Good afternoon, everyone, and welcome. Welcome to another a keynote session at JP Morgan Technology Conference. I have the pleasure of hosting Cisco Systems with on a keynote on securing and scaling AI. With me is Chuck Robbins, CEO and Chairman of Cisco Systems. Chuck, welcome, and thank you for coming to the conference.
Thanks for having me. It's great to be here.
Yeah, no. Our pleasure. Maybe I'll start you off with something you said on TV after your recent earnings spend, which by the way, I think anyone I talk to, even outside technology, took note of. It's great when energy analysts come and tell you Cisco had a great brand. You know that it's gone well.
Thank you.
Maybe you did mention on live TV going through a networking super cycle. Maybe just flesh that out a bit or unpack the various factors you're seeing, the confidence that it is a super cycle as you see it?
Yeah. I think the notion is that with this AI revolution, everything is dependent upon a network, and everything's dependent upon modern networks, secure networks, and you know, this whole notion of the node has a certain amount of power, but when you network the nodes, you get exponentially more power. We see this happening in when you think about what's happening in the hyperscaler business, which we, you know, we took our AI infrastructure target up to $9 billion for the year from $5 billion. You look at what's happening in the enterprise space and you've got every enterprise was previously focused on modernizing their infrastructure in preparation of AI. Now we've got the Mythos issue that we could talk about. You've got Sovereign Clouds that are popping up. You've got Neoclouds, they're all networking dependent.
Right.
Then you have public sector organizations all around the world that are looking at all those trends that we just described. They're also gearing up their infrastructure for defense reasons and other reasons. It's just everything seems to be moving at the same time right now. I just called it a networking super cycle. I didn't know it was gonna get picked up like that.
Okay.
It's I guess I shouldn't say this, but I've been in this business since these networks started to exist, and obviously the dotcom era was one thing, but this is something that I think even subsumes that.
Maybe let's talk about AI. You did mention it as one of the drivers. When you think about the magnitude of changes, and everyone wants to know what are the magnitude of changes you foresee coming through, how would you compare that to the magnitude of changes the industry's gone through since the early 2000s? Clearly, a lot of investors want to compare everything to the early 2000s. How do you when you think about the next five years, how would you describe the magnitude of the change we're gonna go through because of AI?
Well, I think it's a combination of two things. It's bigger than anything we've seen, and it's happening faster. When you have that occur, it just blows away anything that we've ever seen historically. You know, I lived through the dot-com days, and in those days you had customers that didn't have the financial fortitude. They had an idea, they had a plan, and they needed networks, they needed to connect. Today, if you look at most organizations that are investing heavily right now, they see this transition as being existential to their success. They have some of the best balance sheets in the world. They have the highest cash flows in the world. This, it's from that perspective, it's very, very, very different. You have some startups in this space, obviously.
I think that, in general, it's different in that the companies that are spending most of the money or a large percentage of the money are very well funded.
Okay.
Very successful companies that have shown their ability to navigate different trends like cloud and others. I think that's just a meaningful difference than what we saw the first time. It doesn't mean that, you know, you get a lot of questions about, is this a bubble or not? I don't even know what that means, but I know, I mean, there will be misplaced capital. There'll be companies that won't exist, that'll try something that doesn't work out. I think you'll see the winners emerge, just like we're beginning to see now, and then you'll see them succeed as we go forward.
Okay. Let's layer on agentic AI on top of this.
Yeah.
What kind of changes that does that bring? How does corporate America then sort of function with agentic AI? What does it mean overall for enterprises like even JP Morgan as how things work eventually?
Well, I think what it really does in our world is it drives home the absolute requirement to have security merged into the network, fused into the network. We've said that for a very long time, and agentic actually makes that a requirement because these agents are gonna be moving and processing and working in your infrastructure independently, and there's absolutely no way with the latency requirements, particularly as you get into manufacturing and you get into robotics, there's no way that you can conveniently route this traffic through appliances to apply security policy against it. It's going to have to be done in the network.
A lot of the work we've been doing with Cisco Smart Switches and merging security services onto our into our networking platforms, starting first in the data center and then moving out into campus, and people originally didn't understand why we would do it in the campus, and now with the agentic applications, particularly at the edge or robotics, manufacturing and those kinds of things, you're gonna see the need for that in ways that we haven't yet. I think that, as I've said before, none of my networking competitors have a security business, and none of my security competitors have a networking business. We, assuming we execute well, I think we're gonna have great solutions in this space for our customers.
Got it.
If you look at the Smart Switch technology, which effectively is either CPUs or DPUs built into these networking products, so you can run security services in line at line speed with no degradation. As a percentage of what we've sold, our customers have probably about 5% of what they've been buying has been these Smart Switches, and that's, we think that's gonna escalate pretty quickly.
When we think about scaling AI, clearly supply chain is emerging to be one of the bigger topics as well.
Yeah.
Along with revenue acceleration, we're seeing maybe potentially worsening supply constraints for the industry. You highlighted on the earnings call last week the magnitude of advanced purchase commitments and inventories that you're sort of accumulating just to be able to address that. When you run the company and the supply chain the size of Cisco, how has that changed over the last five years? What are you having to take note of now that you probably didn't need to five years ago?
Well, if you go back 11 years ago when I became CEO, I can tell you I knew very little about supply chain, now I know a lot more than I ever anticipated that I would know. Mark, our CFO, will be up tomorrow, he'll talk a lot about some of the purchase commitments and inventory levels that we drove. I think in its simplest form, we have a world-class supply chain team that's been ranked number one in the world by Gartner on multiple occasions, that's across any industry. This team's very good to start with. I honestly think what's happened is this, our supply chain has now turned into a competitive differentiator for us.
Okay.
We're leveraging our balance sheet, we're leveraging our size, we're leveraging our ability to get ahead of this. If you look at some of the creative things we've done, we've actually, I never dreamed we would be making equity investments into companies in our supply chain in order to help them diversify geographically in some cases, or in the case of a recent investment we made with Nanya, it was connected to securing memory supply, which is a big issue right now. I know we'll talk about it, the other big thing that has completely changed our supply chain dynamics is Silicon One, is having our own silicon. We have a much higher degree of control over our supply chain, and I think that's something that's helping us right now as well.
I think the big difference is it used to be that you only heard about supply chain when something screwed up. I think it has gone from that to being a competitive differentiator for us as we move forward.
Yeah. Let me follow up on that and bring Silicon One in as well. One, it sounds like you think supply constraints are now structural given the demand we're seeing. Silicon One stands out to be what helps you as part of that differentiation. Maybe flesh out the Silicon One, your thinking on the roadmap, and how does it position you in this AI sort of demand, infrastructure demand landscape relative to your competitors?
First of all, if we didn't have our own silicon, the $9 billion that we announced would probably be close to zero because we would simply be a sheet metal distributor of merchant silicon. When you look at what hyperscalers are looking for, the value in the products are either in software or in silicon.
Okay.
The rest of it is important, the real value are in those two things. Many of them have moved to their own proprietary operating systems that we all write expose APIs so that they can build their automation platforms and their management platforms. If you don't have software and you don't control your own silicon, then your future role in these networks, I think, is just gonna continue to diminish. We're very fortunate that we made the decision in 2016 to acquire Leaba. Now what that's done for us is allowed us to build a single silicon architecture across our portfolio. By fiscal year 2029, we're wrapping up fiscal year 2026 now, by fiscal year 2029, all of the high-end systems across our entire portfolio will be powered by Silicon One.
Got it.
In fact, I did bring one, you know.
Yeah.
Gotta have props, right? There's a chip.
I don't know if we can zoom into it.
Yeah, you can't see it much, but it's not Cerebras, right? It's really turned out to be an incredible differentiator for us. What it allows us to do now is in our traditional business, you have the systems, the silicon, the software, and the entire stack for our enterprise customers. When you get into the hyperscaler, we made a decision in 2019, and we launched and told them that we would sell them our technology in whatever way they wanted to consume the technology. We'd sell you a fully integrated system with silicon hardware and software. We would sell you our software if that's what you wanted to buy. We'd sell you just our silicon if that's what you wanted to buy.
I think that gave them a higher degree of confidence that we were serious about really meeting them where they were. We really You know, anybody who heard me talking about this seven or eight years ago, I was very open about the fact that we had missed the cloud revolution and we had missed, and the relationships weren't where they needed to be eight years ago. Now fast-forward, you know, all the hard work and all the time we've spent there and listening to them, making sure our ears were working. And making sure that our actions were aligning with what they were telling us they needed. You know, these customers are big enough, you can look at each one as a market of one.
If you look at as an example, we did announce a restructuring of, you know, something less than 4,000 jobs in Q4. Part of that is we need to fund more silicon tracks. You have to almost run these things concurrently now with the speed at which these things are changing. You know, we did that in order to fund silicon optics. We haven't even talked about optics yet. I think we had a $1 billion+ Acacia quarter on long-haul optics. Then AI and internal AI tooling and things of that nature for our employees. I think silicon side is probably the most meaningful differentiator we have right now.
Okay. You mentioned the hyperscalers. We all look at their CapEx outlooks and can figure that they have aspirations of scaling very quickly.
Yeah.
Supply chain aside, how do you think about what are the other hurdles they need to overcome to be able to scale relative to their own aspirations where they want to go with their scaling?
Well, I mean, clearly, energy is emerging as one of the biggest. I think supply chain, energy, I think the technology and the scale is just moving at a pace we've never seen before. You know, when you look at in the hyperscaler space in the last quarter alone, the way they award business is they give you a design win, and that's sort of like think of it as a franchise inside of their infrastructure. There's defined sort of places in their infrastructure. We were awarded five of those last quarter, two in the optics space and three in the system space.
One of them was with this little chip, which is a G300, and then the other two were with a chip called a P200. This gets back to your question. The P200 is for what's called scale-across technologies, which effectively is connecting multiple AI data centers together. Basically, these models in these clusters are no longer, in many cases, fitting inside a single data center, so they're expanding them into multiple data centers, and they need scale-across technology, and that P200 is our first silicon that was custom-built for scale-across use cases. That's helping them alleviate those kinds of issues.
We won two different hyperscalers with scale-across use cases in Q3, and we also announced on earnings last week that we also have been awarded a 3rd hyperscaler's scale-across architecture in the first couple weeks of this quarter. I think, you know, you look at energy, you look at supply chain, you look at just pure location, geography, the population impact and the general belief of local citizens about whether you should be building data centers. I mean, all those things are becoming impediments, which is why there's a lot of discussion about potential for data center pods in space or whatever those may look like.
Yep, yep. Maybe let's go back to optics a bit. When you're helping your customers scale, optics and systems both are playing a big role now. I remember a time when Cisco had stopped talking about optics. Clearly now the vision of putting these two together has worked well for where we are headed. Where do you see synergies between optics and systems? How is that playing out across the portfolio with your customers, particularly as you look to help your customers scale?
Yeah, I think we made a couple of very strategic acquisitions over the years. One was Luxtera and one was Acacia, and odds and ends around it. They've turned out to be You know, again, we had great rationale for making the acquisitions then. We had great rationale for making the silicon acquisition in 2016. In today's world, they look even more brilliant than they were at the time. I give my team credit. I'm sure they saw that coming. But if you look at the beauty of the optics are that these pluggable optics work with any vendor's equipment. When we win, we think there's a high likelihood that those customers will plug the optics in.
The optics is at least equal to the size of the networking spend, of the switching spend. You really double your addressable market by having both. We sell them when our competitors are being chosen for use cases, then we sell them for our own connectivity. You know, it's critical to have both when co-packaged optics actually comes to fruition or some of these other emerging optical technologies. The fact that we own both really gives us a massive differentiation as we look to serve these customers more effectively.
Got it. Yeah. Let's maybe pivot a bit. I know you've talked about this or you mentioned this before, Anthropic's Mythos model. How concerned should we be, first of all?
Yeah, I shouldn't laugh probably.
Uh-
It's just the number one.
How do you think it will affect you?
We had never heard of it 90 days ago, and it's the number one topic. Isn't that amazing? I mean, that's how fast this thing is. It's the example of if you really wanna know how fast the market's moving, think about Mythos 90 days ago versus today, and that just tells you where it is. You can look at Mythos and OpenAI's latest models, and if you're a pessimist, you can create a very, very scary future. If you're an optimist, you can create a lot of positive ideas about what it's gonna enable us to do from a security perspective, et cetera. There's gonna be a little of both that are gonna apply. There's the ability, it's, they're very good, just to be clear.
We had it for seven weeks before it was talked about publicly, and we have the version that has no guardrails.
Okay.
It is a powerful platform that first and foremost, it has helped us running our code through it, understanding where we have vulnerabilities that humans couldn't even pick up. It's also very good at creating exploits to take advantage of those vulnerabilities. We think using it to help our customers defend against attacks is gonna be very important because, you know, one thing that a lot of people say is that these models are as bad as they're gonna be right now, and they're pretty good. I think that look, what it's really done is it's become a CEO level discussion.
The number of CEOs that called me when this hit, who don't, who still today maybe don't have the model, but they just want to, they want to know what they should be thinking about, what should they be looking at. I'd say there's a few things here. We actually created what's called a harness. By the way, we restricted usage of this model to our security team only. No one else inside Cisco has access to this model right now. They've been running our code through it. They also created a very sophisticated harness, which is a framework for taking advantage of the model to run your software through it.
Okay.
We open sourced it. What's, Sami, what's it called? Do you remember?
I don't have the exact white paper, but yeah, we did open source white paper.
We open-sourced the framework last week for our customers to be able to use, and it'll give them the ability to take any model and actually create this harness, which is what you need to actually really fully take advantage of the capabilities. The second thing is that we did release a white paper called Shields Up, which is a set of recommendations for all customers as we prepare for this new world of these models. You just have to assume they're going to be in the hands of our adversaries. They're going to. The one of the big things that is highlighted, though, is the amount of unpatched and perhaps last day of support equipment technology, not just from Cisco, but any vendor inside of our customers' environments.
If you remember Volt Typhoon and Salt Typhoon, both of those had a lot of instances of exploitations of old equipment that couldn't be patched. We see a lot of customers now who are beginning to really inventory that. We announced Cisco IQ, which is something our customers can put into their infrastructure and actually begin to, on the Cisco front, inventory what they have, give them a view of what's past last day of support, what's past last day of patching, all those kinds of things.
I think this is gonna be, we talked about last week on our call that we really don't have any, anything baked in for what customers may do from that perspective, but we think it has the potential to expedite the whole refresh cycle that we're going through right now. We'll see how that goes. I think Mythos is a, it's just a very powerful and it's gonna fundamentally and the corollary from OpenAI, to be clear, are both very good, I think they're just gonna continue to get better and it's just gonna require us to operate at a different level from a preparation perspective.
Maybe talk about the opportunity here a bit more. One, like any early thoughts in terms of the size of the refresh opportunity about products that are past old and what do you, for support? Secondly, what does it mean for your security business? One is updating the infrastructure. Second, what do you need to do and what monetization opportunity does it offer you or your security business in the future?
We've put together some estimates on how much end of life equipment our customers have, and it's a pretty meaningful number. It's at minimum $10s of billions and perhaps $100 billion-$200 billion. I mean, it's a big number out there. What we see customers doing now is they're segmenting their infrastructure so that they can prioritize where they go tackle that first. I'm gonna segment my infrastructure and put deep security connections between those two segments in my infrastructure. Where I'm exposed to the internet, I'm gonna go hard and I'm gonna really screen that off from other areas and have a very high security posture when that traffic crosses those zones. We're helping customers do that.
I'd say on the refresh cycle in general. We're in the top of the first in baseball jargon. I mean, it's early. We've only released some of the early products in our campus switching, as an example. I think you're mid-single digits of what's out there that we need to replace. I would normally say it'll be a very long cycle, and it'll just be a steady long cycle. With Mythos and whatever else we don't know about, you know, three months from now, six months from now, what things lead customers to perhaps move faster, we'll have to see, but it's very early. Now, on the security side, we've now refreshed the entire portfolio.
It has been a part of the portfolio that we had been hoping would improve, it's beginning to show that everything we believed is going to happen, it's just happening at a slower pace than what we had hoped, to be honest. We saw significant firewall growth last quarter. We've seen five consecutive quarters of really high win rates against our competition in the firewall space now that they're all refreshed. I think that again, when you look at Mythos and you combine the Cisco Smart Switches and the new technology that we had for the campus with our security services that we can run in line, I think it'll be good for both our infrastructure as well as our security business.
Got it. Got it. Okay. Let's take a step back. Wanted to get your thoughts on something broader. Clearly investors are concerned about what AI means for software businesses in aggregate. You have your software business extends beyond security. How do you think about AI disruption risk to software businesses or your own business? However you want to cut it, like give us your broader views of how you're thinking about it.
Our software assets are largely around security, observability, network infrastructure, monitoring, and things of that nature. They're not these single use case SaaS products, which I think are the early low-hanging fruit for AI to disrupt. I know what we're doing with it internally. I actually reviewed a presentation on the plane coming here today that's gonna be given to our Board next week. I've talked to our team, I think that the piece of the software business today that is, I think, most likely and being replaced by AI are these single use case SaaS companies. They're just. Then the other thing we see is that on some of the major, bigger platform players, some new modules that you might have to buy, you can actually obviate using AI in some cases.
We haven't had a lot of concern, maybe we should, but we haven't had a lot of concern yet about AI disrupting the software parts of our business. We think AI is gonna meaningfully improve our software businesses. We think that, you know, when you look at security as the parts of our security solutions that are software, getting those implemented into the network, working in conjunction with the models, like Mythos, I think is ultimately gonna be what's gonna solve these problems. It's not gonna be our stuff being displaced 'cause we know how to communicate with the network in a way that is very unique.
Yep.
So.
Maybe just following up. As much as you don't see that concern, how much does it help you on the operational side?
Yeah.
Operate the business at a much lower OpEx intensity given the help from AI?
I would say that in certain areas, we're seeing some benefit. The good news for us is we're able to reallocate those funding into silicon and things of that nature. That was, the restructuring we talked about was I made a point to say this is not AI replacing people's jobs, right? We're not at, I mean, at broad scale at that level yet. What we do see is, you know, it's a reallocation to these areas that we need to grow. Internally, we're using AI a lot. We're using it a whole lot in our customer service organization, much like every customer in the world is doing. We're using it in our legal space.
We have 27,000 daily users of our own proprietary AI model or our front end to the model called CIRCUIT, and it automatically, you know, determines which model is most effective for the types of queries that we're getting asked. We're using it big in sales now. It's being used meaningfully. I would say we're at the phase where we're in our programming, obviously, our coders. I think that our software, you know, at the end, I think G2 has said at the end of 2027, you might see 75% of our code being written by AI.
I think the next big topic, we were talking about it on our earnings call, is gonna be tokenomics because the cost of all this and driving the productivity of your people up by using these tools, the token usage is getting pretty crazy right now.
Yep. Okay. Let me try to get in two more questions before we need to wrap up. Quantum. How do you see quantum interplaying with AI? Cisco's made progress on quantum, including quantum networking, but where do you see quantum intersecting AI?
Well, it definitely is going to, I think it's just gonna make AI that much more effective and in some cases, that much more scary, right? We began investing in 2020 in this space, which again, is great that we started. We built a prototype networking chip for quantum because when you have these quantum devices, they're gonna be nodes. When you put them together, you're gonna exponentially increase the power of these nodes. The network is gonna be even that much more important, and the security is gonna be deeply important. I think that when you combine AI with quantum, the power of what you're gonna be able to do is obviously gonna be massive.
Moving the data around at quantum speeds as well as providing the security is gonna be something that we're gonna continue to spend more time on.
Any timeline you would-
Still early .
Okay. Good.
Take the different dates that you hear from everybody and average them out. you know, 2030 something, I don't know.
Okay. Last one for you and asking you to make predictions here. If you had to make three predictions about how AI changes the broader economy and AI, IT infrastructure in the next five years, what would those be?
I really do think that, you know, we had this term called cloud native. You went from embracing the cloud to being cloud native, and I think being AI native is gonna be necessary to be successful five years from now. I think that's a given. I do believe that the fusion of security and networking will happen, and it's gonna happen over the next 24 months probably, because it has to. I think the other thing, this is not necessarily IT related, but I think it is going to drive an incredible productivity boom. I really believe that in the world, but particularly in the U.S. I think you're gonna see a productivity boom like we haven't seen in probably in our lifetimes.
Okay. Well, great. Any closing remarks before we end?
Just thanks for having me. It was great to be here. Appreciate giving us the opportunity.
The pleasure is all ours.
Thank you.
Thank you.