Good morning, everybody. My name is Brandon Nispel. Welcome to the KeyBank Technology Leadership Forum. I cover comp services for KeyBank and towers, data centers, and such. Super happy to have Ed Knapp, CTO of American Tower Corporation, with us. This is your eighth year at the conference.
Eighth presentation, I think, too.
Maybe just help us understand, you know, what's your role, what's your function as CTO at American Tower?
Thanks, Brandon. It's great to be here. Brand new venue, so really, I think everyone's appreciating that. At American Tower, I joined eight years ago to try to help them get more technical, in the sense to bridge the gap with just towers being a passive business to looking at additional extensions of those towers. Some of the areas I've worked on and we've been successful in is data centers, for example. We purchased CoreSite in 2021, and that's a big part of our capital plan now. Another part of it is satellite. We invested in AST SpaceMobile. I sit on the board there, and we're seeing the convergence of satellite and terrestrial networks. Other areas, power as a service is something that we do in Africa out of necessity.
If you think about how we need to transform the data centers, already do space power cooling, but the tower business really never got into the power part of it. As we look at pushing more and more virtualized platforms out to the tower, it's going to look like small data centers. We've been doing a lot of work on the edge. That's a big part of my remit. We've dabbled in fiber, done a bunch of other things with our assets, but for the most part, that's what I focus on. 6G, of course, is a future technology that we can get into.
Let's talk about the intersection of sort of towers and data centers, maybe to start. You know, American Tower acquired CoreSite, what, three, four years ago? At the time, there was a view and like strong conviction that the edge would play out in terms of towers and data centers being sort of converged infrastructure. I think companies admitted that that thesis hasn't played out as quick as you might have liked. What keeps you convicted that that's sort of a thesis that you can continue to pursue?
I think it has moved to the right compared to what maybe people thought. I do believe, just from being here and you hear a lot of the work people are doing, for example, with agentic AI, the movement from a lot of training into production to models that would be more interesting. You could talk about them being on-prem, but the power consumption says they need to be in a more, let's call it, comprehensive facility. We have been working on a Raleigh data center for the last, let's say, a year and a half, two years, to try to prove that concept. Part of it is we have CoreSite, so we know the data center business really well now when we look at it as, let's say, campus-like co-location facilities with large interconnection. The thesis was that that will distribute, right?
At some point, you need to look at bringing more and more peering points closer to the edge, and you need to look at how is the wireless and wireline networks going to converge. We see that, in fact, previous speakers talked about that today, that the wireless network has been a thin pipe and is getting bigger with 5G and ultimately with 6G. That is going to drive a lot more data, and that data gravity is going to say things need to happen at the edge. It is going to be a longer journey than I think most people would have liked. The reality is I still believe with conviction that we're going to see more distributed compute. We're starting to see customers excited about what we're doing in Raleigh, and that trend will continue.
We're in a great position because we have the hyperscalers and the co-location facilities and essentially the wireline side of the business. We have all the MNOs and the towers on the wireless side of the business, and we're at the center of that convergence.
Talk about sort of in the last three years, what type of progress you've made in terms of establishing these edge facilities. You talked about Raleigh. How many relocations do you have? Where do you see that going over the next three to four years?
Early on, we had a couple, even before CoreSite, we started dabbling in edge. There were some misconceptions by certain players in the industry that you would just magically put a shelter at the base of a tower and the traffic would break out. Even today, we're still not at 5G with SA, where you can break out the traffic everywhere you'd like in a more efficient way. Most of that traffic always got backhauled to the mobile core. Early on, we dabbled in trying to place small facilities, like on the order of 100 KW, 10 racks, believing that communities would look at those as local data centers. We did one in Jacksonville, one in Austin, two near Denver, one in Pittsburgh, and one near Atlanta. We also bought a facility before CoreSite called Colo Atl, which was a meet-me type room.
We learned a lot from the fact that nobody wanted to be in those sites at the level we thought. They were small, and the persons and players we were working with thought that they would fill up and become localized, outsourced from on-prem for small, medium businesses. There was some demand, but not enough to really justify it. We made really small investments, and we learned a lot about what not to do, given the overhead of how many racks you can sell, what the price points would be. We sort of doubled down after we bought CoreSite, learning what we know about how to operate there, how they operate their business, how great they run from a customer perspective, the overall colo market, and where the trends were. Originally, everybody was focused on latency. That never played out.
It's going to, but it's still one of those things that the long term. Then it was data gravity, and we never got a lot of that heavy data, which we're starting to see transform on the uplink and more balance between downlink and uplink traffic, especially as we start to move to more sensors and video. The thing that's driving it now is power, right? Power is getting constrained in a lot of these centralized locations. That's saying, where can you find power to distribute out to the edge? When we built Raleigh, we said, look, we're not going to build 10. When we do CoreSite, we build sort of on the order of anywhere from 10 MW- 30 MW type of facility. We have a campus where we can interconnect them with fiber.
When we look at Raleigh, what we did is we looked at something on the order of 5 MW or less. In this case, we deploy them as modular increments of 1 MW . The first megawatt is out there, it's being sold. There are a number of different types of customers that are interested in that. Bare metal services, GPU-as-a-service companies, the optical companies that are local access networks want to be in there. That's bringing even some local enterprises who think of saying, can you get me higher power density, which is our second drop of modular? We can add liquid cooling, and we can do that faster than a lot of the retrofits that are going today in the larger scale facilities. We're seeing this need through power as almost being the entry point. Once you can get to scale, you start to see that next phase of not just the data gravity part, but the latency part kick in.
I wanted to ask, I mean, I think we hear headlines of all the big numbers of megawatts and gigawatts even being built. These are smaller facilities. There's also, you know, telcos have a lot of central offices with probably in the range of 5 MW type of power. What do you see ultimately as like the driving force to get customers? What's like the inflection point around like really establishing that business?
Today, I think in the last, let's say, two years or so, the 100 GW data center and largely interconnected GPUs, that trend is going to continue as far as CapEx will drive it, right? People need to have that. That's not our business, right? That's not what we do. We do need to interconnect to those facilities, and we do that through our cloud interconnection. In CoreSite, we have the ability to sort of handle those 100 MW to gigawatt plus facilities through our relationship with the hyperscalers. More importantly, the question is, what happens next? When you start looking at transformation, we heard a lot even in this morning's panel about where AI is, agentic AI, production of real tools that take advantage of LLMs and even small language models to really transform businesses.
We're still in early stages, and frankly, a lot of large companies are worried because it's moving so fast. Do I put money in and then it's obsolete? At the edge, you kind of have an opportunity to experiment without building out a major facility on-prem. If you don't have them today, you're not going to have, let's call it, tens of kilowatts of rack power density available to you. Maybe you had a couple of kilowatts when it was old CPUs and you're running Dell and HP servers. Now you got to run something that's a much more substantial platform. You don't want to do it also on your own. You can share it, right? Some of these folks believe that that investment, while it's moving so fast, it's not worth it. Let's just rent it.
We can do that locally where we have tech companies in the Raleigh area, two of them, that are saying we need to run it for, for example, managing networks. This is networks that are in the service provider place that they actually have equipment that they're looking at alarms and downstream management so they can make predictions of what failures occurred and how to correct them. They're building that on a small scale basis locally. There are other people who are looking at it to try to transform how do they manage multi, let's call it OEM platforms for adding GPUs to those hardware components. There's people starting to experiment saying, you can get me to 75 KW a rack, 80 KW . I can do my liquid cooled H200s.
That's something they can't get quickly, especially when most data center space before we even build it is being leased out. The edge is happening because there's this power demand. When you transform, like people talk about transforming legacy, it's a lot harder to do that because you're saying, I have a system that was built for 1 KW or 2 KW racks when it comes to space, power, cooling. I now need to transform that into a computer room that's modern in terms of what the requirements are. There's a lot of CapEx investment, and the question is, what's the chicken and egg problem? We try to lean into it only when we're not going to speculate on a location. We want to have anchor customers and tenants. In the case of Raleigh, because it's a new product, we did lean into it speculatively.
We believe because we've done a lot of work on, let's say, 30 more locations where we have land, we have a tower, we have fiber providers, we have the capability, we do the zoning and permitting, and we do the power discussions with the local utility to get to several megawatts is a much faster path. We can do that in 12- 8 months versus building a three-year facility that's on the order of these 10 MW- 30 MW
Interesting.
There is a speed element, and there is this ability to transform your business with risk knowing that the underlying GPU hardware is changing so quickly. You could rent it. You do not have to put it on-prem and buy it. I have even talked to universities about that, saying departments are getting grants saying, oh, we need AI in their research. The IT department is like, where are we going to put it? They go, build a new data center. All these departments have their own variations of what they need. Why do you not just put it in the community and have that be shared, which is really the basis of cloud?
Got it. I want to take a little bit of a step back. I mean, you've been a CTO for a long time. You're a technologist. Where would you sort of characterize where we are from an AI perspective?
I mean, look, at the end of the day, people will say ML has been out there for a while. You move from algorithms to other types of deep neural network models. A lot of that work I saw in my tenure at Qualcomm, we were doing a lot of research looking at those models when they first came out, when they were looking at ResNet and the ability to go to deep neural networks, lots of layers, lots of parameters, that early recognition of, you know, let's call it images or other types of attributes that you can test against and train against. The probabilities were still low. As that started to get higher with deep neural network models, the game changed. It was really until we got to these large language models, which democratized it, made it available. Everybody's talking about AI.
We are at the point where we're still in the early, I think, innings of transformations. You're seeing that in every company is that we, even ourselves, looking at it, how do we use these tools to improve productivity? Part of our initiative now is efficiency, right? We're trying to globalize our business, get all our data sets, and let's call it labeled data into data warehouses so we can start to make more efficient, productive use of that data set. We had a lot of acquisitions over the years, and we had a lot of towers that were built and operated in Africa and Latin America and Asia, you know, Europe. They all had different data sets. We're trying to harmonize that now as part of an initiative that's sort of going to be multi-year. We believe that AI internally will be a game changer for that.
I still think when you think about it in the context of quantitative and reasoning, to me, that's the world I live in. It's been early days because we had hallucinations, people entrusted. You get a lot of creativity out of the AI. How do you really get it to get to an answer that allows me to move forward as an individual more productively? I start to see that happening now. More and more companies don't need the old entourage of like all the people when they start out. They can say it's a few folks, and then there's a lot of AI around them to help them navigate the landscape. I think we're at a point where it's being adopted, and it's going to get pushed out. We need to move from training to actual use cases.
I see a world where there's a lot of value in orchestration. Being able to orchestrate agents across use cases at the consumer and enterprise level will be super productive. You'll be able to have your own sort of world that serves you and is much more productive than administrators or others who are trying to take care of your daily needs. All that goes away and becomes kind of a twin of your needs and learns who you are, what you do, and almost predicts what you want. It brings you that cup of coffee with the robot in the morning. I mean, those things will happen sometime in the future. I'm excited. I think this is a really great time in technology. AI is a big driver of that.
Got it. Let's switch gears, talk about the tower business a little bit. Before we get to towers and sort of the spectrum opportunity over the next five years, AST SpaceMobile, I mean, direct to cell service from satellites has been a topic. How do you think about it in terms of maybe an opportunity or maybe a competitive threat?
Yeah, I think I've looked at that for many years. First, we met them like six years, plus years ago. We made an early investment in them. I thought that that was really important because the way the AST team was looking at directed devices was very different than what I had seen before in my days at Qualcomm. It all required changes in components. I said, if you can keep the same cell phone and you can use a large aperture array in the sky and you can get that up there and unfold it, that's a game changer, right? We made a couple of investments in the business. I always believe it's fully complementary. You're not going to take thin layer coverage and a few number of satellites and displace the tower business.
Towers are there for, they're like railroads and tunnels and infrastructure that I don't see going away. Wireless spectrum will continue to be used in many different ways, even shared in the future with AI using it to make more predictive intelligence around where users consume spectrum. We get more efficient in bits per second per hertz. I think the satellite really transforms this connect the unconnected first, the white space, but more along the edges of the boundary where it was uneconomic to really build towers and then ultimately provide a thin layer to create more of a seamless user experience so that you always have a backstop of maybe texting or maybe some low bitrate voice to be everywhere. That's what the SCS order from the FCC did, it said you can't just do this on an old school licensed by licensed territorial basis.
You have to do it nationwide. AST got through with some of the operators to be able to get access to low band spectrum nationwide. It's not a lot of spectrum, but it gives you that thin layer. More recently, through a lot of their efforts, they've started to go after and work with partners on MSS spectrum. There was the L-band Legato announcement, and then there was this S-band that just came out earlier last week. Those things are adding a little more depth and spectrum, but you're still talking tens of megahertz versus what you're going to see is hundreds of megahertz that are going to be available terrestrial with fiber. The two of them will have to seamlessly work together. I think it's very complementary, and I think it's super exciting.
We'll get to the point where it's like connect, drop, reconnect to the point where it's like stay connected but thin to then a more integrated view when we get to 6G where NTN gets built in. Now we even have layers of resiliency in the network. I think that's really going to help us for disaster recovery and for the future.
Now, not just to put you on the spot, DISH earnings call a couple of weeks ago talked about building a satellite network to serve mobile use cases, right? Where they're really talking about trying to take some of those high-cost sites for the carriers in rural areas and really complement those or even potentially have them offload a lot of the traffic that would otherwise be on a tower site in a rural market onto the satellite service. What do you think about what they were talking about from a sort of like satellite-to-cell connectivity?
Yeah, I mean, I think as an idea and conceptually, you can say yes, I can do that. In practice, think about in-building. Think about the way a terrestrial tower is helping in those communities, you know, be more, let's call it pervasive in terms of signal strength. You're not going to say, okay, let's take out those towers. First of all, there's an infrastructure investment there. To put beams from the sky directing them directly to a particular area is going to consume a lot of the capacity. The satellite's got thin capacity trying to spread over a large area. It wants to focus it where it's going to be most valuable. If you already have a tower, are you going to say, oh, I think I could save a few thousand dollars on OpEx and I'm going to use that from the sky?
Who are you not going to serve? Because you're making that choice. You already made that investment, I think it's there to stay. You could argue that around the edges. There's some benefit. I think some sites might be just completely like not covering a lot of pops and just covering land. I think that it's such a small, small, you know, error in the scheme of things.
Okay, fair enough. Switch gears. I mean, a new tax bill was passed. It gave the FCC spectrum auction authority. They have to auction a certain amount of spectrum over the next several years. Maybe what do you see as the opportunity for American Tower? What spectrum bands? What's sort of up and coming in your view from a spectrum standpoint?
Yeah, super excited that we finally got some auction authority back. I think the OBBB, I guess, is what we call it. 800 MHz, 500 for the government NTIA, 300 FCC needs to get done in two, four, eight years. We've looked at all those bands. I think any, so from our business, very simply, the tower business is there's three phases. First phase is spectrum. Spectrum drives new radios, drives new deployment for coverage. That's all amendments, generally speaking, but could be new colos depending on the frequency band. Second bucket is technology. Every G is really driving higher bits per second per hertz. How do we get more capacity out of what assets we already have? The third leg of that is colocation and densification. You can't squeeze the technology any further. You need to go and build more sites. That's our business.
When spectrum starts out, it really opens up that framework. We'll see that towards 6G with even more spectrum around the world harmonizing. The bands of interest are, first and foremost, upper C-band, 3.9- 4.2. We had our altitude or altimeter arguments maybe a year or two ago. I think the filters and those things can get changed to the point where that 100 MHz initially in those frequencies will get picked off fairly quickly. There's some re-auctioning of AWS that was never put out. The debates start really around two areas. 6 GHz, which is generally the unlicensed Wi-Fi. There's a lot of spectrum there between upper 5 and like lower 7. The government's also taken off the table 3.1- 3.4 and 7.4- 8.2 or 8.1. Some of those bands are really the ones where worldwide harmonization of 6G is going to happen.
We have some soul searching to do there. I always believe 6- 7, there could be room there to work with the Wi-Fi community through more advanced techniques of coexistence and spectrum management, shared spectrum management. That's going to be probably a debate. The other one is CBRS. There's some proposals out there that say, oh, let's reclaim CBRS and move them to 3.1- 3.4. I don't see that really happening. There's so much invested in private 5G. I've done a lot of, we've done a lot of work too in private 5G during my tenure. I think that that market is really starting to take off. There's a lot of companies that are benefiting from that. I see though that there's also 12 gig as you get to the upper mid band. Those bands will take more time in terms of components and relocations.
The top of the list is the 3.9 GHz–4 GHz for the FCC, followed by what they could do with the UNII bands at, say, upper 5- 7.8. What are they doing there? On the NTIA, they have 4.9, they have 4.4, 5. They have some of the other spectrum that they're looking at, beyond that in terms of 2.9, 2.7. There are some things in there that they're looking at as well. NTIA is really the key here going forward because the FCC doesn't have as much wiggle room with spectrum. We just need to do a better job as an industry of getting more efficiency out of what we have.
The upper C-band should be sort of first to auction.
That'll be great for, you know, again, the operators to get more capacity.
What do you see as the timeline for potential other?
They have to do that in two years, right? This is the first.
For the upper C-band.
Yeah, that's the first step is two years, four years. They have to deploy it by, I guess, the auction will be in two years, is what they're saying.
In between that two and four-year period after the upper C-band auction, 6 GHz, 4 GHz
I think there's a fight on 6 GHz. There's a fight on CBRS that has to happen. There's a little bit more work up in the upper 12 GHz that has to happen. The government, again, with the NTIA, has to work to get 500 MHz. They have a lot of work to do to find out in the 4 GHz, what are they going to give up in those frequency bands to get more spectrum out there in the marketplace. I think these are all really good opportunities.
Got it. Any questions for the audience in the last minute?
Just a quick one. 5G seems to have rolled out slower than most of our expectations. Is the new tax bill going to help the carriers finally start, you know, moving forward a little faster, or do we need native AI on our iPhones to get the edge going?
A double-edged question.
I think the depreciation piece does help the operators accelerate. I do see, I spent a lot of time looking at the next change into 6G. The specs aren't there yet. It won't happen until 2030. I see that this next level of, I was just showing Brandon the smart glasses, with Apple turning in that, let's say in 2027, but these being in the market today and growing rapidly in terms of one of the use cases for consumers, the next 24 months will be exciting to see how many of these types of products end up and will drive uplink and will drive demand for growth in the 5G. We never really had true 5G because we didn't have a 5G core with a 5G radio. We always had a 4G core with a 5G radio. Some operators were faster than others.
I think they'll all get there now. We could start slicing the network. We could start monetizing parts of 5G that we hadn't been able to do. Before that, it was just best efforts. It really was more incremental ARPU if you could get it by charging more, and consumers weren't ready for that. They said, like, I get everything I need with 4G. We're in a position now, I think, that's really good at monetizing 5G, getting more use cases out, getting these new devices in, and sets the table for why we need 6G. You could argue 5G has been slow. Why do we need to invest in 6G? It's the spectrum, and then it'll be new applications. In that world, we're looking at taking the physical world, which think about AI, right?
We're only dealing with text, maybe some other multimodal, like in terms of image and video. Today, you're going to sense the environment and integrate that. You'll have physical models, as Jensen's talked about, taking the physical world and modeling that. That's going to have to get pushed down to the edge. Machine to machine, so whether you're doing a drone or an autonomous car or a robot or a DoorDash delivery, that's going to require a lot of localization of sensing and communication between that. I see a world of what I would call telecomputing, right? The telecommunications network and computing have to merge in a way at the edge that gives you that user experience. That'll lead us into the next decade. I think that's great for the operators. I think it's great for consumers and businesses and obviously for infrastructure companies.
Great. With that, we're out of time. Ed, thank you very much.