Why AI could change mobile networks forever

Great to be here, Arjun. Thanks for having me.

Yeah, look, today we're a digital infrastructure provider, right? We provide technology across all of the core elements that are enabled by the AI super cycle, data center, you know, to the edge ultimately with what we're building in AI-RAN, which we've made great progress on over the last few months in our partnership with Nvidia. You know, that's our journey enabling everything from agentic AI to physical AI to powering the service, you know, the service providers, the hyperscalers, and even delivering some new innovation around new applications and services that weren't previously possible.

Yeah, absolutely. So, there's two things, Arjun, that we do to keep it simple. One is all of these data centers, these AI factories, they have a ton of networking equipment, IP networking equipment, and that's something that Nokia has been providing for years. It largely came through the acquisition of Alcatel, but Nokia's built on that, continued to invest, and that layer is often talked about within the compute infrastructure. But as importantly, it's what provides all the connectivity for scaling. So if you hear about AI factories, that's networking that's providing that, or if you hear about data center connectivity, we call data center interconnect, that's also enabled by networking. We call it wide area network routing. The other part of that that we provide is the optical network, and that's actually the layer that connects these big networks together, these big data centers together. And from that perspective, you know, we're one of the few players in the world that has advanced optical technology connecting these data centers. And what's changing so rapidly is the importance of speed and bandwidth and latency and of course security of these networks. So connecting these data centers in real time because more and more as AI becomes prevalent in software development, in agentic bots that are running parts of networks for our traditional telco customers to financial services to other applications, other vertical applications, to what's coming with physical AI, autonomous vehicles, it's not enough to just have the AI model give you an answer. You need it to give you the right answer at the right time so that you can make the right decision or take the right action. And that means more bandwidth, more connectivity, and all of that is powered by Nokia optical and IP networks.

That's right. And that's that connection of IP networks and optical. And to your point, Arjun, if you think about, you know, in the internet era, you know, like I'd call a web page or even if I'm streaming a video, right, that's largely going to one data center and there's maybe something at the edge that's helping buffer, serve that, you know, serve that internet access, but it's usually pretty monolithic. I go up and hit one site, maybe I go hit a couple places because I'm accessing some data. In the AI era, I've got agents that need to take actions on processing and inferencing data. They need to then make a decision, take an action. They may kick off another query. All of that requires multiple access points. And so that means these data centers need to be very, you know, connected to each other. And that's something that we hadn't really seen in the internet era. But that's a big part of what we're seeing today in data center demand.

Yeah, I think first of all, if you look at where we are, the demand we have is still very early. I mean, you know, again, AI is quite early. So, you know, yet it's quite early, but it's a third of the internet traffic we see already. So it's already massive in terms of the amount of data it's consuming. It's much richer than what we've seen in prior. And it reminds me, we talked about Nokia on the phone days. It's a little bit of the same experience, right? Looking back at history of, you know, the internet was so much more data than what we had seen with SMS or, you know, mobile phones. So we're seeing that step function of data. I think we'll see the same thing with AI and that's why we feel pretty confident we're at the beginning of this cycle. Having said that, if you look at the value that AI is delivering, I mean the value, you know, for companies like us in terms of accelerating software development, providing greater intelligence and insight on decisions we're making, the value is also there as well. And I think what's different about the AI super cycle is this is much more of a business or enterprise cycle at the beginning versus a consumer-driven. We all remember the dot-com era being very much about pets.com or consumer information, but I wouldn't discount that. So the business value is there. And then the other thing I would say is when you look at what's happened over the last couple years, we've been in a situation of persistent constraint. So we've had energy constraint, we've had silicon constraint, we've got supply constraint across other parts of the curve and the ecosystem. So from that perspective, I don't think you can say in any manner we're overbuilding today because reality is that if we could build 2x faster, our customers could build 2x faster, they probably would. And so that gives me confidence that we're still in the early days. Now, we all know that we're not going to grow at this kind of rate forever and we will see digestion periods. But I think when you look, you know, the analogy I tend to look back to is not the internet boom, but more the beginning of the cloud build where you would see these periods of growth and digestion and then growth and digestion. I think we're still in that first phase of growth. And at some point we'll see digestion, but right now we've got a lot of constraints preventing that digestion period.

Yeah, we did 4G, we, you know, we did 5G, but we didn't innovate in anything that was differentiated in new technology. If I look at what we're just starting to do in AI-RAN, Arjun, what I see there is very different, you know, innovation. We're talking about sensing, you know, hyper accurate location positioning, providing that data as a service. So, we're providing new tools. Now, ultimately, our customers who buy the infrastructure from us need to be able to monetize that and deploy those services. But there's no question in my mind that the data is going to continue to grow at the edge. The need for processing at the edge is going to continue to grow. And that means there's an opportunity for us to continue to innovate.

That's right. And today you can think of that connectivity, Arjun, as largely running, it's running like a data query, but we're already seeing elements of that connectivity being very different because in the internet era, largely what I do is I send a request and I get data down, but it's very download heavy. It's almost a 12 to 1 ratio in mobile networks. If you think about what we're seeing today with AI workloads, it's 4 to 1. So it's still download dependent but at a much lower rate. Why? Because some of those queries, I do this quite a bit in Finland. I take a picture of something, I send it up and I ask, you know, tell me what that is or, you know, I ask a question about where I can find something, right? And that's images. And sometimes it's audio for processing. And then you think about what's coming. We're seeing with physical AI, the early, you know, autonomous vehicles are going to process things locally. They're going to send images up. Robots, same thing. You can't have all of that processing be local. There would just be too much, you know, too much weight on the device, too much battery consumption. So, it has to happen at the near edge. And that's the opportunity we see for networks and certainly for telcos should they seize it is to deliver these differentiated services. And this is why we talk about the network being a part of the AI fabric in the future.

Yeah. So first of all, you know, if you think about, let's take one of those examples. Let's start with your example, Arjun, of cars. If you think about those cars, those cars today will have, you know, arguably a lot of compute power. They'll process some things locally, but today they operate as independent vehicles. And what happens in the future is you want those cars to start to talk to each other. There's a traffic incident here. There's an accident here. Oh, I'm slamming on the brakes because someone's crossing the street that I didn't expect. You know, you want the autonomous vehicle three cars back to know that because then it can start to slow immediately. Now, think about that action. That action is the car that's slamming on its brakes needs to make sure it's slamming on its brakes. Needs to communicate that that's happening. Now, when it communicates that's happening back up to the network, you don't want to send just an image or, you know, here's all the data that this car is doing. You want to transmit to that car, this is happening. This vehicle is taking this action in this location. Now, that little activity requires the action that's being taken by the vehicle, knowledge of the location. You don't want to send that to a car that's five miles down the road because it'll jam on it, it'll slow down in the wrong place. So it's, you know, you want to make sure it's the relevant location and then it's very time sensitive. So you think about just that example. It's one minor use case where the network needs to be aware. The network actually has to have the intelligence to say, I'm going to share this out to these vehicles, which means the network needs to see location, needs to see time, and it needs to have some awareness of the application and the service it's providing. That's what you, when we think about AI-native networks for mobile, that's what we see. We see these radio towers becoming much more intelligent and that's just the beginning. That applies to every element of physical AI.

Yeah, what we see with AI-RAN and why we need a computer at the, there's always been a computer at the base of the radio tower. It's been that way since really since 2G, but that computer has been very focused on one thing. That computer's been very focused on delivering, you know, our connection, voice calls, data calls. Now when we think about this world where the computer needs to deliver tokens, intelligence, the computer first of all needs to have the flexibility to make sure it can deliver the intelligence it needs. That might mean actually doing some processing. It might mean sharing additional information back up to an edge inferencing environment where that token can get processed so the right action can be taken. So it means these, you know, these networks need to become networks that sense, plan, and act. It's very different than I'm just transmitting a packet. I'm just transmitting a data packet, right? And the other element of this that becomes important is these networks also have to become inherently much more secure. Today the radio network is quite encrypted. But we need to make sure that when that, you know, when that computer starts to become a part of the intelligence network, we need to make sure that it's seeing information that's trusted. It can verify, you know, that that packet that came over from that vehicle is a trusted vehicle, right? I know it's in this place, I can validate it. So, we need to add a layer of capability so we can have that kind of intelligence. And that's what's different about AI-RAN. And that's why we see the radio network being so different in the future. Even if the fundamental thing, the radio isn't going to change that much, right?

Yeah, there's two things that I think they were traditionally pushed together that I would separate in these G's. So one is the G's become a discussion around new spectrum, new frequency, right? Which is ultimately for the industry a way of getting more capacity.

So sometimes when you hear people talking about 6G, they're talking about new spectrum. And with all these things we just talked about, there's going to be more data, which means more bandwidth is needed and more spectrum, right? We see that coming in 6G in terms of technology, we see this being much more around machine-to-machine communications, these sensing applications and services. That's why we partnered early on with Nvidia, not just because of their incredible computing capabilities, but their understanding of full-stack AI, right, and the services they can build. But it's not just Nvidia. We're partnering with other players in the space. But the key thing for me in this is this is really about software. And when I talk to a lot of our customers, that's the shift they would like to see as well. The past in the industry has been every one of these G's requires a huge forklift hardware upgrade. And the reality is most of the cost of that is changing radios even if the radio itself doesn't really need to change. And so the shift that we're making with AI-RAN is also to say, look, one, we don't want you to switch radios unless there's some reason to do it. You know, the radio's 20 years old. It needs to be modernized. You're adding this new frequency I talked about before, new spectrum. Okay, you need a new radio. That makes sense. But we want to have interoperability at the radio layer because that allows customers to make software upgrades. And there's some customers that we see that really understand this, that are visionary. Some of them are our partners in early AI-RAN trials and they understand this subtle difference. And what that allows the industry to do is not to worry so much about doing a 2G, 3G, 4G, 5G jump, but rather let's just build the services. And I think what we'll see differently with 6G is we'll probably see the little logo on our phones that goes from 5G to 6G, but more importantly, I think what we'll see is different services. And I think if you look at 5G, the reality is for most customers, maybe the bandwidth got a little bit better, but there really wasn't anything different. And I think in 6G, what you'll see is maybe we won't see it as consumers, but what you'll see is, wow, this network can now handle sensing. It can now handle physical AI. And that's where we see the opportunity in 6G. How that plays out, I think, is probably going to be a little bit different than what we've seen in the past. And I think it needs to be because the reality when you look at the industry is the industry hasn't been very healthy when it's gone through these big forklift upgrades. And we see that as a partner to our customers, we also see it from our customers' perspective because when you look at their return on capital, it hasn't been there.

Yeah, I think it's the opportunity of the industry to charge for these applications differently. You know, charge for differentiated levels of service. It's also that the business model evolves because a lot of what we're talking about is not a consumer device in a subscription. It's really an enterprise device. And we have some early indications of that with things like smart glasses and even, you know, autonomous vehicles where these connections are mission critical for the devices. But it's also the opportunity for the operators to charge differently. But the other key thing is don't force this big capital bubble because that means that they don't have to make these huge investments.

Look, I think the opportunity for there to be a different story here is there. You know, how it will play out is probably a little bit harder for me to predict. My focus is on making sure that we are delivering the cutting edge technologies and capabilities that allow this innovation to happen at the edge and then to let the market take shape. But I think the opportunity is there. The key thing for, you know, what I tell my customers and my partners all the time is, but you have to be a leader in innovation and if you're not differentiating on the performance of your network and delivering these services, you're probably not going to be successful. And I think, you know, when you actually go back and look at what's going on in the industry, even with the challenges that you described in the past, the companies that have built great networks that have differentiated on performance have delivered outsized returns. So the opportunity even to monetize the services today are there, but you have to invest in the network. And I think that's going to be the biggest, when we look ahead, that's the biggest differentiator that we see in the short term is invest in the network because it's not everything we just talked about is about new features and new services. It's not just about that. It's also about trust and security, right? You know, there's no way I want an untrusted network managing my AI devices, right? That's a huge risk. That's a risk as an enterprise. That's a risk as a public provider, a service provider. It's also a risk for a government. So trust is going to become increasingly important.

There's lots of different discussions. I think there's a view of end of the decade. I think we're going to see 6G services faster than that. You know, obviously if you look at 5G, there's places where we don't have full 5G deployments. So I think what you'll see is this will play out over time, but my expectation is we'll be, you know, 6G will be the beginning of this transition to services that we deliver versus just these big forklift events that we've been used to in the past.

Yeah, I think a couple of things. First of all, you've got this concept of scale across. Scale across is the connection of multiple data centers to scale training. That solves the problem of some of the things you talked about, just the limitations of I can only get so much power here. I have this much space in terms of what I can put in. What we're seeing is as the demand continues and especially as we get to more inferencing, right? A lot of the scale across problem was originally around solving training. As we move to inferencing, you don't necessarily need these large clusters. Now those large clusters are still an asset because if you're not using them for training, great resource for inferencing, but you can build these smaller inferencing centers. And this is also important when you think about things like sovereignty and security and locality and latency even. So these smaller data centers become powerful and that's where this data center interconnect with our optical and our IP technology, you know, we think there's going to be just a massive boom. And I think, you know, we also got to recognize, you know, even with some of the, you know, there's some pushback around, you know, there's been some pushback around data center buildouts. I think that's, you know, I think those things, these things will play out. But even with that, there's also a tremendous amount of data center infrastructure in the country that can be upgraded to support, you know, AI-native workloads, right? Legacy infrastructure that can be modernized. And so when you think about that in the US or you think about that here in the UK, for example, or across Europe, that all is an opportunity as well to put modern, you know, GPUs or AI-native infrastructure into, but that also needs connectivity infrastructure. And that's one of the reasons particularly here in Europe I'm so vocal about the need for digital infrastructure investment and connectivity investment because without the connectivity, those data centers become islands and that won't work for AI, right?

Yeah. We, I mean, with what we're doing in AI today, we see it in the thousand kilometers of distance. I mean, there's, it's pretty powerful the innovation that we're driving. And under the sea, under the sea is probably, that's probably a little bit too hard on the latency side. But at the moment there are data centers connected under the sea, but these are more traditional. This would be more of a traditional connection. But the importance of that, you know, sub-sea connection for passing large amounts of data is obviously going to continue because I may train a model somewhere, I've got to distribute that for inferencing into different localities. So these long haul and sub-sea networks continue to be important.

Yeah, absolutely. In fact, this is, you know, we, I'm glad you brought this up, Arjun, because this is something with optical that's really driving the transformation. And there's a couple things with physics. As you get these faster and faster speeds, it means you have to power the copper more and more. And what happens is copper actually can't, it can carry really fast speeds, but the distance it can carry it, it becomes shorter and shorter. And so what do I have to do? Well, I have to then reamplify it. And that requires more power. With optical, I can carry much faster, you know, much faster speeds over much longer distances and I don't need to repower it. So it's much more power efficient. And so as we get into these faster and faster speeds across data centers, but even inside the data center, we see demand emerging with optical.