprise space. As I look out on deals we are pursuing, confidence of our sales team, conversations that I personally am having with IT professionals and CIOs around the world, you know, I, I remain optimistic. I mean, I think we will continue to do well, even in a more challenged environment. If the challenges dissipate, start to alleviate, then I think we'll do even better. SP and cloud is sort of a very similar story. First, on the service provider side, you have to understand, like, our aspirations have never really been massively. We, we've always tempered expectations. Our long-term rates that we set in 2021 were -2% to +2%, right? Last two years, we've slightly exceeded that long-term rate. This year, I think we'll be in that long-term rate as well. There, the thing that I think gives us optimism that we can continue to do reasonably well in SP is Metro. It's the investment, it's the bet that we have made. We don't need service provider to grow for Juniper to grow. There's cloud, and cloud is where there's been the biggest discrepancy between the orders, order momentum and the revenue momentum, the backlog draw, the volume of ordering, and so forth. Here, you know, the... like, we've, we were just now sort of exiting a period where there's been, like, unprecedented orders placed, and we're going to have to play out this digestion period. I'm-- I remain very confident about our ability to win in cloud. I also remain very bullish about the cloud opportunity itself. I'm surprised I haven't been asked the AI question just yet, but I do believe that that is an opportunity for us, as well. [audio distortion] Most of the AI opportunity right now seems to be within the cluster, inside the data center, obviously, Juniper traditionally on the WAN, Mm-hmm. Especially the hyperscalers, which strength you see it playing out, to the areas where you're strong. Yes. How do you see yourself entering the areas which are today, really well? Thanks for the question. Any killer app for a cloud provider is music to my ears, because it will lift all networking boats in their networks, in their, in their businesses. If the big hyperscalers or the broader set of cloud providers start to use AI as a means of driving greater volumes of business from themselves, it means more traffic needs to go to their data centers. It means more investments all up, including in the WAN. By the way, the nice thing about our position is, there isn't this whole InfiniBand versus Ethernet thing. It's Ethernet today, it's Ethernet in the future. It's 100 G moving to 400 G, moving to 800 G in the wide area network. Let's talk about the cluster opportunity, which you're right, is an amazing opportunity. I don't think this happens overnight. I think today the path of least resistance is buying a bundled solution from the king of AI today, which is NVIDIA, right? Including their GPUs, their NICs, and typically it's gonna be an InfiniBand-based networking technology. However, I do believe that Ethernet will win long term. This ecosystem and openness has always won out in networking. I think it's gonna play out over the next year or two. As far as Juniper's position, we're already in AI clusters with our Ethernet-based technologies and learning. Yeah, not in a hyperscaler, but in fairly large AI clusters and learning from that today. I believe that we can win in the AI cluster space. Today, nobody gives us credit for it. That's fine. You know, I wanna delight you and surprise you with what we can do by selling our Ethernet technology into that space. It doesn't need to be hyperscalers. There is a long list of cloud providers and even large enterprises that are starting now to think about what their learning, and more importantly, their inference solutions are gonna look like. All of them are looking at Ethernet as a long-term technology for that. Is AI already driving DCI traffic? Intra DC, obviously, it's driving huge amounts of traffic within clusters, between clusters. Are you seeing it benefit the WAN yet? It's, it's very difficult to know exactly how much it's benefiting the WAN, but to the extent that you're seeing people using ChatGPT and other generative AI-type applications, of course, this has to go over a wide area network. I'll just follow up on that. I'd be curious if you could get a little bit deeper in terms of where you are seeing wins for Ethernet in the non-hyperscale customers. What are the strengths that get you in the door to use Ethernet versus InfiniBand? Then, conversely, from the experience, and obviously, there's a short amount of time, we've been only talking about this in reality for six months or so, as it relates to the opportunity set and bringing to the forefront. Where's that critical threshold where as you learn with your customers, where you start to lose the battle and they start to eye non-Ethernet InfiniBand as the necessity in their use case? to the second part of the question, I mean, I don't see that happening at this point, so it's a bit of a hypothetical, right? You're saying, is there a point where we might get to where there's gonna be this realization that Ethernet's not gonna win out, and InfiniBand is the- Not gonna win out, but in the instances where you're seeing where they're continuing to scale it. Yeah. Where is that threshold where like, "Hey, we really can't use you anymore, we need to roll in Mellanox? Uh- That's what I mean. I think it's more of the opposite that's happening now, where today most are deploying Mellanox and InfiniBand, but are thinking, when does it make sense for me to actually transition to Ethernet? With a few exceptions, as I just mentioned. Right? I don't think there are a lot of people that are actually figuring out Ethernet and saying, "Oh, my God, this is not gonna be the long-term solution. I think I should transition back to InfiniBand." At least I've not come across anybody that's thinking in that way. Going back to the first part of the question around Juniper, the critical components of the solution are, you need to have the right operating system. Our Junos EVO operating system, which we developed with cloud providers in mind, is now battle-hardened. It's deployed. It's deployed inside large data centers. This is, this is, you know, a relatively new development for Juniper that now I have a lot more confidence in. Our silicon strategy. I think we have honed in on a really effective and balanced silicon strategy that includes our own custom silicon with deep buffers, low latency, where it makes sense, but then also leveraging merchant solutions for other elements of the end-to-end solution. Often overlooked is operations. These AI clusters are gonna be massive. There's gonna be a, a huge amount of data, and having an understanding in real time of the health of that network is gonna be critical. This is where Apstra is coming in. If you guys are not familiar, Apstra is an acquisition we made a couple of years ago. It was started by real luminaries in the industry, like David Cheriton, who was one of the founders of Arista, knows a thing or two about the data center, and it is all around simplifying data center operations. Now, this works really well when a customer is trying to balance workloads between private and public cloud, but is afraid of the operational complexity of that private cloud. Today, Apstra is the tip of the spear that's allowing us to win in this space. It's gonna work equally well in AI clusters. We're really thinking about adding enhancements and starting to add enhancements to Apstra that will make data center operations for the AI cluster use case really awesome. Is there a way to articulate whether it's architectural structure or number of parameters in these AI clusters where it, it's your sweet spot? I'm just kinda I guess that's what I'm trying to drill in, is what is the... Obviously, the use case is a large language model. Yeah. Is there something more you can give us to say, "Hey, like, this particular type of customer," or whatever it may be, where, like, that's really where it translates, where they want to come to us because it, it makes the most sense to get the job done for them as best TCO? I think the larger, the better. Juniper is really good at scale, performance, congestion management. Solution attributes that reduce the long tail, like, job completion time, is where we shine. This is a function, again, of the attributes of our silicon technology that we have developed, as well as the operational attributes of our software. On the outside of the InfiniBand opportunities that we see today, you're impacted right now by demand and supply that was out of whack, and that's kind of causing some gyrations. Just the demand trend, are you more tied to new data center construction and lighting up new data centers or replacing older data centers that may be retrofitted for higher speeds because of the need for higher speed transport? I mean, it. Retrofits are always easier for the incumbent, right? If it's sort of adding more capacity, adding more expansion, it's easier for the incumbent. New data centers open up opportunities for new players to come in. That's kind of the, the short answer to the question. Now, there are always exceptions. I have seen large cloud providers that have built out data centers that now want to expand pods or new data center locations that open up the opportunity. You know, we have been successful in these areas. [audio distortion] Not off the top of my head. Yeah. Yeah. Just that, you guys have already had some participation in AI clusters that- Yes. non-hyperscale customers. Is that front-end network or RDMA network or both? Both, including the back-end network that's carrying RDMA traffic. In terms of the positioning there within the silicon strategy, has that been, internal silicon or merchant silicon-based? Both. Okay. Yeah, both. In fact, in one example that I have in my head, it's a we are in the spine using a custom silicon solution where our deep buffers, our scale, our high radix won us the business. In another, it's a merchant silicon-based solution, just because it's a smaller, type of deployment. I'm trying to think about answering again the question of where do we have the biggest advantage? Another area I would add is where said cloud provider knows and loves Juniper in the WAN or this data center edge, it gives us an advantage. We are already operationally tied into their IT systems. They already have professionals that are familiar with Junos, and so it makes it easier for us to position ourselves in a new layer of the network. In this particular case, it would be these AI cluster data center opportunities. What do you guys call your cell spraying fabric for your internal silicon? I've only seen papers from you [audio distortion]. Stay tuned. In fact, I just published a blog two days ago that has a view, like a CEO-level view of our position for the AI cluster opportunity. There will be a follow-on with far more technical details about our position with respect to different architectures. Like, I know there's a debate that's happening right now in the industry around fully scheduled versus Hyper Ethernet and so on. We've been taking our time and being very thoughtful, but more importantly, we didn't want to jump the gun without actually getting some real experience and comparing the trade-offs. I mean, like, take, for example, this fully scheduled architecture that some have been talking about. If you take a step back, it looks remarkably similar to something that we built many years ago that some of you might remember, called QFabric, right? It's a cell-based solution, end-to-end, and we know the good, the bad, and the ugly firsthand of QFabric. These are valuable lessons that we're taking and feeding into the decision-making process of the architectural attributes of our solution. [audio distortion] you guys part of the Ultra Ethernet Consortium? We will be. I mean, at some point, I, I suspect, not all decisions around sort of the problems that need to be solved in the space of AI clusters and Ethernet are gonna happen in one body. There are so many different players and so many different ideas that are floating around. It's important for us to keep an open mind and to work with a number of different stakeholders, including, you know, GPU vendors, NIC vendors, cloud providers, not all of them-- not all of which are in that consortium right now. I do acknowledge that that consortium is a good place for us to participate and to share ideas. [audio distortion] size this tier. Let's call it the Tier 2. A lot of new companies popping up. You've got the CoreWeave and Lambda that are building out all these GPU clusters. Some of the AI startups are building their own compute, not just hyperscale. If you add up all of these, I mean, just the size of them, are we talking hundreds, dozens, tens? If you add all that up, does it equal equivalent to or bigger than a one eye scale? How do you see it? Clearly, this is maybe an area where you seem to suggest you might have a better entree in. Yeah. I, I don't know off the top of my head how these TAMs all compared. However, it's meaningful. Yeah. Even top 10 non-hyperscalers, there are players that are not considered hyperscalers in that top 10, that are coming pretty darn close to, like, hyperscale type scale, like maybe a third of moving to a half of. Those are great opportunities for us, and that's just one in the top 10 that, you know, we would include in Cloud Majors, that will, you know, represent a large TAM, a large opportunity for us. If you add up all of the long tail of cloud providers, you know, beyond top 10, it's sizable. I mean, it's not maybe quite as big as all of the hyperscales combined, but it's I think. There's a win rate in those guys, but not- Very good. Very good. The win rate is... The nice thing about these guys is that it gives us way more at-bats. It's not just about the wide area. We have meaningful presence in the data center. There's one other thing about the Cloud Majors in the data center that I really like in terms of opportunity, which is we have the ability to compete based on our full stack, meaning Apstra, and we are competing effectively now based on Apstra. It doesn't have to be just a hardware story, a hardware comparison relative to the competition. The reason being that in all the hyperscalers, they develop their own SDN layers, their own automation layers. So- Yeah. The digestion is equally as bad in that cohort of customers as the top 10? More or less. More or less. Yes, it is. That, that digestion we said in the earnings call, and this is true, has been pretty broad-based, including in, Cloud Majors. [audio distortion] play an AI question, how do you think about the opportunity [audio distortion] training versus infrastructure? It's a great question. I think most people think of training and inference happening in the same cluster. I do believe to the extent that cloud providers are willing to separate these two types of workloads and processing types, that there can be optimizations made for each of them that are separate. Like some of the latency requirements are far more important in inference than they are in training. Also, inference can be a more distributed problem where you want to do it closer to the end customer, whereas training can be much more centralized. Again, I don't want to get ahead of my skis here. You're gonna see more details from us in technical white papers that we're gonna be publishing shortly. We are thinking about sort of separating that problem and optimizing our solution for each of these problems separately. Quite frankly, the jury's out on how many cloud providers are going to separate their data centers into those two use cases. This is all part of the work that needs to be done over the next couple of years to determine the size and the rate of growth of this opportunity for networking and Ethernet vendors such as us. It's honestly, it's a fun problem. Like, for a technologist like me, that's thinking about, like, working on cool stuff, this is a fun problem to solve. [audio distortion] Mm-hmm. [audio distortion]. Could have fooled me, but go ahead. [audio distortion] days of scaling both training and inference, one of the best ways to manage costs as an operator is to make sure that I'm driving consistently high utilization. Training, by definition, is going to be, you know, predictable, but then but sporadic in the sense that, like, yeah, I'll train for a couple of days, weeks, months, but then it's done. Mm-hmm. Inference would be much more bursty, but as I scale applications, like, it's probably a more consistent run rate. Yeah. I, I would think that until I have large enough scale on both of those use cases individually, that being able to have an architecture where, you know, from two in the morning to six in the morning, I'm running my training workloads, but then it's kind of reconfigured in real time to handle inference throughout the day. Yeah ... as users are using the service. Like, when, when you think about both positioning the portfolio for longer term perspective, as well as making sure that you have at bats over the next kinda 24-3 6 months. Yeah if operators kinda chose, like, my view of the world, where these things should be used for the same. This architecture should be used for both workloads until I hit critical mass. Yeah. What do you need to do, either from an R&D perspective or like a talking with customers perspective, to kinda make sure that you're hedged on both sides? I think you're right in general, that for the most part today, you're gonna use the same clusters for learning and inference. Maybe you get to a certain scale and it starts to make sense to separate. There are a couple of other ways to think about it. Training is the cost center, inference is the revenue generator. This is why I believe that inference has the potential to far exceed the scale of training in time, especially with more and more applications and, you know, use cases that people come up with for generative AI. Also, as I mentioned, inference tends to be more closer to the consumer of the services, so there might be actually a geographic separation that might happen. All of that being said, I believe we have the technology building blocks, the automation layer, the network operating system with EVO, the silicon technology, both merchant and custom, to address both. However, I also believe that I can optimize for them separately, but I have just, I'm still in the mode of learning and figuring out to what degree and what is the timing of that optimization that I should go in to pursue. I don't... You know, it seems like from your comments around, you know, wanting to get some level of battle testing done before and, and some level of experience [audio distortion] Mm-hmm different technology trends. Yeah. Maybe this isn't a fair characterization, but it seems like there's a lot more kinda design, testing, product development work that's forward-looking than [audio distortion]. Absolutely. [audio distortion] Yeah [audio distortion] sales resources or R&D resources, like what's getting the... What's drawing the short straw? Like, what are you guys cutting back? First of all, I do want to reiterate what you just said. We are committed to over 100 basis points of operating margin expansion this year. We believe we can grow operating margin next year as well, and for years to come. This is a result of just being very prudent in where we invest and very strategic and deliberate. Like, no science projects, right? Everything's got to be aligned around the specific direction that we're taking as a company in three solutions: the AI-Driven Enterprise, which we talked at length about, which is driven by Mist; the Cloud-Ready Data Center, which includes sort of traditional data center networking, but also AI clusters. Honestly, like 90% of the technology and the solutions are identical. They're really just optimizations that you make for the AI use case versus a more traditional use case, and then the wide area network. In the wide area network, it's about SP, it's about cloud. Anything else, I'm not interested in, other than maybe a small, tiny little investment to explore something. That is the way in which we are gonna achieve not just revenue growth long term, but operating margin and leverage in our business. Rami, let me follow up on a couple of things. Thank you everyone for the questions, made my job a lot easier. Custom silicon, merchant silicon, like, what's the use case for each that you're targeting? I get that you're sort of winning on both areas, is there a specific use case that custom silicon is better for versus merchant? Then the second part to some of the questions about front-end and back-end. Yes, you have opportunities on both again, which one do you see to be bigger, and how do you think about timing? Is back-end where you win more near term and then sort of front-end more longer term? Okay. First on the silicon question, and many of you know that I'm a recovering silicon engineer, so I know a thing or two about this area. The most important thing to me as a business leader now is no religion. Every new project, every new system that we develop, we have a decision-making process to determine if we're gonna use custom or merchant. We've, until now, made the optimal decisions for every one of our systems. Generally speaking, if it's a large multi-PFE, multi-line card type system, there are attributes of our custom silicon that make it worth it. From an economic standpoint, from a scale standpoint, it's not just when people hear scale, they think ports, like how many 100 G or 400 G ports? It's not that simple. I'm talking about logical scale, the number of counters, the number of policers, logical interfaces, all matter tremendously for these next generation use cases. Economics, features, and scale are the sort of attributes that we measure custom versus merchant silicon on. Pretty much all of our pizza box type systems are single PFE. Single silicon type systems are based on merchant today, where there is no need for us to go and to pursue that kind of custom development, very expensive, when I don't believe it's gonna give us any meaningful differentiation. Then the second part of your question, around- Front-end and back-end. The front-end versus back-end. I think today the front-end opportunity is easier because it doesn't require some of this exotic sort of Ethernet extensions in order to capture the InfiniBand opportunity. However, the back-end opportunity is way bigger. Like, the amount of traffic that you're carrying in back-end networks between GPUs is massive. The opportunity for Ethernet in that layer is just greater, but it's gonna play out over years. Let's make that the last one. Yeah, just a bigger picture one again, Rami. How do you think about visibility and predictability over the next 18 months? The reason I ask that is, we went through this period where lead times were long, people had to be a lot of visibility. We thought maybe these customers are for now not gonna do this again and sort of short term, but when things change, these surprises will happen. Now next year, you have to have some backlog absorption- Yeah This year. Just visibility into how much capacity it have, starting with the Cloud Majors, it's happening somewhat in the, in the service provider. Yeah. When do you get your arms around, like, how much more there is to go? Or is the next six- 12 months, we just have to sort of deal with this transition period? I mean, I, I kind of addressed this a little bit earlier. Demand and orders next year is going to recover meaningfully compared to this year, meaningfully, across all verticals. Revenue is all going to be based on the timing of the order recovery in SP and cloud. That's sort of the, the, the high level I can provide. The, the only additional thing timing-wise, is still not 100% clear right now. However, it will recover because backlog draw will not happen forever. Digestion cannot happen forever. I mean, I do not believe that there is anything structurally changing in cloud providers that will make them not buy to support their data centers and their WANs long term. Like, they have to invest. Is it gonna be in a quarter, two quarters, or three quarters from now? Honestly, like, visibility is kind of limited, but it will happen, and that's, to me, the most important thing. We'll wrap it up there. Thank you again for coming to the conference. Thank you very much. Appreciate the great questions. Thank you. Yeah. That was great. Thank you.