Why AI Demands a New Internet Architecture: Kate Johnson, Mobile World Congress 2026

I've been here before, but this is the first time with Lumen.

I think it just continues to intensify with the advent of AI. So CIOs are on the hook to innovate right at the cost point that you're talking about. They have to drive insight at the speed of thought and because we're all terribly impatient. We all want everything immediately. And as AI continues to up the game in terms of how quickly we can get and synthesize information, the onus falls on the CIO to do that in a cost-effective, high-performant way for companies who are under enormous pressure to transform. So all eyes are on that problem and it's a complex one because the architectures are changing, data workloads are exploding and users like you and me are getting pretty impatient.

So I think that's an interesting debate that we can have. I think more and more are carving out parts of their budget for AI. They're very focused on productivity. And I think they really need to be focused on their customers and how they can delight those customers because that's where the growth is going to be. We're hearing in the paper every day, oh AI productivity here, thousands of jobs lost over there, etc. Where is the story about transforming the experience for the customer with new workflows, with new business outcomes so that you can really build a growth story? That's the pivot that we need to make.

You know, I think hybrid architecture is here to stay. So there was a day where we said okay everything in the cloud. Well, do you remember before that where we said cloud, I'm not putting my data in the cloud and then it was okay this thing works and you know the hyperscalers can protect our data better than we can because they spend more on security than we do and they can drive innovation for all the development platforms etc faster than we can. So onward into the cloud and then we realized that we didn't want to put everything in one place. So we did multiple clouds and now we don't necessarily want to turn it all over to the clouds because their business models don't necessarily support what AI demands. And AI demands that you have enough compute. Compute depends on space, cooling, power, and connectivity. So there's really this data center explosion. Everybody's building them and the CIOs are spreading workloads to the most cost-efficient place on the curve and that requires something fundamentally different from networking than we've had in the past.

That's right. It's all about where we're going to put it and it's edge, it's data center and it's multiple clouds.

I think we are. So 60% of the traffic on the internet today at large is coming from AI agents. And you and I both know that most companies haven't really hit the curve in terms of steady state for their AI agent, their business workflows, right? They're just getting started. So with 60% now, you can just imagine the proliferation that's going to happen. And what enterprises, there's two pieces to this. There's the hyperscaler side where they're building the supply side of the AI economy and massive data center buildout, massive connectivity spend, massive infrastructure spend to provide services to enterprises. Enterprises have the CIO budget, CTO budget, and they're like, okay, how much are we going to spend on AI? And they're recognizing their networks are not big enough, fast enough, secure enough, or smart enough. They're not intelligent enough. And they're recognizing that they need something fundamentally different. And it's just a couple of things. We all know the bandwidth and latency thing. It's got to be big. It's got to be fast, right? But there are these other sort of things to consider. Are all the data centers in a region or across the globe where I am, are they connected with fiber? Do I have on-ramps into the cloud that are fast and efficient and don't charge me, right, just to connect? And is my network programmable? The old world is enterprise prem to the cloud. One connection static, doesn't move, stays there forever. The new world is I got to connect all my prems together, all my edges together, all my cloud, wherever I have stuff in it and every combination therein. So, it's got to be programmable so it can be dynamic, so it can change as your business requirements change.

So, I think it's like I said, it's all about fastest time to first token. It's about speed to get insight when you're, you know, clicking on your keyboard or you're on your phone or you're talking into a device. And I don't know about you, but remember the old days where there was lots of beeping and humming when a modem was firing up to make a connection. And then when the internet came and when search came we all were like oh thank god because that old stuff was too slow. We thought the old stuff was great by the way when it came out. And then we need more. And I think right now we're at this place where instant gratification is a thing. And if we don't get it from AI, we're just stymied. Like, how is it so hard that you're giving me my board memo, looking across all of my ERP and CRM systems and bringing all the insights to the board and it's taking you more than 10 seconds. That's outrageous. And it's the expectation of the customer experience. Customers expect instantaneous gratification from their technology experiences and AI is making it harder and more possible.

So I think the physical network you'd start with Lumen's network or that kind of coverage. So we cover a large swath of North America. We're within 5 milliseconds from 95% of the businesses in North America. You need that around the globe at every single geography. But I think if we were to build it today, we would cloudify it. You know, compute and storage that's been cloudified since 2006 when AWS launched their business, right? And what does cloudify mean? It means you can fire up any compute node and storage node. It's instantaneous. You can have it as big or as small, whatever flavor you want. And it's all with a couple of clicks. Network just hasn't been like that. It's been analog all that time. And we've got to promote networking to be at the same level as cloud, as compute and storage in the world of cloud. And if we were to rebuild this today, that's what we would do. The good news is we don't have to. We have the physical network. We can overpull with modern fiber solutions, which we're doing. And we can build that digital platform with a single control plane on top that makes the network act exactly like compute and storage do in the cloud world.

Yeah. So, the assignment that the board gave me which was different than the assignment of the prior management team. The board said pivot this company to growth. And the very first hard decision that started two days before I got there because it was a part of our agreement was to cut the dividend. So instead of harvesting cash on declining revenues to pay a dividend, they wanted to harvest that cash to build that digital platform to transform customer experiences and bring the company back to life. We were laden with debt, $21 billion of debt. All declining revenue, no innovation, no growth platform. So we were in a bit of a pickle. And I joined the company in late 22, hired a new management team, injected new talent at every level of the company and frankly took the great talent that was there and empowered them to tell us what was happening at the company. Repositioned the company to be the trusted network for AI, saying, hey, we've got these assets that are deeply relevant to hyperscalers. And we transformed our capital structure. We took the company from 21 billion of debt down to 12 billion of debt by saying we're going to focus on the enterprise business and sell off our fiber to the home and then we did 13 billion worth of deals with the hyperscalers to connect their data centers and we stand here today having turned around the business and now we're driving transformation of the company and the industry.

Well, I had a friend named Brené Brown. She's a sociologist who does a lot of leadership research and development. And I called her up and I said, I got a live one for you. We'd worked together at Microsoft and she has this methodology called Dare to Lead. We rolled out Dare to Lead training to 25,000 employees. Everybody, it was mandatory. Everybody had to take it, including our techs. We worked with the unions to try and figure out how to get them to participate in the program. And it taught them skills that can be learned of how to build courage, how to be more vulnerable, how to show up and actually be more curious and unpack really hard problems instead of operating from fear and trying to protect old revenue. How do we show up with courage and really sort of overcome those insecurities to make bets on our future knowing that we could fail. Nobody wants to fail. That's gross, right? You have to and taking risks is completely antithetical to a fearful culture. If you're a fear-based culture or if you're playing not to lose and protecting old revenues, legacy revenues, and harvesting cash to pay a dividend, that's a play not to lose strategy. A play to win strategy, which was what we wanted the new culture to be, is all about making bets. It's about being just as happy to say yes to something as no. We were bad at saying no, and saying no is a great business decision often. And so building the muscle to do both of those things. And then probably the most important muscle we built is the learning muscle, being curious, excavating what's the truth about what hasn't been said, seeing where the real problems are, and then trying new things. And when you try new things and you fail, face plant in front of the team, we help each other up. And we say, all right, that didn't work. What are we going to do differently? What do we learn? Okay, let's go. And we actually kind of celebrate people who have the courage to do that, which is very different than most boardrooms.