The Network Is the Bottleneck for AI
Enterprises are spending heavily on AI tokens and agents — but the network underneath is buckling. Michael Dickman, SVP and General Manager of Campus Networking at Cisco, joins Alan Shimel on Techstrong TV live from Cisco Live to share a striking finding: in a Foundry survey of more than 3,400 CIOs, 75% said they have more confidence in their AI strategy than in their network’s ability to actually support it. Michael walks Alan through the explosion of east-west traffic from physical AI and computer vision, why agentic AI demands a shift from network access control to network action control, and the looming hardware super cycle as enterprises refresh aging campus and branch infrastructure. He also recaps Cisco Live announcements — Cisco Cloud Control and AgenticOps, Live Protect, Unified Fabric, Digital Twin, and the new Catalyst 9550 Smart Series Switch with 400G uplinks.
Transcript
Hey everyone. Welcome back here to Techstrong TV. I mentioned it when I came home.
I had the best time in Vegas, and I usually don't say that about Las Vegas, let me be honest. But I had the best time in Vegas, I guess about two weeks ago, when I was out there for Cisco Live. I had the chance to meet some extraordinary people, some of the key executives at Cisco, some folks that are running some of the largest college campuses in the world.
The CIO fellow from LSU who's doing this Tiger SOC. What a great story about training people for security. And it really opened my eyes once again to just what a great company Cisco is and in how many different areas they play.
And we don't always think of it, right? I think Cisco, a lot of us take for granted because they've always been there it seems, and they always will. But when you start peeling the onion back a little, you start seeing all of the different places they play and the different things they're doing, and it's fascinating.
One gentleman I didn't get a chance to meet out in person in Vegas is my next guest. His name's Michael Dickman, and Michael is the SVP GM for campus networking at Cisco, and let's welcome him to Techstrong TV. Michael, thanks for coming on today.
It's a pleasure to have you on. Thanks, Alan. It's great to be here.
I'm sorry we didn't get a chance to do this in person in Vegas, but better late than never, and we're happy to have you here. Michael, your experience as a Cisco person, it's a pattern that I saw out there a few times in that people who went to Cisco, they then left. Either they started their own company or they went to work for another company or two.
But they make their way back, like salmon spawning, right? They come back to the stream they're born in. You're one of those people.
Tell us a little bit about your unique journey. Yeah, happy to do it. Yeah, we call ourselves probably boomerangs more than spawning salmon, but that's a new one.
I'll have to think about it. Okay. Yeah, you're right.
That's a better way of putting it. But yeah, no, when I first started in Silicon Valley- Tell us what happened ... Yeah, when I first started in Silicon Valley, I did come to Cisco in a strategy role and then quickly got into product management and enterprise routing.
So if any of your viewers have heard of the ISRG2 or the ASR 1000, those were my babies at the time, which was quite a lot of fun. But then, yeah, like you said, I did kind of go off and explore the world a bit, and so I ran the switching business at Aruba Networks, for some time as part of HPE. And then I did a tour more in the data center and cloud area with security and observability at a company called Gigamon.
But in the last couple of years, and this was more me on the outside until recently- Familiar ... really started to see a renaissance at Cisco about huge investment and focus, and that combination of investment with focus is very exciting. And you talked about how Cisco's always been there, always will.
I think what that means is there's a great responsibility and a platform and a foundation upon which to do more and better, and I see that happening inside the company, and it's just such an energetic time to be here. So very glad to be back. Absolutely.
It is a great time to be back. Look, AI and the whole notion of managing an AI platform, to deliver AI apps and everything else, it's an historic time. It's exciting times.
Now, excuse me. I mentioned you were SVP GM of Campus Networking, right? What exactly does that mean?
Yeah. So campus networking will really be in the classic enterprise estate and what it means at Cisco specifically is switching, wireless, and also our cameras and spaces businesses. So those four businesses make up what we call campus networking.
And so historically that would've been like the Catalyst and the Meraki lines for switching and wireless. But it's really any way that someone needs to initially connect to the network, when they're inside of a building, or even outside of a building. But you can kind of think of it really as your classic college campus, but also, even in like a small coffee shop, when you get on the Wi-Fi, or the switches in the back of the store that might connect everything.
So it's all of that infrastructure that connects all of us to the network every day. Love it. I love it.
Michael, well, actually, before we jump into the topic of discussion for today, for people who want to get more information about the campus networking division within Cisco, where within the website is there a specific section of the website or a special website for that stuff? Yeah. I think if you're more product-oriented, you can just look into switching or wireless or cameras or spaces directly.
But if you really want to get a sense of where we're going, actually, I would recommend to start at the top level, and this is some of the things we'll get into, which is how to solve some of the bigger problems around security, around agentic operations with access, and so it depends if you want to come more from the solution perspective or the product perspective. But yeah, I think either way, the Cisco website is a great place to start. Love it.
All right. Let me change gears a little bit on you. You guys recently had a research report and survey done, I think it was with Foundry and, over 3,400 CIOs.
I don't want to steal your thunder. Give us a little background on this survey and maybe some of the key findings, and we could jump from there Yeah. So like you said, it was a survey done with this third-party foundry, and it was done in the spring of this year of '26.
And I think what I really found so insightful here was it went straight to the CIOs of actually fairly larger organizations. So all respondents were from call it enterprises, which is more than 500 employees, but also more than that. And in fact, on average, there were over 3,000 locations per company.
So these are basically the CIOs- Wow ... of large companies, and other organizations, which really is a great opportunity to hear what is on the minds of the C-level there. And in terms of the focus, a lot of it was on this question of how does AI affect the network.
I think we've seen a lot of how networking can support AI. And we've also seen a lot of how networking operations can use AI, and we can talk about those as well. But this survey was very focused on what is the impact of adopting AI workloads on that campus and branch network.
So not the data center, right? But on the campus and branch networks we use every day. And it was pretty clear voice, with which we heard.
So 93% of the CIOs said AI demand is already accelerating modernization efforts, and that should be no surprise. Yeah. But I think what's maybe more of a surprise is seeing an increase in, or how fast this is happening rather, seeing an increase in that, what we call east-west demand or lateral traffic, right?
Because a lot of times when each of us get on the computer, like the conversation you and I are having right now, is the traffic goes up into the internet northbound and then southbound back to us. So we call that north-south. But with AI and physical AI and camera vision, and all computer vision and all of these things that are happening, we're seeing a lot more lateral communication within the environment.
And CIOs then need to rethink their campus and branch networks to be able to handle this. So there's certainly going to be an increase in traffic, but I think what we're also seeing here that's different from past evolutions that increased network traffic is the type of traffic itself is even very different. And so there's quite a lot of interesting but also challenging characteristics of that AI impact on our networks.
I think it for sure, and it's causing angst, right, among those people- Yeah ... who are tasked with managing it. I had a look and there were a couple stats that jumped out to me.
One is Captain Obvious. 85% of organizations plan to expand agentic AI in two years. You know what?
I thought that sounded low. Believe that's the crazy world we live in. Yeah.
85%? Yeah. Nah, it's low.
It'll be 95%. But the other one, it is a little interesting. 75% have more confidence in their AI strategy than their network's ability to support it.
So let's think about what they're saying there. Three-quarters, a critical mass of them, they have more confidence that their AI strategy's the right strategy, what they're pursuing is right, but they're not so sure about their network's ability to support it. Why you see that gap, if you will?
Yeah, no, it's an even interesting phrasing to the question, right? Is they're more confident in one than the other. I think it comes down to starting at that business level of what needs to change to really harness the full power of AI.
And we're definitely seeing a shift in the industry now from experiments or pilots to, okay, here's what's really going to work and drive value. And that's been moving apace for a little while now. But as those things actually start to scale, you need to handle them at scale, and I think that's where that gap is that you pointed out, which is, yeah, I think people are getting more confident.
They know where is the value, how am I actually going to get business value from AI? And there's all kinds of examples, healthcare, manufacturing. Some are same, some are different by vertical.
But when you come to the network side, I don't think we've seen a lot of customers yet really prepare. Some aspects are actually security oriented, right? Like new types of attacks and threat vectors.
But a lot more just around the sheer traffic and connectivity. How do you know what an agent is when it's on the network? How do you rethink what we would've historically called network access control?
Now you may want to think about network action control. Like what is actually allowed by a computer that's able to take actions on our behalf. And so how do you manage that?
How do you observe that? How do you secure that? And obviously, how do you connect that so everything can work?
And yeah, those are new types of challenges. And I can see why CIOs would be more confident in the former than the latter. And that's certainly something that at Cisco, and especially me and my group, are working very hard on the connectivity with security and confidence.
And this does go back to something our Chief Product Officer, Jeetu Patel, has talked a lot about, which is the trust deficit. Sure. In the past, you would have a lot of business updates or upgrades just come at us, like bring your own device or the BYOD trend.
I think with AI, trust is so important that until we can fully trust how these work, how they're connecting and securing, that we won't actually be able to get the full value. And so solving that trust deficit is so important, which is why we need to be able to trust that the agents can get on the network as they need. We can trust that only the agents that should be on the network are there, and we need to be able to trust that what's happening with them are secured, both that they're ingesting proper data versus poisoning and some of these other threats that are out there.
And that the actions that they take are compliant with what the organization intends for them to do, which is obviously its own area of challenge. And all of that happens on and over the network. Absolutely.
It's interesting sitting here listening to you explain it, it makes all the sense in the world. Now, when I had a look at just the survey stuff that got sent over as part of scheduling today, I didn't have you. So, and this gives you an idea of the inner workings of my mind I'm thinking to myself- Living in China?
Sure. They spent all their money on AI tokens and agents, but they didn't invest in the network. They didn't invest in the network to keep up or to respond to this investment in-- It's sort of like an equation.
If you put stuff on this side of the equal sign, you got to put a like amount on this side of the equal sign, otherwise it's no longer equal. Okay, good. And so in my mind is, are companies thinking along those lines, Michael?
Are they saying, "Hey, we need to invest in the network if we're going to put more load on it with all these AI things"? What do you think? I think you're right that that network consideration came second.
But I do think that CIOs are starting to be aware of it. Even the fact, like the statistic you shared, that 75% are agreeing or stating that they're more confident in the AI strategy than the network's ability to support it, does recognize there's a challenge that hasn't been solved yet. So that's where we need to really lean in and help solve that.
Some of that is just sheer performance of the network and making sure that it can handle some of the massive increases in traffic. And then a lot of that is also just how that network operates. And so just as one example, you can think about a normal human type of traffic is very occasional bursty traffic.
And so that on average, the network utilization is actually somewhat low in most campus environments in the world. But when you start to have more agentic traffic that's always on and always communicating, it becomes different, and the average network bandwidth utilization can become high. And so these are the kinds of changes that CIOs and network planners need to be ready for and plan for.
And like you said, yes, as AI scales up, the network will need to scale up to support that. I think the other aspect I do want to mention is going back to that security question, is there are new challenges with AI agents on the network in terms of security, and being able to have a layered approach around how you do segmentation to make sure that you have guaranteed ability to keep guardrails around where the agents can and can't communicate becomes a really important method of trust. Right?
Because you may have good citizens who register their agents and say, "Hey, here's an agent. " And then you're going to have situations where well-meaning employees, or potentially not, are going to be just putting agents on the network. And we've probably all heard stories.
I'll just share one in a moment that really made my eyebrows go up about well-intentioned people with well-intentioned agents that are doing some things you won't expect. So one friend of mine here at Cisco, he was telling a story how he was interested in a certain accessory for his truck, and so he asked his OpenClaw agent, which he had running at home on a Mac Mini, to keep track of when this accessory might be available. Well, the agent had a hard time finding information online, so it decided to download the ability to make voice calls and actually call up truck dealerships and have a voice call requesting information about this accessory.
And you think that's in a way good because the agent was creatively solving the problem, but it also is a little bit creepy to have this agent making voice calls without explicit permission. Mm-hmm. Yeah.
And so that's where you think about that idea of agentic control. What are you really authorized to do? Not just least privileged access, but least privileged action, and the value of network segmentation to actually build fixed guardrails that cannot be violated around that becomes even more important.
So it's both the performance side and the security and segmentation side. Agreed. At some point, though, the laws of thermodynamics or what have you, have to kick in here, and the network has to catch up to all this AI investment.
Now, a lot of people are saying it's going to create sort of a networking super cycle. Mm-hmm. Right?
And in doing so, it's a question of not if, but when. And so is 2026 the year we see a super cycle, or will it take a little time for the network cycling to catch up to this tremendous AI investment? How long, how much runway do enterprises have before it becomes critical to make these kinds of investments to AI ready your network for agentic?
Yeah. That's a great question. I think we've already seen that networking super cycle happen or is happening rather in the data center, and it's going to happen next in the campus and branch.
Right? Because first you have where do most of today's agents live, and now in the future, you're going to have where most agents consumed and new agents will live, as well as physical AI. So it will happen.
And as for when, use the crystal ball, but pace of change that we're seeing in some forward-thinking customers, I would certainly say, this calendar year, we should see that starting. And you might argue it's already started, because of some significant growth we're already seeing, and then certainly continuing into next year. And that's really also compounded by one other factor, which is sort of just the age of infrastructure in most campuses and branches in the world today.
So there has already been a long-running kind of, call it sweating the assets, letting these devices live already in the network for some time, and so already they're aging. Some of those devices even going last day of support. And so now we're being faced with this immense challenge of AI, and we need modern security.
And so I think the combination of that will definitely drive this networking super cycle. Agreed. Hey, Michael, we could dive more in, but for people who maybe want to get a deeper dive themselves into this survey, is that available off the website?
Yes, I do believe there is a... Yeah, I believe if you just Google it, there's a little link. It's called No Time to Wait, I think is the phrase- Okay ...
beginning the survey. But yes, that's available from Cisco and Foundry Research, and it really is packed with a lot of data and detail around challenges and concerns, but also just what everyone is expecting to happen, like some of the questions you're asking. When do we really expect these changes to happen, and in what areas?
So yeah, everyone should go ahead and find that report and learn more. Excellent. Hey, while I got you, if you don't mind, as I mentioned, I didn't have a chance to meet with you at Cisco Live, but your group put out some big news.
There was some significant announcements. For folks who maybe weren't obviously at Cisco Live or didn't catch the news, can you give us a quick summary? Yeah, sure.
So we talked really about three pillars of innovation that we've been talking about, but new enhancements in each of those areas. So we talked about agentic operations, fusing security into the network, and scalable devices for this AI era. And so in agentic operations, I think the number one headline from Cisco Live was Cisco Cloud Control.
And so my group in particular talked about applying Cisco Cloud Control, which is kind of like a single governing control interface across all of Cisco, and applying that to some specific campus and branch use cases like unified fabric, running a fabric all the way from the campus to the data center with industry standard protocols like BGP, EVPN, VXLAN. So kind of in the guts of how to deploy a fabric end to end through Cisco Cloud Control. Other example of agentic operations would be like a digital twin, and so actually being able to not just mathematically model but fully emulate a network, and so that if you wanted to make a change, configuration, plan an upgrade, deploy something new, you can actually emulate that change and then see what happens.
So digital twin there, as well as much more. Experience metrics is a great one about taking a user view on their experience as they connect to the network. In security, we talked about Live Protect, which is shielding against vulnerabilities that may come up without requiring a patch or upgrade.
So actually deploying what we call a shield inside the network infrastructure itself and giving time to plan a full patch and upgrade cycle. And so that's something that's super interesting. So if anyone's interested in that, just Google Live Protect and you'll learn more.
There's a lot more I could go into, so I'll only share one more, which is the Campus Core of the Future was actually launched. So this goes back to the need for new hardware as we look ahead. And so we did launch the Cisco 9550 Smart Series Switch, which is a fixed switch, so super compact and powerful, that actually can go up to 400 gigabit for uplinks.
And it has eight times the routing capacity of previous generations. And so really just dramatically able for this high scale we expect of flows and traffic and more, and all of that is built in with all the latest security as well. So really chock-full of innovations across the board.
But the one thing to remember would definitely be Cisco Cloud Control as to how to make all of this consumable and built with agentic operations around both the networking and security from the beginning. I love it. Michael, we're way over time.
I apologize for hogging your time like this, but I think it was well worth it. I think our audience enjoyed it. Thank you for coming here on Techstrong TV.
Don't be a stranger. Come back. Keep us posted, okay?
Thanks, Alan. I appreciate the opportunity. All righty.
Michael Dickman, GM SVP for Campus Networking over at Cisco. Appreciate having him here. You're watching Techstrong TV.
We got more coming at you. Stay tuned.