Cisco Systems, Inc. (CSCO) September 2026 Conference Transcript

Great.

Well, thank you, everybody, for being here. Welcome to the Cisco Fireside Chat at the Goldman Sachs Communicopia and Technology Conference. It's my privilege to welcome Cisco's CFO, Mark Patterson, alongside Cisco's President and Chief Product Officer, G2 Patel. My name is Mike Ying, and I cover Cisco and ComTech here at the firm. We have about 35 minutes for today's session. So first and foremost, Mark, G2, thank you so much for being here today. It's an absolute privilege to have you on the page with us.

Thanks for having us.

Wonderful. G2, I'll start with you. In the most recent earnings call, I think you described this as the beginning of a multi-year networking super cycle. So talk a little bit about some of the demand signals, the customer deployment milestones that give you the confidence that this AI-driven network demand represents a multi-year durable ramp. and also elaborate a little bit about how Cisco's position not just for the networking piece of the super cycle but also the growing security needs that you've been investing and planning for?

Yeah, so that's a great question. So if you just take a step back on what's happening right now, I think it's non-debatable at this point that this is one of the largest secular shifts that we have seen in a long, long time, if not ever. and if you think about there was a very interesting data point that came out with a study from Open Router which was in February the consumption of tokens by agents actually exceeded the consumption of tokens by humans. Since then agents' consumption of tokens has gone up 14x and now agents consume about 60% of the total inference capacity or the total token volume that's out there. And so that's a pretty meaningful shift. And the second thing that you're seeing is the global compute capacity in the world, 60% of that is being consumed by inferencing rather than training. Why is that significant? Because we are no longer just in the experimentation phase of building a new model, hoping that the volume is going to come. It's actually the models have been built and people are using them and they're using them at a pretty hefty rate. And if you think about the fact that there's probably only like sub-2% of the humans in the world that are probably using agents in any kind of power user capacity, right? And so this is the first time that you've seen in the secular shift that you don't need to have 70% of the humans using it in a power user capacity for there to be massive kind of supply shortage at a macro level on the infrastructure that's needed. So I actually don't see an end in sight as far as the demand signal because we're just getting warmed up with agents. The agents have not even gone to like 2% of the enterprise yet on going out and automating the business processes. Consumers are just starting to get warmed up with agents. And so this is going to continue for a while from that perspective. So if you assume that the demand signal continues this way, which every piece of evidence shows that it does, then the question is, what are the other tailwinds that are also going to continue to persist? Right now, most organizations need fast infrastructure that's low latency, high performance. These agents tend to be far more consumptive on network bandwidth than humans, so about 450% more network bandwidth gets consumed by an agent to conduct the same tasks that a human would conduct. And so one is you have many, many more agents. You're going to have trillions of agents. These agents are going to be working 7 by 24, and they consume more network bandwidth than ever before, five times as much as a human does. And on top of that, you then have a huge emphasis on machine-scale defense because the attacks are going to be machine-scale with agents as well. And so the post-methos effect is going to need to make sure that people are saying, I need to have the right level of security and observability for agents. It turns out Cisco is in the center of all of these macro trends, whether it be the data platform, whether it be the observability platform, whether it be the security platform, and then the entirety of the networking supercycle, for hyperscalers, for neoclouds, for sovereign clouds, for service providers, and for the enterprise. So I feel like this is a very durable trend for a long time to come. Great.

And just on that, Cisco was among the first to highlight how agentic AI traffic was going to impact the network. And I remember that chart very clearly where the human traffic was very spiky, agentic traffic was high and sustained.

And physical AI is not too far now.

And physical AI is not too far now. But I was wondering if you could bring that back to the Cisco business and talk a little bit about how this impacts the portfolio, the Cisco core business growth relative to what we may have seen in historical networking cycles.

See, in the past, what we used to have, we missed the cloud wave at Cisco. That's pretty well known. But in any of these major shifts that have happened, there was a favorable architecture for Cisco and an unfavorable architecture for Cisco. So in the cloud wave, the favorable architecture was on-prem. The unfavorable architecture was hyperscalers, right? And we missed that wave. Right now, there's three variations of architecture that will persist. There's going to be hyperscalers. There's going to be actually some kind of a minor variation of hyperscalers, which is neoclouds and sovereign clouds. there will be the enterprise for on-prem data centers and then there's going to be the edge it turns out that regardless of the variant that the market chooses Cisco has a tailwind if you go into hyperscalers we have a tailwind because we've actually grown our business on the hyperscaler side in two years from almost zero paper taken on AI orders to 9.3 billion in the last quarter in the last year, in the last fiscal year. You will see the same kind of tailwind happen within the enterprise as people are going to want more control, as models go more open source. They might want to have open weights and open source models be on-prem for them. And then what you're also seeing is there's going to be this whole new category of computing or desk-side computing where you might have a Mac Mini or an AMD or something of that sort on your death side that is dedicated for agents to run, those agents are going to need much more bandwidth, even in the campus and branch. So our campus and branch business for networking, which historically was pegged at, like, growing at the rate of GDP, 3%, 4%, has now been growing at, like, 20% for several quarters. And the reason for that is because you're actually starting to see that we have, one, a huge LDOS cycle, which is the last day of support, and people need to make sure that they refresh their infrastructure because they need lower latency, high performance. The second thing is they can't afford to keep dated infrastructure in their facilities because of the fact that you need to ensure that the post-methods effect doesn't actually create exposure to the business. And third is agents are creating so much more traffic that you have to have a whole different level of traffic pattern accommodation than what you've had in the past. So what you see is not just in the hyperscaler business, but also in the core business, you're starting to see a fair amount of tailwinds.

Mark, if we could bring you into the conversation here. The fiscal 27 revenue guidance at the midpoint implies 15% total revenue growth. And even if you take out the AI tailwinds, the underlying is still up double digits. So could you talk a little bit about some of the enterprise or service provider dynamics that are driving this excluding AI acceleration? How durable is this baseline growth? And where do you see near-term upside optionality, if there is any?

Yeah, it's what we get asked a lot in terms of how durable is it, right, or is this just kind of a cyclical thing? And maybe just to reiterate a couple of points that G2 made, I think the three big things that we're seeing is everything that G2 talked about in terms of AI applications, whether it's the models themselves or it's agentic AI really being built out in the enterprise, or just the preparedness for that and the bandwidth, the latency requirements, et cetera, that are being put on the network. That's obviously, I think, the first thing that I'd highlight. The second thing, I think, is just this whole notion of security and networking coming together and the fact that Cisco is very unique, that we're both a security player and a networking player it's you're seeing it play out with our smart switches and running hyper shield alongside it you've got you know the the live protect capabilities to be able to patch and protect you know from vulnerabilities as soon as those are known until you can actually do the patching and the upgrades and then I think you know just the overall need for modernization and this mythos effect that is being termed that way anyway. But basically, the frontier models and the way that they're being able to provide the capabilities to really identify and exploit vulnerabilities at machine scale. And so the notion of sweating an asset another year is a very dangerous notion, if you will, where in the past, companies, I think, thought of this as sort of a nice-to-have and really scrutinize the ROI, it's now becoming a security imperative. And you're seeing boards and CEOs say, I want to know everything that's in my environment. I want to understand the risks and what's our path to actually upgrading that infrastructure too. And you're also seeing a huge opportunity from just the need to refresh and modernize infrastructure. So what's coming, you know, end of life, end of support for us, just in the next few years, we've identified over $100 billion worth of opportunity just for Cisco upgrade and refresh. And by the way, we're going after our competitors end of life, end of support as well. And you're starting to see some of that be reflected in some of the market share reports here as of late, showing that not only are we holding share, but we're actually taking back share in most, if not all, of the networking categories for us. So I think in terms of durability, as you look at each one of those sort of three big areas that I outlined, those have all been tailwinds for us if you go back a few quarters or even another fiscal year or so but they're all more pronounced today than they were before and they're all very early stage in terms of where they're going to be headed over the next few quarters and years so I think the fact is this is going to be very durable it's a multi-billion dollar opportunity for us in many, many facets, and it's kind of this refresh opportunity is creating a bit of a demand floor, if you will, as you look out over the next couple of years, and you're seeing it play out. I mean, networking's now been double-digit growth for eight quarters in a row. It grew 40%, actually, in terms of orders overall for the company this last quarter, so we feel pretty good.

That makes perfect sense. No enterprise wants to be caught running devices that are end of support just given the emerging threat landscape.

Especially not today. As G2 always reminds us, these are the worst models that we'll ever see. So 90 days from now, the sophistication is going to be incrementally better, exponentially better.

Mark, just on gross margins, you have fiscal first quarter 27 gross margin guidance of 65% to 66%. How much of this gross margin guidance is driven by potentially just lower margin hardware, scaling into the revenue mix, or commodity cost, inflation, memory, and things like that? As we look throughout the balance of fiscal 27, how should we think about the trajectory of margins?

Yeah, I'd say if you look at the last quarter or two, we've actually been able to manage the memory price increases pretty well and been able to pass the costs on to our customers, and largely in part because of the demand tail ones that we talked about. I think that as you look at gross margins overall, there's a couple of things happening here. One is, bottom line, as G2 outlined, this networking super cycle, if you will, we're leaning into these opportunities, and we're shipping a lot of hardware, whether it's the predominantly hardware-driven hyperscale space or this significant uptick that we've seen on the enterprise side. While you've got hardware that gets recognized day one in terms of revenue, the software that gets attached to it is a subscription, and that gets recognized ratably as well as the services that ultimately get put into place, also ratable. So there's a bit of a timing difference on that side. But I think, you know, what we said was there's sort of a slight, you know, headwind to the numbers in terms of gross margin as we go through FY27. We've looked at, you know, the analyst estimates that have come in and feel comfortable with kind of where they're sort of stepping it slightly down as we move through FY27. But I think the biggest thing to look at, frankly, is operating margins. So, you know, while gross margin, there was a headwind, and if you take Q4, for example, gross margin was off about two points relative to the year before, but OpEx was actually down almost four points in terms of the percentage of revenue, And operating margin then was at a record level for the company, up almost two points on a year-over-year basis. So that operating margin, you know, we actually did guide, which was kind of unusual because we wanted people to understand that, look, we are going to be dropping money to the bottom line at a faster pace and at a more efficient pace than we have ever before at Cisco.

Great. Makes perfect sense. G2 you talked about the 9.3 billion dollars of hyperscale AI orders last fiscal year that does not even include the enterprise and rest of cloud orders that are for AI could you talk a little bit about how you expect this goes overall AI revenue mix to evolve how are you positioning your product portfolio within Silicon One, optics, smart switches that capture demand across both ends of the hyperscale AI and then the enterprise and rest of cloud orders. Yep.

So just for those that haven't been following it closely, two years ago we had said, hey, we need to do a billion dollars in orders taken and put a stake in the ground for AI orders. We ended up the year at 2.3. So we said, oh, it seems like it makes sense to double it. So we doubled that the following year and said, why don't we do $5 billion? We ended up at $9.3. And we literally, in Q4, did $4 billion of orders, right? And so there's clearly momentum there. Before we do that, let me take a step back on, like, what is the structural shift that happened in Cisco? Because, you know, Cisco has been around for 40 years, 41 years. And when you start thinking about one of the biggest challenges that we had is we used to operate like a holding company. We had 250 acquisitions. We had a bunch of GMs running their own businesses. Those businesses, actually, each one of them was managed like its own fiefdom. And before you knew it, it was very hard for customers to tell how these things all kind of came together. what we've done over the course of the past few years is we have actually become a vertically integrated platform that is a co-designed full stack we make our own silicon without which we would not have a hyperscaler business because they want choice and diversity from the others in the market we make our own photonics we have our own systems business with systems hardware we have our own systems software we have our own platform for security we have a platform for observability We have a platform for data. We make our own models in some cases for networking and time series and security. We have our own applications. We have our own agents. And we have one unified management plane for all of this. So regardless of what you're managing with us, whether it be WebEx or Splunk or Nexus switches or Catalyst or Meraki, what have you, it's all managed in one management plane, one control plane, right? This is a very different Cisco than what it used to be. That transformation would not have happened without AI, not just from a demand perspective, but also because we now have 32,000 of our engineers using AI to build products on an ongoing basis. And the speed and velocity with which we're building has really kind of sped up. We used to get these estimates that to go out and create a unified platform would be 7 to 10 years. In fact, Mark was in the first meeting when I took over all product with. They actually said that to me. Within nine months, we were able to go out and get that built. We now have by the end of this year, we'll have half a dozen products that are going to be 100% written with AI. By the end of next year, we'll be 70% of our portfolio will be 100% written with AI. No human lines of code. There's a level of momentum that just is there in the business where if you have scale and if you can inject speed. That's a superpower that's just very hard to replicate. And we have scale and speed not just in one tier of the stack. We have scale and speed across the entirety of the stack, from silicon and the physics to the semantics. And so that's a hard thing to go out and just say, let's go replicate that. And as we think about our overall business, We grew our orders 35% for the overall business for the past two quarters, but ex-hyperscalers in Q4, we grew at 25%, which means that we've got these inherent tailwinds. And as Mark said, there's not only durability or demand, there's also a floor because you need people to upgrade your infrastructure. You need people to go out and make sure that you prevent against mythos, and you need to make sure that your agents are satiated with the right level of infra. And so all of those things become massive tailwinds. And then we've got the security and observability where they are kind of fusing together as markets as well. So it's much harder to tell today whether an agent was influenced because of an external prompt injection attack or an agent just decided that it was the right thing to do to go out and escape containment from OpenAI's container and go attack Hugging Face and steal the weights there because it was just following orders to say go out and make sure that you pass the eval. So the fact that we have both security and observability and we can fuse them together and fuse them in the fabric of the network creates a very asymmetric structural advantage for us that is just hard to replicate. And that, along with the fact that we've got scale of the number of customers, over 960,000 customers, when we build our own silicon, one of the things that you need to have is scale so that you can amortize the cost of silicon across multiple different volume segments. And now, by 2029, we will be completely independent of merchant silicon providers because we'll have made our own silicon. All of these things start to compound on top of each other. And so, you know, and these are hard businesses to be in. Like, you know, being in the silicon business is a five-year planning cycle. Being in the agent business is a three-year planning cycle. Being in the hardware business is a 24- to 30-month planning cycle. Being in the software business is a 12- to 18-month planning cycle. We've actually been able to perfect all of those and have them work harmoniously together, which is why you think about a co-design full stack. And that, I think, is what creates the level of excitement both within Cisco and within our customers and partners. And you can see a spring in the step in both our customers and partners as well as our employee base. 90,000 employees ship and operate at the speed of a startup while still maintaining the scale that we've maintained.

Yeah, I was just going to add, I think to G2's credit, he won't brag on himself, But one of the key things, too, in terms of, you know, you used to have all these GMs, as G2 described, and they were all fighting for money, and they all wanted their business to be successful, but maybe not somebody else's business. And so when you really think about secure networking, for example, you've got to bring multiple businesses together. And one of the very first meetings that G2 had me as a part of, everybody's now fighting to have, how do I get security built into my solution? And you don't have disparate, you know, business units asking for money. You're now looking at it holistically underneath G2 that says, okay, we have massive opportunities in silicon and optics and AI, and how do we actually fund what really matters?

And so that's been a big help, too. just on the hyperscale AI orders historically you guys have talked about you know 60 60% silicon-1 based systems 40% optics could you just talk a little bit about which areas of the AI stack Cisco has the most exposure to today and does that change over time I mean I've certainly seen the success that you guys have had to scale across just as an example.

So there's basically, from an architecture perspective, for those that might not be kind of familiar, the way that this all started was a single GPU would be able to load a model, and you would train the model on a single GPU. Then the models got big enough because you had many, many more parameters, and so you said, oh, I need to make sure that I have a model where eight GPUs can act as one coherent cluster. And so you created a server with eight GPUs, and they acted like one GPU where you had memory that was coherent. Then they said, well, we need to make sure that we go rack scale. And so now you have, you know, NVL 72 kind of racks where you've got that. That was called the scale-up networking side. But then these racks weren't big enough, and so what people started doing is row-level scaling, which is like you have rows of racks that needed to be tied together within a data center. And that's what was called scale-out networking. And now what you have is the rows themselves aren't big enough within a data center because you can't pull enough power to a single data center. So what's happening is rather than pulling power to where you want the data center, you're going to where the power is available. And so these data centers are starting to get to be hundreds of kilometers apart, but they need to operate like one logical unit. And when you do that what it requires is a whole different set of architectures a whole different set of chip design because you might want to have things like deep buffering so if you do have some packet loss you don't need to restart the training run and so what we have is we operate right now in both scale out systems as well as scale across systems which I go across data centers the scale out systems, we now have our latest chip G300, which has close to a quarter trillion transistors on a hundred millimeter packet substrate. And then we have scale across systems, which is the P series chips. So what we have is not only the chips, but we also provide these hyperscalers with full systems in place so that they don't have to worry about support and all of those things. And so we provide them the full system. And then with enterprises is in NeoCloud and Sovereign Cloud, they also want to have just full secure AI factories, which also have security and observability and data and log management and all of that with Splunk. So that secure AI factory that's a partnership with NVIDIA, which we just expanded with having Supermicro in there, we're now able to have not just a server, but server with networking and security and observability and data platform all in a single factory that can then be tied to what you... And the beauty about this is we always struggled with the compute business because it was lower-margin businesses. But because of the partnership with Supermicro, you can now do rack-scale compute and actually have a net-margin treatment for us. So that actually doesn't have a headwind on margin, but it does provide you with tremendous amount of drag potential for all the other technologies that you have. And you've got this full stack where we can go to an edge device, we can go to an 8GPU server, or we can go to an NVL 72 class rack, all built out with Cisco technology all throughout with partner ecosystems.

On security, maybe one for Mark and then one for G2. I guess, Mark, you know, you got it to a recovery in security, observability, and services revenue for fiscal 27. Could you just walk through the primary drivers of that improvement? Generally, how should investors think about the Cisco security business and Cisco 27? And then, you know, G2, maybe you can expand on whether this is cross-sell, the refresh products, what's happening to the legacy security portfolio, what's happening with Splunk.

Yeah, so I guess at the highest level, I'd just say relative to security and services, as you mentioned, I think we're turning the corner. both. If you look at the guide that we gave for FY27, it's for kind of high single-digit growth and security. We grew low single digits for the full year of FY26. As you look at security, Splunk begins to lap this sort of accounting issue that we had, not issue, but transition, if you will, from moving from on-prem to cloud. And that really starts to lap, and so you're going to see positive growth from Splunk and improving through the year. If you look at our core organic security profile, if you look at the new products, we actually signed up 1,500 new customers just in Q4 alone for those new products that we've outlined. On the refresh products, Firewall now has grown in excess of 30%, two quarters in a row, so I feel real good about sort of moving past, and I think you'll see that for the full year being high single digits, but I think you'll start to see it accelerate to double-digit growth as we move through FY27. And then on the services side, similar story to what I was saying earlier. I think we've seen massive hardware growth. You're going to start to see those services attach. The ratable revenues start to come through. So again, we said you'll start to see services positive and moving to kind of mid-single-digit growth by the end of the year.

On the security side for us, I mean, the opportunity is massive because, firstly, we talk about this thing called a naked network. You don't want to sell a naked network. You always want to make sure that security is tied to the network because there's such a huge need right now of making sure that if the attacks are going to be machine-scale, then the defenses have to be machine-scale, and you have to assume in security that the adversary has already infiltrated your environment, And what you're trying to do is prevent lateral movement. Where does lateral movement happen? On the network. Who has the most amount of telemetry about the network? Cisco does. If you tie that to the security side, we have such an enormous advantage because our security competitors don't happen to have a network stack, and our networking competitors don't happen to have a security stack. We happen to have security fused into the fabric of the network, where we've actually taken the firewall and the switch and fused it into a single device on a top-rack switch. That's what we call the smart switch. We have a DPU and an NPU on the same device. So the latency goes down. The packet inspection happens at line rate with packet forwarding. And so it's a much, much faster kind of solution. And we have taken security and sprinkled it all throughout the fabric of the networks. If you happen to want to have enforcement for security in a top rack switch, you can do that. You happen to want to have it on a server, you can do that. You want to happen to have it as an agent overlooking what's happening within your kernel and what processes are terminating on the host, you can now do that. We have basically taken security and fused it in the fabric of the network. So then the question is, okay, so what's the next thing? Well, first is our architecture is going to change yet once again because all of the agents are going to need to get secured. Right now, so far, we've only been able to secure humans. When you think about agents getting secured, that same zero-trust principle that you use for humans of least privileged access, you're going to need to now apply to agents. Agents are going to need to make sure that they're constantly being monitored, and when they start to show behavioral drift, where they start saying, oh, I'm starting to see an agent do something that's different from what I want it to do. You have to be able to intercept it in runtime. All of those capabilities we now have within our foundation stack, right? And it's completely tied into the network. So that is what the key advantage is. We've got security and networking fused together, observability fused together. We can not only measure how resilient the infrastructure is, we can also measure the cost of tokens. How consumptive is that agent for token usage? Is the agent using too many tokens? Do you want to quarantine the agent because the tokenomics are not favorable to what you might want to do? Is it going to blow your budget up? All of those pieces now are part of that stack, and that's what we bring to the market.

In the last couple of minutes, I just have a closing question for both of you. I guess, Mark, first, could you talk a little bit about the capital allocation strategy? Like last year, fiscal year, you returned nearly all your free cash flow to shareholders. How's the management team balancing the shareholder return strategy, investment in the core business, M&A, right? Is that critical at this juncture?

Yeah, really, really unchanged for us. First and foremost is to support the growth of the business, be it organic or inorganic. Second is really to, you know, protect and grow the dividend. and you've seen us do that for years. And then thirdly is just offset dilution with our buyback program as well as just return excess cash. And as you mentioned, almost 100%, basically, of free cash flow we return.

Great. To close out, G2, as you look out to Cisco's multi-year roadmap, what's the overarching vision for the company over the next one to two years? What technology or market opportunities are you most excited about? And Mark, if you have something to add as well.

Well, look, I think we are right now at a perfect time with a massive market that we can address bit by bit. We've actually got the most comprehensive stack of any one. We've got a great product set. We've got a great team. We've got a great brand that's trusted. And we've got great distribution. So timing, market, team, product, brand, distribution, we've got all six. We've got them in spades. So that's great. If you think about where you expect us to invest our capital so that we can continue to keep growing, you should expect us to make sure that we are focusing on all five kind of customer segments, hyperscalers, neoclouds, sovereign clouds, service providers, enterprise. You should expect us to focus on networking, security, observability, and the data platform. and then on the, you know, if you think about our stack, where should we be having continued investments that we are actually pouring in is silicon is very strategic for us because everything kind of is at the foundation of silicon. And so we will continue to make sure you should expect us to make sure that I think it would not be a crazy idea to think that we are going to actually participate in different kind of use cases of silicon and we'll keep expanding over there. It's not crazy to think that we will actually keep doing the same thing on the photonic side. And then as you keep moving up the stack, the goal is to make sure that we are fully co-designed in the way that we do it. And so every single thing that we do at every layer of the stack should be intentionally done with what it does to create a compounding effect for every other layer of the stack. And that's what creates this. the goal is reduce the marginal cost of ingestion of Cisco technology, continue to keep compounding the value so that every single thing that you buy from us should add value to what you already have from us, and have an open ecosystem. We don't want to be closed. We want to stay open. We will even partner with our biggest competitors as we move forward, and that actually gives us the right level of customer sentiment where they want to trust us and continue to keep partnering with us.

Quick comment because I know we're at time. I just started my 27th year with Cisco. I would say I don't remember a time when we've had as much opportunity ahead than we do today, and I'm probably more optimistic about our ability to actually capture those opportunities than I have been today. So thank you for having us, Mike.

G2, Mark, thank you so much for participating in the conference. It's been a privilege to have you on stage.

Thank you so much.