Why AI infrastructure Depends on More than Data Centers I Kate Johnson transcript

But AI is a matter of national security. And we have to think through what's the right balance of getting a national framework that can help us go fast, but still respect, you know, sort of the the local rights and and values. And that's complicated. Legacy networks, they're just not big enough, fast enough, smart enough, or secure enough to support the needs of AI. The new corporate workforce [music] includes these autonomous workers called AI agents, and they live in data centers.

And they have to be connected to the rest [music] of us. >> Sometimes the most powerful forces in the world go unseen. A critical layer of intelligence driving possibilities. This is Micron memory and storage. Intelligence that uncovers new cures, models changing climates, and build smarter, safer cities. Every breakthrough in AI is driven by data, and data lives in Micron memory and storage. >> Welcome back to the Building America podcast, a show that explores how our nation is building better for the future.

I'm James Holman, the deputy opinion editor, and over a series of episodes, we are examining how artificial intelligence is just as much a physical story as a digital one. In this episode, we're looking at how one legacy communications giant pivoted from traditional telecom to AI infrastructure, building what its CEO refers to as the nervous system of AI. That company is Lumen Technologies, and joining us is its CEO, Kate Johnson.

Kate, welcome. >> Hi, it's great to be here. >> For those who aren't immersed in this world of AI infrastructure, what do you mean when you say it's the nervous system of AI? >> Well, I think networking was traditionally thought of as just pure plumbing. Dumb pipes, you know, connecting point A to point B so that you could transport data. But, in the world of AI, which is incredibly complex, you need to be able to transport data from anywhere to anywhere.

And so, you really need the ability to program the network to do that for you. And essentially, if you think about the nervous system is the conduit or the pipes connecting all the data centers. And the brain is the digital platform for orchestration. Basically, the control plane or ability to to do whatever you want the network to do in support of your AI ambition. >> So, what actually happens if the physical network that you're talking about doesn't have enough of this central nervous system? >> Well, I think I think legacy networks don't have this capability.

They don't have the intelligence layer that I'm talking about. They're simply static connections, you know, fixed point A to point B. And the reason why that doesn't suffice, James, is because AI is not operating in a single place. AI needs to operate in multiple places, right? It operates in buildings, it operates at edge points, it operates in data centers, and in public clouds. And it it thrives on ingesting data.

And it creates data when it makes insights, right? So, it's all about that data movement. The interesting thing is legacy networks, they're just not big enough, fast enough, smart enough, or secure enough to support the needs of AI. AI has to have the ability to call data in from anywhere, process it, and then send its insights to anywhere. So, it's asking for that adaptability and flexibility from a network that traditional networks can't deliver.

So, it's it's our hypothesis over the past couple of years as as we've been transforming the company that a programmable network is in support of the growth of of AI and it's proving out to be uh on target. >> I'm so glad that we're having this conversation because I think so many people when they think AI infrastructure, they think like data centers. But this is in many ways as or more important. You use the plumbing analogy, uh but you know, you think about you know, your companies literally out there building thousands of miles of underground fiber routes and fiber huts.

Uh in practice, what does that look like? Help people I you know, I visualize what it is that that Lumen is doing. >> So, from I guess it's 2025 to 2030, we're going to have a 10 times uh growth of data centers in the United States alone. That is a massive expansion. And a data center sitting alone without being connected is useless. It's just a brick. >> Right. >> Right? So, uh Lumen, we have the largest fiber network in the country.

Um we are growing it very rapidly. We're currently a little little bit north of 17 million fiber miles in our network. And by the end of 2031, we will have 58 million fiber miles. So, there's this massive expansion of fiber networking in support of connecting all of those data centers. Cuz those data centers are is where the AI lives, right? It's the new the new corporate workforce includes these autonomous workers called AI agents and they live in data centers.

And they have to be connected to the rest of us. You've got people, data, applications, and AI agents. All need to be connected. We've got the network to do that. But what's more, we've invested in this digital layer to make the whole thing programmable, which gives businesses optionality in terms of how they design their their AI architecture, and uh gives them flexibility to really be able to adjust to all the different factors in the environment. >> What do you say to folks who worry that there's sort of like a boom and bust cycle and that there's kind of a you hear the word bubble, but you know, you're building out just I mean a breathtaking amount of infrastructure, obviously very capital intensive, uh but you know, it's it is so important to the workforce of the future, all of that, but I mean, how do you how do you respond to people who say, "Wow, that's a lot of investment uh right now, and what if it >> It sure is. >> Yeah. >> [laughter] >> What if the 10 year X, what if all this stuff doesn't quite pan out as we hope? >> Well, so the numbers are daunting because they're they're kind of bigger and coming at us faster than we've we've ever experienced in our lifetime, you know, trillions of dollars worth of infrastructure investment.

And it's not just networking, obviously. It's the build out of the data center, which GPUs and energy and networking. So there's So there's a lot there. But I think there's there's kind of two angles. The first is, you know, we see traffic on the network, and the composition of data uh is growing dramatically from AI agents itself. More than half of the traffic on the internet today is generated by these autonomous workers.

Which is really sort of staggering if you think about the fact that ChatGPT was just released, you know, for the first time November of 2022. And in the three and a half years, it's you know, AI uh agents are comprising more than half of internet traffic. That's that's pretty astonishing. >> And a lot of that internet traffic is running on Lumen's networks. >> That's right. That's right. So, what we we see this huge growth in data traffic for sure.

So, we can validate that the demand's there for networking for sure. But, I think also if you think about the hyperscalers and the neo clouds and the companies that are building the models, they also understand what AI needs in terms of ingesting and proliferating and generating, you know, data as well. And you know, they're building uh ahead of the demand curve a little bit. They see the demand, but they're building that supply side for the AI economy because it takes so long.

So, what we're seeing is these companies making these huge investments today so that tomorrow and beyond, they're prepared to be able to support enterprises as they need to consume these services. >> To what extent, bringing it back to sort of the focus of this series, is it then a physical story, uh you know, the as much as a digital one? >> Yeah, so the interesting thing is infrastructure just always triggers everybody's mind physical physical physical. >> Right. >> Right?

And you know, in the ground, it's something that I I you know, I don't want to have to think about it. I only think about it if it doesn't work. But, the truth is infrastructure needs to be smart now. It needs to be intelligent. So, it starts the table stakes for entry to play in the space is the physical part. >> Yeah. >> And you know, since I since I became CEO in '22 to now, you know, and then '20, '31 and beyond, we will have more than quadrupled our physical network, which is really kind of profound.

It's the greatest expansion of the internet at large in you know, uh ever. >> Ever. >> Right? And um that's important cuz you have to be proximate to where those data centers are, right? You have to be right near it. We will be covering more than 90% of data centers by the end of 2026, we'll be right there for direct on-ramps into into those clouds um and uh the the GPU capacity that that is so precious. >> It's funny you started it you started at Lumen I guess right before ChatGPT came out, right?

I mean this is been such a >> started the same month. >> Oh well. >> [laughter] >> Three Three weeks before it came out was my first day, yeah. Yeah, so it's been I think they call it good timing. >> Yeah, exactly. [laughter] >> Um and the whole reason why I started was cuz of this physical asset, right? >> Right. >> Because it was so impressive. But the real opportunity that I saw was digitizing that physical asset and having the hunch that uh you know, the the the pipes need to be there, the fiber needs to be there, but if we can't make those pipes intelligent and able, you know, to respond to the needs of the environment above the ground, uh then that's you know, that's not going to serve AI well.

And AI needs all this data, so let's let's go after that. And you just a couple of examples, um as you build out data centers, you need to go where the compute is. Compute is very precious, right? Uh those GPUs are very expensive. You want the data to wash over that and uh you know, as quickly as possible. You don't want anything sitting idle cuz that's just too expensive. But if you're not able to build, you know, a data center in a certain area or if you have a workload that, you know, is taking up more than its fair share of capacity, wouldn't it be great to be able to move the workload to wherever the GPU capacity is today?

And that's what a programmable network allows you to do. The same holds true for energy. You know, it's there's a quest for uh clean energy, you know, affordable energy surplus uh or at least availability. And when you have workloads that are consuming more than their fair share, wouldn't it be great to move those workloads to where there was a surplus. And in the legacy network where everything's static and it's fixed, you can't do that.

But if you have a programmable network, which is basically digitizing that physical infrastructure we talked about, you have the ability to move the data wherever it needs to be, economically, security-wise, or um you know, from a GPU capacity perspective. >> I love your point that infrastructure's one of those things a lot of people only think about when it's not working or when it's broken. Uh we hear a lot about how hard it is to build in America today.

From where you sit, what is the one thing America needs to do to build better? >> So, I think we really need to think about the whole AI system. You know, we often get uh focused on one piece of it. You know, there's a lot of talk about the models today in particular, you know, um open weight versus not. There's a lot of talk about the availability of energy or, you know, scarcity around GPUs and the demand curve over the next several years.

Those are three very important pieces. Connectivity should be included in that list because again, those resources are rendered uh meaningless without having the connection to get the data where you need it to be so that you can get the insights where they need to be. And so, you know, if I think about one thing that this country could do is it could have a go fast button around building out infrastructure. And if you think I'll I'll boil it down to a very simple example.

We do long haul routes. I mean, if you need to connect something from New York to San Francisco, every time you need to go into a local community or municipality along that route, you're going to need a permit to be able to go into the ground. That's a lot of permits. And every single one of those municipalities has a different process, different timing, you know, a different method for generating the permits to be able to execute the work.

And that slows us down and frankly is a huge disadvantage to other countries like China where they don't have those kind of local laws governing, they take it from a national level and sort of express route everything. So, that's one example of things could be doing to go faster. >> You've been supportive of bipartisan permitting reform legislation that's trying to move through Congress. >> Uh and it is turning out to be a pretty heavy lift.

Why do you think it has been so hard to pass? Because I think I mean you're absolutely right that this is like the thing and I think it at a at a big picture level everyone understands how crucial this is, uh but it's in in practice it ends up being quite difficult. >> Well, look it's really complicated. I mean the things that make this country great, you know, is the federation of states, right? Where there's local empowerment and that's a great thing.

It's proven to be an accelerant around innovation, you know, around civil rights and and around the overall experience of this country. So, you know, we're in great support of those freedoms and and want to see them continue. But AI is a matter of national security. And we have to think through what's the right balance of getting a national framework that can help us go fast, but still respect, you know, sort of the the local rights and and values and that's that's complicated and I think really I love what we're seeing, which is this notion of bipartisan support for frameworks to govern how we do things, um public and private partnerships and academic institutions coming in and support as well to get kind of the troika of thinking around it cuz we're better together.

And um so we're headed in the right direction. I just I think clarity is really key, speed is really key. How can we make AI secure and reliable and how can we build out the stack, the whole stack, all of the infrastructure as fast as humanly possible because it is a race. Let's Let's be very very clear. This is an innovation race that matters deeply. >> You say we're moving in the right direction. I want to believe you're right, but we see like what happened in New York recently with the data center moratorium.

We do see some like emerging backlash to new technology. Uh how much does that scare you? >> I don't know that scare is the right word. It's It's to be expected. I mean, you know, state backlash to protect, you know, citizens' rights and and all those things. Um but I I think the administration is is addressing these things and saying, "Okay, let's talk about energy. Let's have, you know, uh a framework where companies can do the right thing and say we're going to cover the cost and that won't be incumbent upon, you know, the citizens." That's a breakthrough moment where maybe now that we took that off the table, we can get down to, you know, how can we have shock clocks or minimum SLAs for permitting, for example? >> Can you for those who don't know what What is an SLA? >> Oh, service service line agreement.

Just a Just a an agreement of how fast you might do something and say like, "Hey, you've got, you know, 10 days, 20 days, whatever it is to get through this permitting process and at the end of it, you know, you've got to render an opinion." And the reason why I brought that up is because you know, when you're building a route, that route that I talked about from San Francisco to New York, it's a pretty long one, if you will.

Obviously goes across a bunch of different segments, but we have something, you know, the physical infrastructure requires um in-line amplifiers or ILA huts. And it's basically a piece of equipment that every 50 miles or so, it needs to boost the signal so that the the data gets to the place that it's supposed to be on time and at at the right level. And it takes a couple weeks to install one of those huts every 50 miles.

It takes a couple of years to get the permit to do one. Right? And it often times it's the same permitting process as commercial or residential real estate. And we're standing in the same line. A great example of how, you know, the government could help us is to say, "Hey, there's no plumbing, there's no parking. This is just a server rack. Let's separate it out from the complexity of commercial or or residential real estate permitting." And and allow for a fast lane uh for permitting in support of this infrastructure build out we're trying to do. >> That seems like a great idea.

Uh on energy, you mentioned it and uh we have seen a lot of moves recently to assuage public concerns about electricity, but how much energy infrastructure are we going to need to enable this build out that that you're talking about uh to support all of this just tremendous amount of data? >> Yeah. I mean, I I think there's a clear need for more and I think there are a lot of companies that are working together to figure out how to do that safely, cleanly, and in an economically feasible fashion.

I would argue that a programmable network would allow us to get more from the energy that's available today. Because today you have, you know, uh communities where a data center might be taking more than its fair share of energy and if you're able to push, like I said, workloads and spread them to communities where there's surplus, for example, then it it might be uh a more economical way to address you know any sort of energy gaps that we have. >> That's a really important point that Lumen can limit the strain on the national grid. >> That's right. >> You talk about this being a race and I couldn't agree more.

This is just like the big race of our time. Thinking about long-term American competitiveness, when you look at the global landscape, what broader strategic steps do you think the United States needs to take to ensure that we are the global leader in physical AI infrastructure? >> Well, I I I think there's a really hot debate right now around open weight versus proprietary models. You know, I I come from an open source background.

I spent some time at Red Hat. I think we know open source is an innovation accelerant. That the sort of counter proposal is access to models, you know, shared at a global level would suggest that, you know, you have some competition. I think that's a great thing. I think competition makes us innovate faster. I think it it forces us to innovate more economically. And when we put pressure at that model level like that, we start to realize that it's not just about the model.

It's about the model, it's about the chips, it's about the energy, you know, available and it's about the the connection of all of it and you have to consider it all as one system together. And you know, I I guess I'm trying to advocate that networking be promoted as first class citizen in this AI discussion because it's is of paramount importance. It can alleviate energy, you know, demand problems. It can alleviate GPU capacity problems and should really be considered all as one discussion. >> What do you think the best way is to do that?

Cuz you I mean, on the public mood piece, you know, this this kind of earlier in the summer we we saw the booze for some commencement speakers as they talked about AI. There's a lot of anxiety among young people. But what you're talking about, I mean, you're essentially you're enabling partners with these AI bots for these workers of the future. In a lot of ways it will make their jobs much easier, allow them to accomplish a lot more.

You're talking about making the kind of the network just a heck of a lot more efficient. But how do you get through to people who have this sort of generalized anxiety? >> Well, I I look, I think the anxiety is around the unknowable. When there's lack of clarity, there's anxiety. There it's a one-for-one match, right? And so I I do think that it's hard because uh not you know, there's no clairvoyant uh person who can say this is exactly what's going to happen.

And there's going to be adaptation, for sure. And we know that adaptation has friction, it has pain, it has loss. Um it it can be very very uncomfortable, and I think that's what you're seeing in these public discussions, these forums, is uh you know, uh citizens, uh students, uh professionals are saying, "Hey, let's let's slow down here, you know, for a second." And and that's very complex, but I'm an optimist, I'm not clairvoyant, but I have looked at, you know, historical uh trends of technology, and I think net-net after that adaptation period is over, there's a lot of benefit to be had.

So we just got to be uh really cognizant of thinking about how we give our employees the skills that they need to adapt, how we train for this in our school systems, and prepare, you know, our employees of the future for the same. And um and how we help everybody through that adaptation process. I think I think that's part of the story of of, you know, of how we get everybody ready for this because it's happening. And you can either get in the game or, you know, you can hang back.

I mean, Jensen Huang says all the time, you're not likely to get replaced by AI, but you'll probably be replaced by somebody who uses AI better than you. And what does that say? It says you need to know how to use it. You need to be comfortable with it. You need to get in the game and you need to start practicing and learning and figuring out how can help you do your job better? How can help you learn faster? You know, how can make us more efficient and more productive? >> You've been transforming Lumen over the last 4 years from a legacy telco to this modern technology infrastructure company to be a trusted network for AI.

What's the most important lesson that you've learned about rebuilding the foundation of a company? I know you've dramatically reduced the amount of debt. You've done a lot of things to put the company on on solid ground. What uh what takeaways can others get from your experience? >> Yeah. So, uh we have done a lot There's like three big things. For sure, we transformed the capital structure. That was incredibly important.

Uh doing $13 billion of deals with the hyperscalers to connect their data centers was an essential part of that story board. Um we're in a great place now, strong balance sheet. We also repositioned the assets uh to be the trusted network for AI and we declared a major, James. We said, "We're going to serve enterprises, right? And we're going to be very focused on making sure that all of our resources and our capital is allocated to do that better than anybody else in the world." And that's why we sold off our fiber to the home business.

Um but probably the most significant lever that we pulled was on culture and talent. And we said, you know, I don't I don't foresee a world where it's just AI. I foresee a world where people, you know, actually are supported by AI. And And they're going to be in charge, they're going to be the ones that lead the way. They're going to either stand in your way or they're going to help make that transformation happen. And so we invested in them from day one.

We said, "Okay, you know, we got to be learn-it-alls instead of know-it-alls. You know, we have to be open and have an open mindset to to just about everything. We have to be learning organizations. So, if something goes wrong, rather than, you know, having that uh postmortem witch hunt thing that a lot of companies do, we go right to, "What did you learn and how would you do it differently?" Okay, let's go. And I think that that has given us a serious advantage because right now companies need to adapt in in a inside of a day, inside of a week or a month.

Instead of this setting this annual budget and having everything fixed, "Okay, here's my strategy, here's my budget, here's my head count, my sales targets, my capital allocation and my my commitment to a, you know, my investors, and then let me wait a year and see how that all goes." That's not a thing anymore. You have to be able to pivot based on the signals you get back from the market. And those signals are coming in hot every single day, and you have to be able to separate the noise, you know, from the actual learning.

So, you need smaller teams, they need to be empowered, they need to be scientific in terms of how they test and process. And that's what we're teaching our people to do, and it has really paid back. So, recapitalizing, yeah, that was really important. Repositioning the company to be the trusted network for AI, yes, that has redefined who we are. But it's those people that have done the work to transform this company.

And in case you can't tell, I'm a little proud of them. >> Just circling back to what we were talking about a few minutes ago, I was thinking about kind of the the security in the physical infrastructure. Uh how I mean, when you're thinking about building something so big, and you're talking about these huts every 50 miles. How are you thinking about like physical security of all of this infrastructure? >> I mean, so obviously uh security is of paramount importance.

We take it very seriously. Um we actually support many of the government agencies and they choose us because of our security and it's physical and it's cyber. We have incredible muscle in this space that you know, Black Lotus Labs uh has threat detection capability. That's our That's our security lab um uh team. You know, and and we have threat detection capability that is world-class and we work in partnership with the government to identify threats and to thwart them on on the regular.

I think an interesting thing that we're seeing is uh you know, there was a time where the world kind of said, "Hey, public internet on in networking uh is a less expensive way to transport our data and we've got a lot of data, so let's start using the public internet to do that in between our facilities." I think with with the cybersecurity landscape changing as dramatically as it is, there's this you know, sort of renewal around the idea that private networking is a great way to limit the blast radius of of malicious attacks and that's part of our story board.

We We've got, you know, by the end of 30 the 2031, we'll have 58 million fiber miles and a lot of capacity for private networking to interconnect data centers for enterprises to safely and securely transport their data. >> One of the things to that we can uh wrap our conversation with is um we've been asking other CEOs uh and I'm curious for your answer. Who is an American builder or innovator you admire from either the past or the present and why do they inspire you? >> I'm going to have to go with Warren Buffett.

A great builder. He's a great business builder um and he's inspiring because uh of so many reasons. First of all, he started uh with a textile mill that was failing and he transformed it into one of the greatest, most valuable companies of all time. But it's the how that I find, you know, incredibly inspiring, right? He was enormously focused on culture. Uh he looked for passion in the founders of all the companies that that he was buying.

Um always high integrity. Always took the long view and was rooted in value and the companies that that comprise Berkshire Hathaway today, they're just they're just great foundational companies. There's manufacturing, there's retail, you know, there's home home building, et cetera. And um just really kind of a salt of the earth uh orientation. He's a great capital allocator and a great financial steward. But he did that rooted in people.

So there's a lot of heart there. So I I have a a great, you know, affection for the idea that you can drive incredible business impact by focusing on people. >> The Oracle of Omaha is a is a great great answer. Uh well, Kate, I really enjoyed chatting and hearing about everything that you're doing. Thank you so much for joining us. >> Well, thank you, James.