AI in the Network, AI on the Network - with Iain Gillott, WIA

Speaker A: Welcome back to Industry Therapy. Today on our show, we are joined by Ian Gillett, Vice President of Innovation and Technology at wia, to talk about AI both in and on the network and the important distinction there, as well as what it means for our society. Looking forward, Ian has also been a special guest speaker on our show multiple times.

Speaker B: This podcast is provided by the Telecommunications Industry Foundation. This podcast as well as much more additional content can be found@tifonline.org Please welcome your hosts, Michelle Kang and Scott Stecker.

Speaker A: Welcome back to the show, Ian.

Speaker C: Hi. Thank you, Michelle. Good to see you.

Speaker A: Now we hear a lot about AI, and we hear about AI ran, inferencing, hyperscalers, etc. How does all of this relate to mobile networks?

Speaker C: Uh, yes, good question. And, um, it seems that you can't turn anywhere without something being AI enabled these days. Um, um, I bought a new pool cleaner this weekend, a robotic pool cleaner, and it's AI enabled because as it goes around my pool, apparently it's mapping itself and recognizing dirt. So literally everything from the pool cleaner to everything on your desktop has got AI these days. Um, but you're right, um, it seems we're kind of surrounded by it. Um, the thing you've got to remember is quite simple, is that to use AI, you've got to have a connection to the hyperscaler class, because that's where everything runs. ChatGPT, um, Claude, Gemini, etc. Etc. Um, so if you're sitting at your desktop, then you're probably using WI fi, and then maybe a fiber connection from your home or your building up to the cloud. But increasingly, of course, a lot of people sit on their mobile phones and connect up to the cloud. Um, so those mobile networks, if you're using 4G or obviously increasingly 5G, most of it's 5G actually these days. Um, and, and you may be, uh, doing whatever you're doing, asking ChatGPT something or Gemini or Claude or whatever you're doing, um, that request is going over the mobile network, connects up to the cloud, the hyperscaler data center gets processed, the AI comes up with the answer and then sends all that information back back over the mobile network to your device. And so when you look at those data centers are very big, um, but they don't work very well without those connections to them. Fiber Tower, Wi Fi, 5G, et cetera, et cetera, et cetera. And so that's where the wireless infrastructure becomes part of the AI infrastructure, because you need that connectivity.

Speaker A: Okay. And can you just maybe take that a Little bit further and talk about the distinction between AI on the network versus in the network and what that looks like.

Speaker C: Yeah, so there's different versions of this. Some people say, you know, AI next to the network. I always question that one as the left hand side, the right hand side under the network, over the network. Um, but basically the simplest form is in the network and on the network. So on the network would be that, um, just like I said, um, let me think of an example. Um, there's an ad that Google has for Gemini where somebody is looking for flowers and they want something for uh, dinner party I think, and something that will last two or three days and they hold up their phone and look at the florist and Gemini kind of circles. You want this one, right? Um, so that is an example of using the mobile connectivity in order to process that request. So what happened there was the phone took the image of the florist's uh, different flowers, sent it over the network. It was on the network, on the mobile network, went up to the hyperscaler, it processed it and came back on the network to the phone and used it. So that's when the traffic part of our mobile data traffic, which also includes obviously voice and um, teams like we're On now and FaceTime and text messages and everything else. Um, part of that traffic is now AI traffic. So it's AI traffic on the network. Okay, um, AI in the network is when the mobile operators, the equipment vendors, uh, uh, all the professional services that combine within the mobile network ecosystem to build and run that network. When they starting using AI tools, it could be a variety of different things. But basically we're now using AI tools in order to run that network and operate the network. Okay, so that's AI in the network. So within the network itself we've got some AI capability. Um, it could be deciding to power down a cell site at a certain time to save power. It could be reorienting an antenna that's out of alignment to get better uh, efficiency or coverage. Um, it could be engineering designs, it could be digital twins looking at the tower equipment and all those types of things. So variety, wide range of things if actually include um, using AI to process amendments and permitting requests for the towers to build the equipment in the first place. So uh, so that's AI in the network.

Speaker B: So there's another way to kind of phrase this. Like AI on the network is basically anything that an end user is consuming data wise. And then the other part is just kind of enhancements and things that the network Needs to survive, so to speak.

Speaker C: Yeah, yeah. Um, uh, I'll go a little bit further than the end user, uh, because the end, well the end user could be a thing, right. It could be um, a video camera on, ah, a uh, factory or warehouse or whatever that's monitoring the situation or looking for a fault on a production line. Right. Um, so that's again connected to the network, going up to some processing, coming back. Um, it could be a person as you said, it could be the example I gave. But yeah, anybody's basically got an AI application or service that needs that network connectivity in order to um, connect to its compute, uh, power.

Speaker A: Okay, in looking at AI traffic today, what impact does it have on the M Mobile networks? And how is this changing?

Speaker C: Yeah, well, quickly. So what we did last year, uh, it was kind of an interesting exercise actually. Uh, but towards the end of last year we actually looked at AI traffic on the network and how much of we know how much traffic goes over the network. Right. Um, uh, there's a lot of uh, measurements for that and things like this. Um, and you can look at all the users and look at how much data people use per day and calculate it all out. It's pretty straightforward. Um, the carriers also put out lots of information. What we did was actually look at how much of that was AI traffic. So how much of it was due to an AI application or service. So um, it could have been the person at the flower shop, it could be that security camera, it could be a robot, a factory or whatever. Right. Uh, and what we came up with was 4.2%. So 4.2% at the end of 2025 of all of the US mobile data traffic was resulted from AI applications and services. So when you look at that in terms of how much spending that is, if you look at what it costs to build and operate the networks is just under $3 billion a year. Okay, so now the interesting thing is, well, two things with that. Uh, one is it's growing very quickly and I think at the beginning of the year we said, hey, probably by the end of this year we're probably closer to, I think we'll be over 10%. I believe right now it's somewhere in the 6 to 7% range. Um, because the more people use flawed Gemini, ChatGPT, et cetera, pool cleaner, that pool cleaner, by the way I use it. It's fascinating. It is actually cloud connected. Um, when you set it up, you actually set up an app on your phone and control it. Even though it's in my Pool, it's got to come out. But yeah, Michelle, you're shaking your head, but it's kind of crazy, right? So that pool is not contributing 4.2%. It's probably 0.01%, but it is something. Um, so that's the first thing. It's growing very quickly. We'll look at again at the end of this year. And I said my estimates, it'll be. If, uh, it's not 10%, it'll be over. Um, uh, and then the second part of this as well is that, uh, people say, well, the mobile data traffic didn't grow by 4% last year. Well, no, I didn't say it grew by 4%. I said it was 4% of the total. So some of the traffic is getting, the existing traffic is getting replaced by AI applications and services. So, for example, if you take your phone and you do a Google search, right? That was traditional web browser search traffic. Now that may be an AI search, AI prompt. So instead of using the Google search, you're using Gemini, right? So now we kind of change the bucket. We're putting the data into. Um, and so it's not that the mobile data traffic grew by 4%, it's that 4% of it was from AI. And just to give you an example, we didn't look at the m number in 2024. The end of 24, I kind of went back and did an estimate of what I thought it would be. It's probably under. It was definitely under 1%. Um, um, so it's growing very, very quickly. And even the last six months it's grown enormously.

Speaker B: So is the overall mobile data growing at a faster rate than 4%, or is it slower than that?

Speaker C: Uh, it's, uh, mobile data as a whole. Uh, the general rule is it grows by like, uh, uh, is it 50% every two years, something like that? Um, uh, I can't quote the numbers right off my head, but it's still growing significantly. Um, most of the traffic on the mobile network is video and tech and. Sorry, and imaging. Right. Because it takes a lot of, obviously a lot of data to do that. Um, what I find interesting for the AI tools is that the ones that are going to really contribute to that growth and take us up to 25 and 30%. Are those examples where you take your phone, point it at something and say, what is this? Right. That means that image is. There's an image there that's being uploaded or to be processed and come back. And that's much more data than asking Chat GPT. Hey, what time is the next bus? Right? Or something like that, right? That's just text. That's just bits and bytes. It's easy. But the video, when you start doing video, uh, so taking a video on your phone and then asking, AI, hey, can you just clean this up, uh, and make it look like, you know, I really did mountain bike off that jump, put my face in there or make me look like a really good skier, please? Right? And you upload that, it processes it and brings it back. That is a ton of data. Um, um, the other difference with this is we do have a lot of video and text usage today, sorry, video and data image usage on the phones today. But most of it's downlink, right? It's pulling up your phone and watching Apple TV on here, uh, or Netflix. It's, it's downward traffic. But when I start taking video and images and uploading them, uh, and asking it to be processed. Now I'm increasing the amount of uplink traffic. So you hear the operators and the equipment guys talking about this split between uplink and downlink and how it's shifting and the uplink is increasing. So, uh, the pattern of traffic on the network is shifting as well.

Speaker B: Right. And kind of, you know, talk a little bit more about the design, building and running of the networks. How are some of the ways these tools are affecting the network and helping the, you know, design and operation of the network itself.

Speaker C: Uh, let me think. I was, I uh, was on a call last week, we were talking about digital twins. So increasingly, obviously you've got a tower. On that tower in the U.S. you could have two or three operators, right? Uh, you have older equipment, newer equipment, etc. Uh, so what they'll do now is when we have site changes, they'll take a drone, fly the drone up and down the tower, build a digital image, a digital twin of the tower. So then you can say, okay, we now need to move, remove this antenna, uh, this radio, upgrade this. When we climb up there, we've got to bypass this. Looks like the power equipment needs upgrading, etc. So think about that. You flew a drone up to take images of the tower, and then the AI took those images and worked out what all the equipment was and said, okay, this is a radio, you know, 8792, this is etc. Etc. Um, so that's been increasingly. It's not even flying a drone. You can stand at the bottom of the tower with a high def camera and take a picture of it, which is uh, it's pretty cool. But I heard last week that uh, of an example where the AI actually, uh, noticed that a bolt was missing. Um, so an antenna should have been put up with four volts and one was missing. And uh, the uh, when. When the vendor puts up new equipment, they have what's called a closeout package where they have to show before and after and what was done and account for everything. And they've missed this bolt. And uh, the. One of the vendors, the other vendors, contractors found it and um, so they had to go back and fix it. Um, so, you know, minor thing, it seems like minor things like that. But that's when you get into health and safety. There's a reason there's four bolts and not three on those things. Um, that's one example. Another one is we always. In the tower industry, uh, there are a lot of buy and sell of tower portfolios. Goes on all the time. Tower, uh, companies are buying and selling between each other, between investors. Every time they do that, there's amendments or contracts, there's the ground lease contracts, there's the permits to require upgrade and things like this. Just think of the paperwork. It's quite significant. And every time you do a buy or a sell, you're going to review everything. So using AI for that, to review all the documents, highlight potential areas, uh, double check things is increasingly happening as well. So that's kind of the, you know, the tower end on the network side. Um, it could be, as I said, moving antennas, reorienting, reorienting antennas, looking at maintenance schedules, uh, looking at the performance of specific equipment when it's about to fail. So before it fails, you go out and look at, um. Uh, could be uh, scheduling the trucks to go out there and optimizing and all these types of things, the schedules and the routes they, uh, drive. And also the weather comes in as well. Increasingly looking at the weather and deciding, okay, we may have a problem in this area, let's move some trucks into that area because, you know, a hurricane's about to go through, we need to react quickly, things like that. So it really is everything from the network operation itself all the way through to that, the, the legal processing and the engineering side.

Speaker B: What about the actual decisions of where more capacities needed more coverage? And then the actual design of the rf, for example, is it getting into those type of things?

Speaker C: It does. Um, I would say that it's. Right now those are tools they are assisting because that tends to be a little bit of a, um, there's kind of an art There, Right. So hey, I need to, I think I need a tower over here. Okay, we're gonna, we need it here. Well guess what, that location's not available. We've got to move it over this location. Um, but certainly on traffic patterns and analysis of traffic patterns, predicting when the load is going to be, we've always had rules that you, you know, you build to 85% of your peak load and all these types of things. So that's all getting refined now. So it's not just a 85% rule across the board, but it's a lot more precise than that depending on the local situation. Um, but we're not there yet. Where the AI actually build, designs the network and says go put this here and this here and this here. Um, and then runs it and not there yet. And it is yet.

Speaker B: Someone's probably working on that.

Speaker C: Oh absolutely, they absolutely are. Um, uh, yeah, that's just a matter of time before you know, increasingly you've got somebody overseeing the AI design rather than uh, today they're using the AI tools as the engineer designs. Over time it'll be the engineers overseeing the AI, uh as they design. It's a different thing.

Speaker B: And what about uh, just anything we missed? Any final thoughts on this topic?

Speaker C: Um, we've got a long way to go. Uh, it gets fascinating when you start to think of the potential. Um, and I'll tell you a question I get asked a lot is what will kick it, what will kick it from 4% to 4 to 10 I think is a matter of growth. But what takes it from 10% of mobile traffic to 30 or 40%. And actually my answer is when you take your iPhone Siri right now and she's going to jump. She's, she started up. Siri, um, is not very smart, don't tell her this but she's not very smart. Um, she's based on last generation technology. So you can ask Siri, play this music, you know, etc. But Apple's announced they've been working with uh, I think they're working with the Gemini model, um, uh, from Google and to build the next generation of Siri that is AI powered. So now you ask Siri, hey, can you take that photo I took yesterday and can you just remove grandma, um, and put her from the left hand side to the right hand side or something or whatever it is, right. And be a lot more sophisticated about it. Now the reason I find this interesting, there's nothing there you can't do on uh, uh, Google devices and things like this. But when that rolls out on, um, iPhones in the US overnight. Overnight you can hit 30%. You get a hundred million people suddenly using that capability overnight, right? That's when you get a step up in what to do. That's when people like my wife who don't want to use AI won't go to on the desktop, but suddenly she does use Siri, and her interactions with Siri become a lot more sophisticated. So without her knowing it, she's using more AI. And I think that's when we see that, that step. And when is that? We don't know, but it's. It's not five years away.

Speaker B: So the, uh, equipment manufacturers and app developers, do they work closely with the carriers and the operators to plan for these things or they just kind of.

Speaker C: No, they do. Uh, I mean, we've, we've. I spent a lot of time talking to the operators. When you talk. I'll give you another example. Um, we've got the World cup coming up, right? So, um, I was at a 49ers game last year, and, you know, the amount of images and video you take and upload, etcetera, and you may sit in your seat and order some food, right? And it's delivered to you. Things like this. That's what we do today, right now. Now you take that to an AI world and you start taking video of the game. And then you say, hey, AI, uh, can you just put me in a jersey and make me stand behind Brock Purdy and make it look like he just made the pass to me? Right? And you get the video down while you're sitting in the stadium and go, yeah, that's really good. Can you actually make me a little bit taller so I look bigger than him? Okay, great, thanks. Now, can you send it to everybody I know? And you're doing that while you're sitting in your seat, Right? That is a lot of power in the network, the processing bandwidth. I mean, that the carriers are, uh, um, it kind of scares them a little bit. But having said that, they're preparing for that level of interaction because it becomes part of the game experience, the stadium experience. You don't want to go see Taylor Swift and not take photos and video and tell everybody with it. It's part of the whole experience of being there. And so that's what these networks have got to deliver. So, yeah, they look ahead.

Speaker A: Well, Ian, thank you for being on our show to talk about AI and how it's being used both in the network and on the network, and how that use is growing and what it might look like in the coming years.

Speaker C: Okay, thanks. Good talking to you guys.

Speaker B: Thank you.

Speaker A: Thank you for listening to Telecommunications Industry Therapy, presented by the Telecommunications Industry Foundation. The views, information, and opinions expressed herein do not necessarily reflect the views of tiff, its board, or the hosts. These podcasts are not meant to supersede regulations, standards, or AHG requirements that govern the reference subject matter. Rather, they are offered to help educate listeners so that they may gain information that will lead them in the search for additional information and opinions. To view more TIFF podcasts and other educational content, to provide feedback on an existing podcast, or to submit a potential topic for a future podcast, please Visit our website tifonline.org.