AI Will Triple Your Network Traffic: Will Your Architecture Keep Up?

The Impact of AI On Network Traffic

Stephanie Hamrick – 00:00:01

Hello, and thank you everybody for joining us today. We at Bluewave are partnering up with experts from Comcast Business and Cisco to discuss how AI will triple your network traffic. Can your architecture keep up? And before we jump in today, I’m gonna help set the stakes for our conversation. So, Cisco and Foundry recently did a study where they found that AI driven demand is projected to triple network traffic over the next three years. This is not linear growth driven by your employees talking to LLMs. This is a compounded increase across generative, agentic, and physical AI.

And one more key statistic I want you all to keep in mind during the session, they surveyed more than 3000 IT leaders as part of this study and found that 73% of organizations are already facing or expect to face AI driven capacity limits in the next 24-months. So those are the stakes.

And today, we are going to unpack why these numbers are important and what you can do about it. And we have a great discussion ahead, but before we start, I’m gonna get into some of those housekeeping items.

So first, we are gonna aim for the discussion to run about 60 minutes today. This webinar is being recorded, and you will get a copy after the event. And lastly, there is a q and a section in our webinar interface if you’d like to submit questions as we go. This is an unscripted discussion today, so if we make time at the end of the session, we will get to those.

And if not, we will answer your question by email after the session. So, with that out of the way, let’s introduce today’s panel.

First, joining us from Bluewave, we have Solutions Adviser, Bob Schweiss. From Comcast Business, we have VP of Solutions Architecture, Joe Austin.

And from Cisco, we have Solutions Engineer, David Antonson. Thank you all for being here today.

David Antonson – 00:02:09

Thank you.

Bob Schweiss – 00:02:09

Thanks.

Joe Austin – 00:02:09

Thanks, Stephanie.

Stephanie Hamrick – 00:02:12

And I will be your moderator along the way. My name is Stephanie Hamrick, and I’m the Director of Demand Gen at Bluewave. So next, we’ll do some brief introductions for those who don’t know us, and we’ll try to keep these short so we can get right into our discussion. First, about Bluewave, we are a technology advisory, and our mission is to bring confidence and clarity to technology decisions by partnering with IT leaders. We have expertise across a wide range of technology areas from security to CX to cloud, networking, and a whole host of others with both our clients and advisers located across the United States.

Now let’s talk a little bit about our partners. I’ll talk first a little bit about Comcast Business. They are a leading provider of advanced technology solutions, helping businesses of all sizes adapt, evolve, and thrive in an ever-changing digital landscape. Comcast Business works across industries and around the globe. They design and deliver reliable, flexible, and scalable solutions that empower business growth. Those solutions include connectivity, networking, cybersecurity, and unified communications with a range of service models that help Comcast Business meet the needs of any business at scale.

And, next, I’ll hand it over to David to just talk briefly about Cisco.

David Antonsen – 00:03:42

Yes. Thanks.

Cisco is a worldwide leader in networking and is the critical infrastructure in the AI era.

Only Cisco brings together networking, security, observability, collaboration in one platform, plus the trust and the expertise and innovation to help customers like yourselves thrive in the AI driven world. And Cisco helps companies of all sizes connect, secure, and automate their digital infrastructure.

Stephanie Hamrick – 00:04:07

Awesome. Thanks, David.

David Antonsen – 00:03:42

You’re welcome.

Stephanie Hamrick – 00:04:10

So now let’s get to our agenda. Today, we have four sections that we’re going to cover. First, what does that 3x number really mean? What does it mean for traffic to triple?

Then we’re gonna cover what this traffic overload looks like in real life with some real-world scenarios.

Third, we will discuss what it means to design your network with these constraints in mind.

And lastly, we’ll talk about what you can do to prepare for that 24-month wall we heard about earlier when your network may reach its limit.

So, let’s jump into our first section here where we are gonna talk about what does tripled traffic actually mean.

And I’m gonna take our slides down here so we can run this discussion style. And our first question for the panel, and I’ll hand this one over to David first, is why are we even having traffic conversations at all? What is actually generating this load?

David Antonsen – 00:05:11

Well, thanks, Stephanie, and thanks, everyone, for joining.

So, one thing is the networks that we’re using today were really designed for predictive traffic patterns. Think of, like, Office365, Salesforce, SaaS applications, CRM traffic. Right?

And when we start implementing agentic AI to start automating a task requiring other systems in our organization or in the cloud, what we’re starting to see is these unpredictable traffic patterns.

So, suddenly, we’re starting to see the three AI agents are starting to talk to each other and even on your own network and solve a problem at machine learning speed, which is now over 400% faster than a human than doing the work and also happening all the time. So, this is going to be a big thing as companies start to realize their AI goals. And there is a widening gap between AI ambition and their network readiness.

And we’ve done from the survey, we’ve actually found that only 15% of organizations say that they’re ready and have networks flexible and adaptable to support AI at the necessary scale. So where we’re seeing this 3x, well, surveyed over these 3400 IT decision makers over 15 countries. And already what they’re saying is in the last twelve months, there’s a 34% average increase in campus and branch network tied to the AI workloads. Now, on top of what’s already grown 34%, there’s an expected 96% growth just in the next year.

So that’s just one year. Now what happens after three years is we’re gonna compound that and that traffic is predicted to reach 3x the current levels over in the next three years. And that’s gonna be compounded across generative AI, agentic AI, and also physical AI.

And then from that second point you said about the 73% were expected to face their capacity limitations, those individuals are very highly engaged in AI and they do expect to face those capacity limitations.

Now, this report that we’re showing is focused more on the branch and the campus enterprise, but we know a lot of you people on this call are running small, businesses and the conversation is still relevant there. There was another study done roughly around with 750 US business leaders with director level and above with companies less than 100 employees.

And we talked about their AI objectives.

And 62% of them were very confident in handing these high stake tasks to their agents. One in three of them considered AI agents mission critical to their company’s strategy, and 30% right now are actually using autonomous task execution, and that’s that traffic that we’re talking about.

And these are small medium business pilot workflows. Think of customer service bots. Think of Copilot, Salesforce, similar SaaS AI tools that are being built into these tools now to connect across your stack, right?

So, even Gartner even said that enterprise applications are going to include AI agents within these, right?

So, these are the points that we’re talking about and worth mentioning when this is really kind of where this 3x traffic increase is coming from.

Stephanie Hamrick – 00:08:23

Joe, Bob, anything to add on that?

 

Shadow AI and East-West Traffic

Bob Schweiss – 00:08:26

Yeah.

What I would say is it’s a little bit compounded too by the fact that there’s a lot of shadow AI use right now that corporations haven’t really been able to get their hands around.

So, when a company is rolling out an AI strategy, in a lot of cases they’ve done their due diligence and they’ve made sure that they have the adequate bandwidth to be able to support that.

What they can’t really anticipate is channel IT, right?

It’s people who are going out and leveraging tools that are just compounding to that consumption, right? So, that’s really what a lot of companies have been struggling with. It’s not only a bandwidth challenge, but it’s also a security challenge for a lot of companies. So, a lot of people are talking about that.

How do we get the proper governance around all of these AI tools to make sure that we can actually do some thoughtful planning around our network and make sure that it’s going to be able to sustain those loads as we go forward?

Joe Austin – 00:09:22

Yeah, and I would just add, Stephanie.

The one thing that we’re seeing right now with conversations with our customer base is not just the ChatGPTs, the Copilots, those intelligent assistants where you’re having an LLM query go out and in.  But it’s really that east to west traffic that we’re seeing now where AI agents and assistants are talking to other AI agents’ assistants and now you have that east to west movement across your organization.

So, not only are we seeing the amount of traffic going out to the internet, the amount of traffic going to the data center really changing what it used to be.

But now we’re even seeing kind of internal to the branch, the land networks are becoming more important than ever. And as you think about that east west traffic, security of that east west traffic also comes to mind.

So really what AI is driving right now, we are gonna see it as one of the largest network upgrade cycles that we’ve really seen in decades. So, when we’re talking to the end enterprise, it’s really about how can you be prepared?

How can you be ready for what’s next? And I think you made a really good point in your opening.

We see this kind of happening in two years as it’s about two years away when networks have to be ready.

But as the buyer starts thinking about “how do I get my network ready for that two year cycle?”, in the past when network administrators and network engineers and network architects were building their networks, they would think I need a five year plan.

Those plans are going to be much shorter now.

We believe that every two to three years you’re going to be seeing that network growth and you’re going to be seeing changes happening in your network.

So, building that redundancy, building that resiliency, building networks that can be easily upgraded are going to be in the forefront for all network architectures going forward.

 

Packets Per Second Versus Raw Throughput

Stephanie Hamrick – 00:11:21

Let’s talk about what some of that traffic looks like. Let’s talk about packets per second versus raw throughput. And how does this affect security and the firewall? A couple of you have talked about security so far. Joe, I’ll hand this one back to you.

Joe Austin – 00:11:33

Yeah. That that’s like the really nerdy bits and bytes of it, right?

And I think David made a really good point earlier that it’s not just a user talking to a machine using an LLM or something to that effect anymore.

When you have machine to machine connectivity, those machines, they’re going to exchange information a lot faster than what a user can. So, that packet per second count’s gonna go up.

And as you start having that packet per second account, you gotta start thinking, what is my memory? What is my utilization on my boxes?

It’s a different thing you have to think about versus just raw network traffic.

And so, in the past, as you were thinking about a building a scalable network solution, a lot of times you were just thinking what’s the total bandwidth?

You weren’t thinking about the packets per second but as a machine is talking to another machine to another AI machine and you have chains of AI machines talking to each other with data going back and forth, a lot of times those packets are gonna be really small.

And so that’s gonna put a lot of strain on a firewall that’s different than just raw bandwidth that you used to look at before. So, thinking about utilization, thinking about the network from a switching perspective, from an overall security perspective.

It’s just gonna be a different way to think about “how am I going to build my network for now and in the future”.

Stephanie Hamrick – 00:12:59

And speaking of that bandwidth conversation, I think a common misconception a customer might have is, hey. I have a two-gig circuit. I’m fine. Right? What are they missing?

Joe Austin – 00:13:11

I think from a com I’m just gonna speak from a Comcast business perspective to start and then I’ll I can pass it over to the other guys here.

When we’re having the conversation about bandwidth right now, one of the great benefits that Comcast is bringing is flexible broadband and Ethernet solutions to our customers for that bandwidth.

Being able to have broadband solutions that can scale over a gig of upload and download speed without having to replace a cable modem or replace any hardware on your premises is a huge benefit for our customers. And so, if they’re using a 200meg service today but need to go to a gig, it’s as easy as placing an order to make that change.

And similarly, from an ethernet perspective, we don’t have to worry about replacing hardware. It’s just an order in the system so to speak to get that additional bandwidth.

And so not having to have a new circuit installed and just having it be logical bandwidth changes is a huge benefit. Because a lot of times what you don’t know is when a user or an enterprise adds new applications that are gonna use AI on the network, they might not know exactly how that’s going to affect the network, how it’s going to affect the bandwidth, how it’s going to affect their firewalls, how it’s gonna affect their east west switches and things to that effect.

And so, sometimes you don’t know until you actually turn up the brand-new AI application and network administrators need to be ready to react to what’s happening from their software developers.

But I’ll hand it over to the other guys to comment on that as well because that’s a really good question.

 

Edge Device Capacity and Throughput Limits

Bob Schweiss – 00:14:46

Yeah, I think people underestimate what the capabilities of their edge devices are, right?

So, it’s very easy to call up a Comcast or another provider and have a gig circuit delivered to you, but the real question is, “are you able to consume that gigabit of connectivity”, right?

Is the firewall within your edge device capable of processing a gigabit of activity at a given time? And that’s where a lot of plans fail, right?

They think, “well, we have this large capital investment we made, it’s got a gigabit ethernet interface on it, therefore we’re getting gigabit speeds”.

And that’s not always the case.

There’s lots of other components that go into that edge that impact performance.

So, I think that’s where people are underestimating the most is, “what are the capabilities of my edge device” and even if I order a gig circuit, am I going to get a gig of throughput or am I going to be dumbed down to whatever my appliances keep capable of processing?

So yeah, Stephanie, there’s a lot more to let’s just call up and upgrade our circuit. If it were that easy, we’d all be out of work, right?

So, I think that’s where a lot of customers are underestimating. They don’t want to have to go back in and reinvest, right?

They think the reinvestment is simply by ordering a new circuit, it’s not. It’s really looking at the underlying technology and making sure that we’re able to consume everything that we’re subscribing to.

Stephanie Hamrick – 00:16:18

And that kind of leads really well into my last question for this section, which is gonna be for you, Bob, which is you talk to customers, a lot of different customers across many different providers.

What is that disconnect between, you know, what we’re seeing and discussing here and what teams think are happening?

And I think, of course, that starts with what you just mentioned is they think you could just get a new circuit. There anything else that you’re seeing where they’re just it’s not connecting?

Bob Schweiss – 00:16:48

Yeah. Well, honestly, the whole AI conversation the last probably 15- to 18-months, it’s sort of been caught in the fog of war, right?

Because there are so many facets to it. There’s “what’s the right tool?”, “What’s the right security posture to put around those tools?”. And unfortunately, bandwidth and network have been something that’s sort of been pushed down that stack a little bit, right?

A lot of people have been deferring those decisions, waiting on until they have a clear picture of what their AI strategy as a whole looks like.

But the key point with Cisco’s study is those days are coming to an end, you can’t ignore the network any longer. Because if you do you might have the best strategy for AI roll out, but if it doesn’t perform people aren’t going to use it, right? So that’s sort of where we’re at right now. I think a lot of people have sort of figured out what their AI picture looks like.

You know, they’re still in the early adoption phases trying to fine tune that system. But I think a lot of people now are starting to focus on that core infrastructure and say, “okay, things have to change”.

A big area around that is security. You know, we have a lot of customers who really just threw bandwidth and who threw hardware during COVID to get people connected, right?

But there wasn’t an overall strategy on is this the right solution for those people long term? I don’t think anybody post COVID thought that work from home was going to stick around. They thought people were going to go back to offices, and it was going be a much more finite problem to deal with.

But the reality is, work from home is here to stay for a lot of folks.

And now we really have to start thinking about how we can bring those remote users into our overall AI strategy, and make sure that they have the resources and the tools they need to be just as successful working remotely as if they were sitting in a corporate office somewhere.

So, those discussions are starting to happen today as the picture around AI becomes clearer and clearer.

Stephanie Hamrick – 00:18:42

Yeah. That’s a great real-world scenario, which leads us really well into our second section of the day here, and that is going to be talking about where we have seen this actually happen and what it actually looks like.

So, I’m gonna start here with a customer story that I know, Joe, you have, about seeing the consequences of this play out in real life at a restaurant chain.

Walk us through what happened there.

Joe Austin – 00:19:13

Sure. Yeah.

So, at Comcast Business, we are probably the largest provider in the industry of SD WAN and network connectivity services for the restaurant industry, specific quick serve restaurants, working with our partners at Cisco, of course, to provide those services.

And we had a customer of ours and they had Cisco SD WAN services.

They were running Meraki MX67 at their restaurants and they added some brand new AI applications to their restaurant environment. Three of them to be exact.

One for the drive through, one for, restaurant management in the kitchen, and then another one for, basically like an employee chatbot that was helping their employees understand what was happening inside of their restaurant.

So, they added these three brand new AI applications, not exactly understanding what was going to be happening across the network with network traffic. Right?

They did all the work to make sure that the AI application was gonna work correctly and they work to make sure that the data center side would be able to support the AI application.

But they really didn’t understand how it was going to affect the network. How it was gonna affect their local firewalls and their local SD WAN services or their cloud-based firewall services and how all of that routed in and out of their partner at the time which was their LLM was living inside of their cloud service provider.

And so, adding these three brand new AI capabilities to a single restaurant running on broadband, running on Cisco SD WAN, it really created brand new network patterns that the customer just really wasn’t ready for.

And I think that’s one of the key items here, Stephanie.

You talked about it in your opening, there is a huge unknown right now. And that unknown of what’s gonna happen to the network, what’s gonna happen to your firewalls on the security side is, I think one thing that’s going to, if it’s not already, is going to be keeping these network administrators up at night, right?

It’s understanding my application developers are bringing new AI tools into my environment. I have machine to machine AI talking now. What’s that gonna do to my network and I’m ready?

And in this case, we really outpaced what was happening from a network perspective with these brand-new AI tools that they brought in.

And so now the networking team, they had to react to what the application developers did for this restaurant chain and they had to make sure that they were ready. So, we really had to lean in, help this quick-serving restaurant company, make sure they were successful.

We worked very closely with Cisco on the SD WAN portions, but it was just one of those examples of new AI coming into an environment and then the SD WAN and the network teams having to react to that very quickly.

Stephanie Hamrick – 00:22:16

And, David, the Cisco research that we’re talking about today has a similar story about a retail organization and their VP working on an AI loss prevention tool that was defeated by some latency.

Can you tell us a little bit about that?

 

The Necessity Of Edge Computing

David Antonsen – 00:22:32

Yeah. Sure. And thanks for recognizing that.

Yeah. So, AI systems are really creating new categories of workloads that are, you know, not only distributed dynamic, but also sensitive to latency and reliability that many traditional enterprise applications don’t really care about.

So, we all know that, like, video is sensitive to latency.

So, like, when the video clips out, that might be a small inconvenience if it’s just you and I talking. But latency is becoming really the utmost important in certain AI driven workloads.

And this particular example that you’re mentioning, so an executive from a retail chain created an AI initiative for the company where there’s physical security across the store.

And so, they were losing millions of dollars every year in theft and decided to use AI to solve the problem.

So, what they’ve done is they created a model, they trained their models and wrapped that AI around what the camera sees. So if a person puts something in their jacket or maybe they walk to the bathroom with an item or put something in their pocket or even with becoming aggressive, right, that the AI tool was able to recognize that.

Now, in this case, it was a network latency that was a problem where this AI loss tool didn’t work properly and it ultimately failed because the network latency made the system operationally ineffective.

So, what eventually happened is, so that the tool is responsible for collecting and analyzing all this data from the camera and determining when a theft was occurring, but the network delay was significant enough that it just undermined the whole thing, right?

And the VP of infrastructure, he said, quote, “There’s a delay about five seconds before we know what really happens.”

And in that five seconds, that person could be already out of the store, right?

So, he said the whole system was kind of useless and it was really the network latency that brought it down, right?

When it comes to like architecting these things, a lot of people, you know how we’ve spent our whole life in this industry, bringing things, centralizing things, then distributing things and then bringing them back down to the edge. And I think this is one of those moments in time we’re bringing back things down to the edge.

AI is driving demand back to the edge where we need to have some of that processing power, some of that inferencing at the branch.

So, this particular example, the local network had to be upgraded. It had to be equipped with a local AI inference server here instead of going to the cloud. And everything had to be brought back to the branch in order for this application to be working instead of hosting in the data center in order to eliminate this latency.

So that was the use case and the problem, and it really came down to latency and bringing everything back into the branch and upgrading the local network to get everything working.

Stephanie Hamrick – 00:25:16

And it sounds like from hearing both of these stories that one of the big takeaways here is there’s a process of re architecting and not just thinking, oh, how can we can we fix this with another app or bolting on something else? Is that what our audience should be taking away from this? And I’ll let any of you answer.

Joe Austin – 00:25:40

Yeah. I can jump in real quick.

So, one of the things that we see is this cycle of where workloads live, right?

And so, every couple of years, depending on bandwidth cost and processing speed and things to that effect, you’ll see data and applications moving from the cloud to the premises to the near edge to the far edge and it’s very cyclical and it changes.

And I think what David was just saying is right now the edge is really becoming one of the most important places for AI.

For like the last two or three years, everyone’s been telling you, “Hey, go to the cloud”, “move to the cloud”. And AI is really changing that cloud conversation.

A lot of those AI workloads now they’re requiring near real time responses.

And so, having an AI application in the cloud, even if you have five seconds of latency, ten seconds of latency, think of it with a retail customer, right?

And a retail customer swipes their credit card or whatever they may do. And if you swipe your credit card and you sit there and wait thirty seconds for that transaction to actually go through, you’re gonna be like, man, did my credit card get declined?

What’s happening here? Why am I waiting thirty seconds?

They’re looking for that to be instantaneous. So, as you move those types of AI, those very fast responses needing that intelligence to move closer to where the data is created and where the decisions need to happen is a change that we’re seeing right now.

And so not only do network administrators and network engineers and architects need to start thinking about data flow to the cloud, how that’s actually happening, they need to think, “do I have to rearchitect my edge”?

And I don’t think that the future of AI isn’t just cloud versus edge, but it’s really how can we get the cloud and edge to work together the best possible way.

Bob Schweiss – 00:27:35

Yeah, I sort of like the analogy of data at the speed of life, right?

Another great example, Stephanie, was we had a large radiology practice with about 75 locations. Well imagine they do a CT scan or they do an x-ray. In the past they would store all those locally and then they would feed it up to the cloud for analysis. Well nowadays when you go into a radiologist you sort of expect to be able to walk away with information. It’s not the old days, where they take the film out of the device, send it off to be developed, you wait a few weeks and find out a result. Most people expect for a notification to pop up on their phone that has the imaging right there and already has a person who’s reviewed it.

So, it’s really sort of delivering on the expectations of the consumer, is everything has to happen in real time, right? It’s not acceptable for anybody to wait for a week for somebody to analyze document. And now specifically in radiology, there’s a lot of AI tools that are detecting things that the human eye can’t see.So, it’s even becoming more and more important that we’re living it real time. We’re not living off of a delay any longer.  And it’s just the expectation that people have now, right?  The same culture that wants their food delivered to them from McDonald’s in five minutes, right?

That’s the expectation that everybody is operating off of going forward. And companies have to step up and deliver on that expectation or they’re gonna lose customers.

 

Designing AI Ready Network Architectures

Stephanie Hamrick – 00:29:02

I think what you both, what you all are talking about right there really sets us up nicely to talk about our next section here, which is AI ready architecture.

And my first question was gonna be, you know, why is it not enough to have connections to Azure, connections to the cloud?

And I think we’ve answered that a little bit here, but I’ll open the floor again if anybody wants to add anything to that before we move to our next question. It sounds like the cloud’s far away and this stuff needs to happen in real time.

Bob Schweiss – 00:29:34

Yeah, I guess the one thought I had on that is we do hear a lot of customers say, well, we’ve already got connections up to Azure or to AWS or to OCI.  And usually, the follow-up question we have is, “yes, but do you have the right connections”?  Because there’s a big difference, right, between a site-to-site VPN tunnel up to a hypervisor and an on ramp into a hypervisor. One, you’re sharing a pool of bandwidth, one, have dedicated bandwidth that’s getting you to your applications. And that also plays for inter cloud connectivity too, right?

With AI tools rolling out now, people can be tapping a sales database at the same time that they’re tapping an HR database for a single request, right? So how does that east west traffic handle? How are you connecting your hypervisors together to make sure that your AI tools are delivering you the best possible experience? And those are the questions we seem to be having a lot more customers lately.  It’s now whether you have a connection, that’s an invitation to the dinner. It’s whether you have the right connection, right? Is what’s on the menu appealing to you? So yeah, the world is sort of transforming with respect to those.

And once again, not to harp on it, but security is playing a huge role in that too. I have two disparate databases sitting in two different cloud hypervisors. How do we make sure that we have the same security edge wrapped around both those providers even though they’re different technologies?  And then how do we secure that data loss and transport all the way out to our end users? So, it’s not let’s go buy this security solution for this, this security solution for this and this security solution for this…that doesn’t work anymore.

You have to have a single pane of glass in which you can see exactly what your security posture looks like regardless of where your data is stored at or where it’s being consumed from.

Stephanie Hamrick – 00:31:28

Actually, oh, Joe, were you gonna jump in?

Joe Austin – 00:31:29

Yeah. I was gonna say, I mean, it’s a good segue if I could just kinda add on that of how Comcast and Cisco work together.  And so, the one thing that Comcast and Cisco we Comcast is, I believe, the largest provider of Cisco SD WAN services in the United States today. And working together between Comcast and Cisco, what we have really found is that AI requires a more intelligent WAN, not just a bigger WAN, right?  Not just fatter pipes, it’s not bigger circuits.

It’s really as this AI traffic increases, what we know is that not all traffic is the same. Not all traffic is equal, so to speak. And so, you have some applications and some AI applications that are mission critical. You have some AI applications that require that ultra-low latency on the edge. You have some AI applications that involve large amounts of data.

I think earlier David was talking about kind of camera vision and some of these other AI applications where you have very large amounts of data. And then kind of what Bob just talked about. Right? You have some AI applications that have very strict security requirements. And I think this is where a modern SD WAN solution really becomes incredibly important. As you say, how do I build that more intelligent WAN, not just a bigger WAN? Kind of putting all of that together, everything we’ve been talking about so far. And the network must be able to intelligently understand those applications, prioritize the business-critical traffic, optimize the performance, secure the communication exactly what Bob just said across your branches, across your data center, across your cloud for those AI environments.  And we’re just not, I think Bob just said, “building VPN tunnels anymore”, right?

We’re no longer doing that. It’s just not a simple connection, but you’re really orchestrating this end to end AI experience.

 

Expanding Attack Surfaces and AI Security

Stephanie Hamrick – 00:33:26

And let’s talk a little bit more about that security piece that both of you have brought up, and I’d like to ask, to David.

I’ll put a quick stat again from the study on the screen. 77% say AI has expanded their attack surface in the last twelve months, and leaders now see the network itself as the most effective enforcement point for AI related risk.

What would you tell, David, leaders who are worried about this?

David Antonsen – 00:33:57

Yeah. Well, they should be worried. You know, one in four malicious breaches are now done with AI. And over 85% of small medium business breaches involve ransomware in some way. Right? So, it’s a highly repeatable process now, and I’m sure it’s going to get even worse. One thing that probably a lot of people don’t know is 40% of the top 100 vulnerabilities are against end-of-life gear.  And the truth is a lot of customers need help just addressing their old stuff more than a customer just starting up brand new. Right? So, the last day of support play is real, and it’s a real fuel, to the fire here when it comes to security for companies being hacked. Because there’s AI tools now you can just download that even helps the laziest of bad actors to weaponize. Right?

So, we have to help our customers close that vulnerability window. The time to explore it used to be years, and now it’s just down to minutes with the help of AI. So, it’s a time to be worried and actually start, preparing. So, a lot of these IT leaders in the in the report said that their existing security models are not keeping up with their complexity of the AI environments, not only from the packets perspective that we were having before that combo, but because the need for added security controls.

So, we have to protect not only our users, but we have to protect AI from AI.  And over 60% are not even scaling their AI initiatives any further because they don’t have their posture, security posture right. So, what we need to do is we need to create AI guardrails to protect not only the users from divulging sensitive information but also protect this machine-to-machine trust. And what a lot of people don’t realize now is identity is the new security perimeter, right? So, zero trust is a must.

And, you know, the truth is we can’t solve all the vulnerabilities for our customers, because they’re gonna have a lot of different things going on.  But what we want is we want them to have a really good foundation, so that the infrastructure that supports those things, number one, they’re sound, they can’t be hacked, and they can help our customers secure the rest of the estate.

So, what it really means is that we need to help our customers with this new approach, build resiliency, observability, and guardrails in their environment to kind of tackle these challenges. And again, 61% of the people that we talk to are holding back on scaling AI until they trust their security posture, which is a big deal.

Bob Schweiss – 00:36:32

One other challenge that we have Stephanie too around the security conversation is at the very same time that we’re trying to secure the world for AI, we’re also dealing with quantum computing.

And the ciphers that we’ve grown up on that we’ve used over the years are very vulnerable now. And it’s constantly evolving.

And that adds an incredible workload on the core networking equipment.

When they have to encrypt and decrypt and those calculations keep getting longer and longer because of the threat that AI poses, as well as the threat of quantum computing and the ability to attack those ciphers.

So, security is a much broader conversation than connectivity. It has to be part of the design. It cannot be an add on any longer.

 

Risks of Reactive Network Upgrades

Stephanie Hamrick – 00:37:24

So, environments have to evolve, security has to come with it. Let’s talk in our next section here about what happens to organizations that wait. And, of course, when we talk about wait, how long do you wait?

We started the session talking about how you’ve got 24-months to think of a to think of and plan and work on the solution here.

And before I hand it over to the group, I want to pull one more statistic onto the slide.

75% of IT leaders agree that there are higher long term costs from reactive upgrades and remediations, and I liked how this was put in the report.

“You can pay now on your terms, or you’re gonna pay more later on the outages terms.” So let me take this down and pause it to the group here.

What does reactive actually cost these leaders?

What do we need to be thinking about when it comes to getting ready now or just waiting until hoping that this AI traffic doesn’t affect you?

Bob Schweiss – 00:38:36

Well, I would make the argument that delivering a poor customer experience and losing a customer probably costs you way more than whatever you’re going to spend revisiting your security and your bandwidth.

You know, like I said, we sort of live in a technology and real time world where people expect instantaneous results, and when you don’t deliver on those expectations, that in the long run is going to cost you far more than a network upgrade or revisiting your bandwidth to make sure that you’re appropriately cloned.

So, I think the big question is, “what does kicking the can down the road and ignoring this do for my customer retention, for my customer experience, and for my ability to generate revenue off of my existing customer base”?

That’s really where the focus should be at, is delivering on our customers’ expectations and matching and exceeding those expectations, not necessarily on the cost of a new firewall or the cost of a circuit, right?

Those are things that should be relatively easy decisions for a lot of companies to make.

The other thing I will tell you is there’s an incredible shortage right now of memory and processing power in our industry.

And just like customers who did not heed the warning about POTS lines, and now are struggling to replace those on a real rapid basis, and probably throwing money that they didn’t need to spend at it to get the problem solved, they’re gonna have the same challenges.

And I don’t see in the foreseeable future that over the next 24-months memory is gonna become much more plentiful. And I just think waiting 24-months is just setting yourself up for disaster, right?

Because everybody else who procrastinated is going to be entering the marketplace at the exact same time too. Scarcity is going to be there. It’s gonna be much more challenging to manage these projects and roll these projects out if you wait till the last minute.

Stephanie Hamrick – 00:40:31

David, it looked like you were agreeing there. What do you think?

David Antonsen – 00:40:34

Oh, I think these IT leaders are fully aware of the financial and competitive risks because they fail to adopt other environment for AI driven demand. Right?

So, I think I think the wait is coming from a number of things.

Obviously, and expertise is always a part of it, but, they’re scared about falling behind and affecting part of the business process, like the inability to meet customers’ expectations quickly, what Bob was saying, missed business opportunities, not to mention the increased security risks that we were just talking about.

I think these are some of the reasons why a lot of them have been waiting for so long. But now they got to jump on it.

Stephanie Hamrick – 00:41:15

Joe, anything to add there?

Joe Austin – 00:41:17

Yes. Stephanie, I mean, I would say I mean, the answer to your question.

Or if to reframe your question, like, “if I’m an enterprise and I was they were to ask me, like, how much network capacity do I need”?

“What kind of firewalls do I need”? “What kind of SD WAN solution do I need”? Because of AI and what’s gonna happen to my network. What I would say is, today, like nobody really knows.

Right? You’ll go back just two years ago.

Two years ago, we were talking about teleworker solutions. We were talking about how do we support video conferencing and remote users and the SaaS applications and cloud connectivity to support those remote workers.

“How do we support cloud migrations”? Those were the conversations we were having just two years ago.

And now just two years later, the conversation is how do I support Copilot? How do I support AI assistance? How do I support this AI agent talking to this AI agent working on this intelligent automation platform.

And so, you have all these autonomous agents working back and forth. We’ve talked about computer vision today. We’ve talked about large language models today. We’ve talked about AI-powered security.

None of those were part of the network conversation just two years ago.

And now those are core parts of the network conversation and how I’m gonna build my network out. And so, I really think like we’re in the early innings of this baseball game.

We’re literally, it’s like the bottom of the first inning and we’re just understanding what AI is going to be doing to networks today.

And so, to kinda like, answer your question. Over the next two years, if I’m a network administrator, what should I be thinking about? What are those challenges?

And I think that’s where and I think we’ve said it a couple times today about building flexible networking solutions.

Building solutions that are agile, that are going to be able to meet those ever-changing needs. If I’m coming into a network today and I’m thinking, “where am I what’s my upgrade path”? “What do I need to do”?

I think agility and flexibility are like number one and number two on that list as I think about what is next for my network.

 

Strategic Planning and Network Agility

Stephanie Hamrick – 00:43:34

And that is great, I’m gonna put this stat up one more time.

That’s a great segue.

My next question is gonna be, what can you be thinking about today? What do you need to do immediately?

24-months, that’s one budget cycle, one procurement cycle.

These things come around much faster than we think.

So, we heard you answer that question, Joe, when you just said here, you gotta think about agility and adaptability. Bob. David.

Bob, I’ll hand it to you first.

What would you tell IT leaders? What can you do right now to start preparing?

Bob Schweiss – 00:44:07

Yeah, customers might not be ready to execute today because their AI vision might not be complete. So, they might not understand everything that’s going to impact what that design ultimately looks like. But my message would be, even if you’re not able to act today, you should have a plan.  You should have a good executable plan.  You should have a vision for how you’re going to transform your network over the next 24 to 36 months. Usually when you’re reacting to an application not working or an expectation not being met, you’re not making the best decisions in the world about what really should be happening in your environment, right?  And usually, they’re very expensive decisions that you’re making. Whereas, if you have a plan that you can execute and matriculate through your system over the next twenty four to thirty six months, it’s gonna have much less impact on your users, it’s gonna be more financially predictable and it’s gonna usually provide a better end result.

36-months from now you’re gonna have that agile adaptive network with a good plan. Without a plan, who knows where you’re gonna be in thirty six months. It’s like building a building without architectural drawings. It just doesn’t turn out well. So, my advice to people is even if you’re not ready to pull the trigger today, these conversations are worth having.  It’s worth sitting down with somebody. It’s worth looking at the financial side of it. It’s worth looking at the security side of it. It’s worth looking at the networking side of it.  We’ll make sure you have a good plan, that you can execute on over a prolonged period of time.

Stephanie Hamrick – 00:45:42

David, what about you? What would your advice be?

David Antonsen – 00:45:44

Yeah. I mean, so I guess my, you know, the question I would ask your customers is, “where are you currently with AI”? “What are your plans”?  If you’re deep into it, then you probably are aware, already know, or maybe were even part of that study, right? Again, 40% of those top 100 vulnerabilities are due to end of support equipment. So, I just kind of look at that, the most vulnerable part of the network. Start there. And next, see where you are with network traffic now.

Do you need, I know we talked about how bandwidth doesn’t solve all the problems, but sometimes it does. Because you might have 100 meg link down, but only 10 meg link up, right?  So, it might be asymmetrical traffic.  Well, maybe you need to call Comcast Business and make that symmetrical, 100 meg up and 100 meg down, because now we have all these processes going back and forth, so bandwidth could be a concern there.  Do we need to upgrade the WiFi?

In this study, they said 50% of the actual, basically, load was coming from the Wi Fi network.  This is just because of how users connect to the network. So, do we need to upgrade to Wi Fi7 built in security for capacity? And then, of course, we talked about before security. We want to create these AI guardrails for your own company to protect users and machines. So, these are the things that you really want to be thinking about. And if you’re already behind, then it’s time to catch up, but it’s not too late to start.

Stephanie Hamrick – 00:47:08

Yeah. I love the way you phrase that. And, Joe, I’ll hand it back to you again.

I know that you kind of were talking about what you would do, what you would advise, but I would like to give you the opportunity if you have anything else to add.

Joe Austin – 00:47:21

Sure. First, I wanna add, I did not pay David to tell him that you need to upgrade your Comcast bandwidth. I did not have anything to do with that part of what he said. But no. All joking aside.

Yeah. I mean, ultimately, think where we’re all going and kinda where all this is trending towards is saying like the biggest risk here isn’t underestimating AI or the biggest risk isn’t your organization is going to be bringing AI in. It’s really underestimating what AI is going to do to your network. And so, understanding can my network adapt? Can it evolve to what’s gonna be happening within my organization? And going back to kind of how Bob outlined it right at the very beginning of the call, right? Is if I’m going to have more network traffic, if I’m gonna have different network patterns on my network, can my network support that?

So, if I have to have double the network capacity, can I support it? If I have triple the network capacity, can I support it? And if I can’t today, do I at least have a plan to get there?

I think the plan is just as important as having the network because nobody wants to go and spend hundreds or thousands of dollars to overbuild a network today.

But I need to have the plan and have that flexible architecture in place. So, when I need to increase that capacity, if it does double, if it does triple, then I’m not caught flat footed, then I’m ready to make that jump with the AI applications that are coming into my enterprise.

 

Network Readiness Assessment

Stephanie Hamrick – 00:48:58

And all three of you have touched on it now.

So, I’m gonna wrap us up here with a few thoughts, which is that you need to have that plan in place.

That’s the most important piece. Understanding what’s in your environment, understanding where you need to get to, what it could take to get there, and how to actually get there.

Those are things that we all can help you with.

So, if you’re not sure where your network is right now, if it’s AI ready or even just future ready, if it’s ready to evolve to to Joe’s point, you know, in two years, we maybe not be talking about AI.

We’ll be talking about something else. So if you need help figuring out if your network can get there or if you can plan the investments to get there, those are all things, again, that we can help you with.

So, we’re all partnering together to offer our audience here a network readiness assessment. We’ll assess your environment, which includes looking at things like traffic and utilization, infrastructure, wireless, your segmentation and IP, and, of course, also security hardening. This will help us uncover any gaps that could turn into bottlenecks as AI driven traffic growth or whatever we’re dealing with in 24-months, grows exponentially across that time frame. So, we’ll help you find clarity on what to do next by working with you to translate those gaps into an AI or IT investment plan that ensures your network is ready for AI and your organization gets where you want to go.

So, with that, I’m gonna give a big thank you to everybody in our audience for joining us today.

I have a QR code on the page here where you can go if you want to read the Cisco study that we’ve been referencing today or if you want to learn more about getting a network assessment.

We’ll also email you a link to that page in our follow-up email for this session. If you don’t like scanning random QR codes, you’ll still get that information there.

And so, we do have a few minutes where we can jump into the q and a, and we have a few q and a questions submitted here.

This first one is for Joe, and it’s kind of long, so I’m gonna paraphrase it for you.

But this person is asking if you could go into a little bit more detail about how you helped that restaurant chain when they ran into those AI problems. How did the network adapt?

How did you tag, identify, and act upon the traffic?

You needed to scale the beat bandwidth by how much? Can you talk a little bit more about some of those details?

 

AI Traffic Management

Joe Austin – 00:51:35

Yeah. And, actually, the solve for the problem, was we really said, how do we take it off the network and how do we have more responsive?

The solution was actually bringing compute to their premises.

In Comcast, we put a edge compute system inside of the restaurant where we were able to run these AI applications locally versus running them in the cloud.

So, the network was a part of it.

The network was basically setting it up to make sure we had the available switch ports. We had the network segmentation, everything that you would need to be able to bring those.

But the biggest change was running those AI workloads locally on edge compute versus running them in the cloud.

So, it was thinking about the edge architecture at the same time of thinking about how to have the AI applications run closer to the user.

Stephanie Hamrick – 00:52:32

Awesome.

And then another question here, and I’ll open this one to the group for anybody to jump in.

How do I figure out what AI traffic I already have if my teams are experimenting everywhere?

Bob Schweiss – 00:52:49

A lot of that’s gonna be depending on what visibility you have into your current network infrastructure today. If you don’t have a good tool set today that lets you analyze that data and understand exactly what’s traversing your network, then it’s gonna be a challenge.

That’s where a lot of times a 3^rd-party assessment could come in and they can look at those traffic patterns across your network and they can help you make an informed decision about, “hey, is this normal traffic?” “Is this abnormal traffic”?

Is this something that’s going to scale and become more of a problem? Or is this just a one-time anomaly, right?

And it helps you look at those patterns and decide, how’s my network being utilized? Is it being utilized appropriately?

Part of those assessments also is a good hard look into security too.

So really if you don’t have a good single pane of glass that provides visibility into your network, my recommendation would be work with a company like Bluewave, work with a company like Comcast, Cisco, to come in and do a network assessment and help you untangle that picture.

David Antonsen – 00:53:52

One thing I wanna add to that is Cisco had a product called Secure Access, which is a full SSC that can detect shadow AI and that’s exactly what this question is.

That means you’re gonna know who went to chat GPT, who’s using Claude, and what you could do is you could put guardrails around them so that they don’t accidentally do something like divulge personal private information, effectively really securing the user and your data, right?

So, think of like HIPAA compliance, patient records, social security numbers, data loss prevention.

That’s one of the key assets within secure access to deal with AI to make sure you don’t divulge, company information.

Bob Schweiss – 00:54:29

Yeah.

 

AI SaaS Tool Integration

Stephanie Hamrick – 00:54:31

Awesome. Okay. Another question here. Does this conversation apply if our AI use is just SaaS tools like Copilot and not custom apps?

Bob Schweiss – 00:54:44

Well, so yeah, an AI chatbot. The power of an AI chatbot is not the raw data off of the internet.

It’s connecting that AI chatbot to your data sources.

So, there’s context around the results that you’re looking at.

So even though people say, “hey, I’m using Copilot”, really the next evolution will be, well, what is Copilot using as the source of its data, right?

And that’s when the picture gets a little bit more tangled.

If you’re going out and using Copilot to look up recipes or to help you figure out a shortcut in Microsoft Excel, yeah, there’s not a tremendous amount of value there, but when you empower Copilot to go look at your Salesforce instance or to look at your HR platform or to look at your sales tools, that’s when the conversation really becomes much more relevant.

I don’t see a lot of people standing still, I don’t see a lot of people adopting AI posture of “we gave you copilot and that’s the end of the story”.

I think it’s the very beginning of the story.

So yeah, if all you’re ever going to use is a generative AI tool to go out and do a smart search on the internet, this conversation’s not probably very relevant for you, right?

But if you’re going to actually embrace AI and use it to create efficiency and automation inside your business, then it becomes much more relevant.

David Antonsen – 00:56:15

Yeah. The, and these Copilot and SaaS AI tools, right, they’re really becoming themselves increasingly agentic under the hood. They’re doing background syncs, connector calls, autonomous actions across your staff just like what Bob was saying.

And, you know, these are not custom built apps.

These are just the traffic and security, the applications that you’re using already, already having built AI tools to connect into the data that, you wanted to manipulate.

 

Upgrading Network Infrastructure

Stephanie Hamrick – 00:56:45

Awesome. Alright. One question here that might be a little specific, but I’ll ask it to the group.

Most of our sites are one gigabyte circuits, and our Meraki firewalls can handle the speed.

Most of our internal networks are still at one gigabyte switches. And next, are we needing to bump this up to handle more of the east west traffic discussed?

Bob Schweiss – 00:57:09

I’ll make a few comments.

So many of our customers that we see putting servers today into their enterprise environments, a lot of those servers only have ten gig ports on them. It’s something we’re running into more and more frequently is that those interfaces on those servers are not one gig ports.

Some of and I don’t know what kind of Meraki switches you have and some of the Meraki switches have four or so.

Even if they’re a one gig switch, they have four or so ten gig ports on them. And so, you do have some flexibility there if your servers are ten gig ports. But a lot of it just depends on what AI applications you’re putting in, how much traffic you’re going to be originating across those AIs.

But coming back to it, I think it’s about thinking about “how do I have that flexibility within my network.”

So, it goes back to that being the underpinning statement here is that flexibility. And so, if all of your switch ports today are one gig Ethernet switch ports, I would definitely start thinking about how am I introducing 10gig switch ports or maybe even 25gig switch ports.

A lot of these higher end servers we’re seeing now from Dell and HPE have 25gig interfaces on them.

And so it’s really just thinking “where am I going and what am I gonna be needing in the future and building out that flexibility to be ready”.

So, when your application team comes and says, we have this brand-new server we’re gonna bring into the premise and it has 25gig interfaces.

You’re not sitting there saying, “oh no, my switch ports are all one gig”.

I don’t have any 25gig switch ports available in my enterprise today.

Yeah, and I guess my comment would be, what are you using your Meraki devices for? Are they just a router? If they’re a router, yeah, you’re gonna get your one gig of throughput. If you’re using the security elements of the Meraki, you’re gonna get much less throughput because it’s creating more overhead on that device to do inspection on the packets that are traversing that device.

So even though you may have a physical device that has a one gig report, it doesn’t necessarily mean that’s the throughput that the device is delivering to your user experience, right?

That can be impacted by security and definitely once you start talking about wireless, there’s a lot to talk about there too.

But just because you have a one gig internet connection, if you’re using an outdated wireless protocol or an older standard, it doesn’t necessarily mean that that is what the user is experiencing, right?

So, it definitely makes sense to, like I said, have a plan, sort of look at this holistically rather than just what are the raw capabilities of the device itself, because those might not paint a very accurate picture for exactly what the end user experience is.

Stephanie Hamrick – 01:00:03

Awesome. So, we are very close to time here, so we’re gonna make that our last question.

I wanna give one really big thank you to Joe, Bob, David, for joining us today.

I really enjoyed this conversation, and all of you are so knowledgeable and shared a lot of great invite insights and advice with our audience here.

So, for our audience, thank you also for joining us.

We hope you learned something new today, and we hope to see you on one of our next sessions.

With that, one last thank you to everybody, and I’ll say goodbye.