One, thanks for joining us. For those of you who do not know me, I am Nick Del Rio, I cover digital infrastructure at Moffett Nathanson, and I am thrilled to be kicking off our 2025 Media Internet and Communications Conference. It is our 12th one of these over the years. Appreciate you joining us. This is our commercial fiber panel, and I am thrilled to have two phenomenal speakers with us. To my left is Dave Schaefer, the founder, chairman, and CEO of Cogent Communications, and to his left is Kenny Gunderman, the president and CEO of Uniti Group. This is the first year that Kenny is with us, so thank you, Kenny, for joining us.
Pleasure to be here, Nick.
Dave has been kind enough to join us for all 12 of our conferences, so special thanks to Dave. With that, let's get started. There's obviously a lot going on in the commercial fiber space today. We've only got 40 minutes, so I thought we would focus on four particular topics: demand, strategy, network expansion and evolution, and the competitive landscape, with an emphasis on the parts of your business that overlap or are adjacent. Let's start with demand. AI-driven demand has really changed people's expectations and perceptions of the space over the past year or so, especially in light of some of the big dark fiber deals that have been announced. Now, Kenny, you've shared that about 15%-20% of your bookings come from hyperscalers. I think you've said that two of your top 10 customers are hyperscalers.
You said on your earnings call last week that demand remains strong from those customers. Can you dig in a little bit? Can you dig in a little bit and maybe talk about the level of urgency in the discussions and the magnitude of the capacity that these customers are seeking?
Sure. Nick, again, thanks for having us here, and it's always a pleasure to be here with Dave Schaefer, who I admire. Get lots of questions about the hyperscalers. I always try to remind folks that they're still a very small percentage of our revenue, a very small percentage of our business. One of the things that we love about wholesale fiber is that we're diversified across all customer bases and all fiber use cases. Therefore, we pointed out that last year was a really big year for us with the hyperscalers, but still, they were just two of our top 20 customers. Very small part of our business, but growing dramatically.
A couple of years ago, a few years ago, I would say the percentage of bookings from hyperscalers was virtually zero, and the percentage of hyperscalers in our funnel was virtually zero. The TAM for fiber businesses with hyperscalers was also very small. Recently, we estimated for the market our view that the TAM for hyperscalers today is probably $15 billion, and that's up, we think, dramatically. We think it's going to grow to probably $40 billion-$50 billion over the next number of years. Of course, we want to capture our fair share of that, and I think we have been so far.
All that to say, Nick, I would not characterize it as urgency because that kind of sounds a little bit undisciplined for the hyperscalers, but there is absolutely a sense of purpose, I would say, towards building out the infrastructure that I think is enabling AI. We get questioned all the time about the quarter-to-quarter performance, but the reality is I think they are playing the long game. I think they are making long-term investments. They say to us and to the market that if they do not use those investments for AI, they are going to use it for other things because I think it is critically important for their business to have high-speed broadband proliferate. For us, we are seeing dramatic increases in strand count. A few years ago, we were selling 12 to 24 strands. Now we are selling 30 to 40 times that amount of capacity.
Whereas before we were probably selling more lit capacity, we're now selling a lot more dark fiber and a lot more conduit. All the things that customers do to enable for a substantial amount of capacity is what we're seeing from their behavior.
Okay. Okay. That's great. Dave, as you work to build your wavelength business over the coming years, how important are AI-driven use cases going to be to that? And do you expect that to help your transit business as well?
Yeah. The underlying network infrastructure continues to be repurposed for new use cases. Much of the infrastructure that's being used for AI was actually originally constructed to carry voice traffic and then was adapted to carry internet traffic, private data, and is now being used for AI applications. There are really two very different use cases for transport in AI. The first of those is to link large data sets to the locations where training is occurring. Oftentimes, the power and facilities exist in different locations, and fiber is the best way to do that, whether it's lit services or dark fiber that is then lit by those providers.
Because approximately 50% of the cost of a training facility is in the GPUs, you want to make sure that GPU capacity is being used very efficiently, the 99.99% efficiency, 24 hours a day, seven days a week. A fixed connection is the best way to do that. AI has become a large incremental driver of transport needs. Those transport needs can be met either through the purchase of dark fiber, as Kenny had indicated, or wavelengths, or a combination of both. The second use case for AI is just now emerging, which is really the inference phase. That is using much smaller data sets, taking the large language model outputs, the neural networks, and then applying them to a specific question and generating a dedicated dark fiber or wavelengths.
In both cases, this is increasing total bit traffic demand, and that is good both for the internet, which still dominates the total volume of traffic transmitted. The internet is the cheapest, easiest to use, most ubiquitous way to move bits, but it does not provide a defined path with specific latency. For that reason, there's this incremental use case, and it's a very large one for dark fiber or wavelengths.
Kenny, you noted a bit of a shift from lit services to dark fiber. Dave, you noted some fungibility between waves and dark fiber. When do you see customers opting for one versus the other?
It varies, I think, and it's hard to generalize customers. They all have different buying patterns, and those buying patterns change over time depending on their access to capital, depending on their network needs. Generally, Uniti sells a substantial amount of dark fiber. That's really our core business today. We have a growing waves business, and we expect to continue to grow that business. Dark fiber is really our bread and butter. What we find is when customers are making longer-term network commitments that require substantial amounts of capacity, that they can foresee that capacity over a longer period of time, they're going to buy dark fiber because they're making a commitment to light that fiber for themselves and therefore making a commitment to CapEx, OpEx, and ultimately helping manage that network.
There's a lot of requirements there that aren't there if they're lighting that fiber themselves. When I apply that back to the hyperscalers, again, 24, 36 months ago, we were selling waves to the hyperscalers, or we were selling small amounts of dark fiber, and we were selling more metro fiber. Versus today, to today's point, we're selling largely long-haul routes. We're selling some greenfield builds for the first time in a long time, building long-haul routes, connecting those data centers, those large language model data centers. Rather than buying waves right now, they're buying dark fiber, 400-plus strand count, 800-plus strand count, and even in some cases, a second empty conduit next to it. That just implies a substantial amount of capacity needed. We often get the question, does that mean that they're buying for the next five years or the next 10 years?
In some cases, when customers buy dark fiber, they are. They're planning for that long out into the future. I think in some cases, maybe some of the hyperscalers actually are, but what we have found in only the last 12 months is that we're now having hyperscalers come back to us and buy another 400 strand count on top of what they just bought 12 months ago. That suggests to us that they're using the capacity that they're putting in the ground now, and they certainly foresee the need for more of it in the future.
Okay. Great. Let's give it to non-AI demand, which is the bulk of the demand you guys are saying. Dave.
Probably eight minutes in, and we're moving off AI.
Lots to talk about. Dave, you entered into the Sprint transaction and built the business plan for wavelengths before AI became the thing it is today. How is that portion of the demand funnel related to the pre-AI business case materialized relative to your expectations? What are some observations you can share about the composition of that demand?
Yeah. When we evaluated the Sprint GMG network and eventually acquired it from T-Mobile, that acquisition actually was two concurrent acquisitions. The first was an operating telecom business selling to legacy enterprise customers that was in decline and burning cash. We were compensated for that by T-Mobile to cover those burns and try to correct some of the deficiencies in that business. The second thing that we acquired was the physical network of Sprint. It was built between 1982 and 1991 at a capital cost of $20,500 million. It was comprised of 19,000 route miles of owned dark fiber and 482 buildings along the way. All of that infrastructure had one purpose when it was deployed. Long-distance voice. Sprint shut down that long-distance voice network in 2015. For all intents and purposes, the network was foul. It was empty.
It carried a little bit of de minimis traffic. We acquired that asset for a dollar. There were challenges. The asset went to the wrong locations, and it needed to be repurposed. We focused on wavelengths as the highest and best use of that network. We do have dark fiber and will sell some of that dark fiber on a route-by-route basis. The three primary customers that we were initially targeting were either regional networks who needed to link islands of traffic together and had been long-term buyers of wavelengths, two international carriers looking to extend their network, and three content companies that needed to distribute content not over the internet, but rather doing it through a dedicated network. That is particularly applicable to live events that are very gender-sensitive.
With the transition of sports from pure broadcast to now being streamed in real time, that is an incremental use case that has grown wavelength demand. The AI and hyperscaler demand is yet additive to that. For Cogent, we needed to look at the total addressable market and be able to serve it. Sprint was connected to 23 data centers initially. We ultimately extended that and were at 883 North American data centers where we can provision a wave. The network that we architected because we had the advantage of having a clean sheet of paper looks very different than our competitors. Most companies that sell wavelengths and even sell dark fiber started with a network that was selling a variety of heterogeneous products. We were able to optimize that network for wavelength deployment, giving us shorter provisioning, broader footprint, and the ability to lower prices.
The market is dominated by a couple of large players. We do occasionally bump into Kenny in a region of the country, but it's pretty regionalized. Conversely, we're a customer of Kenny's where we buy dark fiber to supplement our footprint. This is definitely an industry where companies both compete and cooperate.
Kenny, how would you characterize the non-AI demand that you're seeing today? Wireless, fiber to the home, and so on.
I would say it's been very, very solid and has been for the past number of years. It's one of the reasons why we always love to talk about generative AI and the hyperscalers, but I always like to remind folks that that's just a portion of the demand. As Dave said earlier, it's an incremental TAM that has really revealed itself over the past 12 or 24 months. You step back and you think about all the different use cases for fiber. You don't have mobile broadband without fiber. You don't have fixed wireless without fiber. You don't have satellite without fiber. You certainly don't have fiber to the home or waves or fiber into the enterprise buildings without it. It's essentially the connective tissue for all broadband. I know that sounds like a buzz phrase that we use a lot, but it's true.
As a result, when you think about the industry, it's growing at somewhere between five-10% over a longer period of time and probably for the foreseeable future because there is no technology out there that's going to replace that underlying fiber, especially in the core backbone. When we look at our customer segments, there are years when some are up and others are down. For example, we get asked about wireless carriers all the time. They are big customers of ours, but when you look at it on an aggregate basis, the wireless carriers are probably less than 20% of our revenue. When we combine with Windstream, it'll be less than 10%. Last year, wireless spending was down, and it was down the year before.
We think this year it might be up a little bit, but last year was a record year in bookings for us because other segments were up. This year, we expect fiber to the home carriers to be our biggest segment of wholesale customers again this year like it was last year. All this money that's being spent to build fiber to the home, those customers have to buy backhaul to connect those markets back to the core. Terrific customer segment the past couple of years. I think that customer segment's going to continue to grow. We think the hyperscalers are going to continue to grow. We think large enterprises are going to begin procuring large amounts of capacity as well, especially as we get into the inference phase around AI. I think just the tailwinds in the industry are strong.
If we're smart about which customer segments we're exposed to, we're going to do well regardless of whether one segment is up or down. I think we're well diversified. We don't have any specific customer segment that represents more than 20% of our revenue. We're well diversified.
Okay. Great. Let's turn to strategy. I think a lot of observers tend to lump fiber providers together when in reality there are a host of different business models involved and very varied strategies to try to capture value. In fact, you guys both have very distinct strategies that you pursue. When I think about the biggest players in the market, almost by default, they've embraced what I'd call a classic telecom strategy. You have as big a network as possible. You try to drive as much revenue over it as possible. You guys have both rejected that approach. Dave, what do you see as the primary flaws with that kind of classic strategy and why you opted to pursue the strategy that Cogent embraces?
Yeah. A common thought for telecom executives is, "I got to move up the value chain. I got to sell higher-level applications," which is really another way of saying, "I can't make money at the network layer." We realized that bandwidth is a commodity. It is subject to very rapid price deflation. You needed to focus on the fastest-growing segment of the market. You needed to be able to capture advances in technology more effectively than legacy networks. You needed to be very disciplined about where you deploy capital as endpoints because you are overbuilding legacy technologies, whether it's copper, coax, wireless. In each case, the internet can overlay those applications. Remember, for the vast majority of the market, virtually all consumers and all businesses, the internet has now become synonymous with data.
We realized that purposing the fiber for the internet was the correct strategy. What the internet was effective in doing is decoupling the application from the network. Traditionally, a carrier built a network in order to deliver a single application, whether it be linear television or two-way voice communication or closed data communication. The internet changed that. The internet was application agnostic. It was cheaper on a per-bit basis. It allowed new business models to be developed. Virtually all of the traffic that is being carried today was not even available 20 years ago for legacy carriers. The legacy carriers have been struggling as their core revenue streams get migrated to over the top on the internet. At the same time, they see their capital intensity going up as their antiquated distribution networks are not effective.
We focused on two endpoints: really big office buildings and the central business districts of major cities. The reason there is we thought there was enough end-user demand per facility to generate an adequate return on capital. You needed the discipline to pass by and not serve all of those smaller locations because your revenue acquisition cost was too high in that very diffuse market. The second thing we connected to were data centers. Think of them as supermarkets for bandwidth. They have continued to proliferate. The use cases have changed. That narrow strategy has allowed us to grow organically and generate free cash where virtually everyone else in the enterprise business has seen those legacy businesses shrink.
Yeah. Now, Kenny, when I think about Uniti's strategy, I think primarily infrastructure in second-tier markets, right? So the supposition is that maybe you trade off some revenue opportunities, but for a lower level of competitive intensity. I guess, is that characterization right or wrong or incomplete? And what do you look at internally to kind of confirm to yourself that not moving into kind of Tier-one markets is the right approach?
Yeah. You nailed it, Nick. It's not complicated. We like to stay as close to the infrastructure as we possibly can. We like to go where there's less competition, which tends to be Tier-two and three markets and long-haul routes that are less trafficked. Dave knows there are many routes out there that are built and overbuilt and overbuilt. There are other regional routes that are more unique. We greatly appreciate Dave being a customer of ours on many of those routes. It's not more complicated than that for us. I think Dave nailed it. I think a lot of companies in our space try to get as big as they possibly can because scale does matter in telecom. There are synergies for being bigger. They also try to add as many products as possible on top of their offering.
Our view is that you do not have to be super big in certain areas. You certainly do not need complicated products. When we look at product set, we always ask ourselves, "Is it scalable? Is it simple? And is it something that we can make money at?" Oftentimes, the answer is no. We try to stay as close to the infrastructure as possible and try to compete on the quality of the infrastructure as opposed to technology and price, which tends to be the case with complicated products. Building in areas where there is not fiber or less fiber, we believe you are right, Nick, that if you look at it statically, the growth potential in Mobile, Alabama, may be less than New York City. In New York City, there are 12 or 14 fiber providers.
Once you're a fiber provider and you're in a city, you're going to try to cannibalize other providers' business. I mean, that's just what you do. You've got salespeople on the ground selling. If you get into a Mobile or a Birmingham or a Pensacola or a Shreveport, Louisiana early, you dissuade other fiber providers from coming in generally. We have a second fiber provider in less than 15% of our markets in Uniti Fiber. That's terrific, which means that, yeah, maybe the pie is smaller, but we get a bigger slice of it. Our growth potential is essentially assured as long as we're smart about which customers we target. That's ultimately true on long-haul routes as well. I think the industry at a time was overbuilding, and we were not very disciplined about building back in the late 1990s, early 2000s.
Dave picked up a lot of those networks for cheap when they went bankrupt. Now there's a lot more rational spending. If you've built a long-haul route, that does not mean you will not get overbuilt, but it is less likely, substantially less likely than it was before. When you take that strategy and you apply it to our combination with Windstream, that is essentially what we are going to be doing with Kinetic, building fiber into these smaller markets. We are going to be the first fiber provider or certainly an early provider going into those markets. It is really simple. Back to your point, you nailed it. Infrastructure; Tier-two, Tier-three markets. That is really our strategy.
Okay. That's a great pivot point to talk about network expansion and evolution. You guys have very different philosophies when it comes to the appeal of building new fiber, right? So Dave, as you noted, your network was either primarily built out of dark fiber historically or you picked up Sprint for a dollar, so a good price for that. Kenny, obviously, laying fiber is an important part of your business. It's kind of interesting that you have two very savvy veteran executives who view the appeal of building fiber so differently. Yes, I guess, Dave, what do you think you see with respect to new fiber construction that others don't? Or are there risks that you worry about that others may not?
Cogent's core business was delivering internet services globally. We have amassed a network of a little over 100,000 inner-city route miles of dark fiber and another 34,000 route miles of metropolitan fiber in 267 markets across 57 countries. All of that was by buying fiber from others, 325 different suppliers, and then using that fiber as efficiently as possible to sell the most commoditized service, internet services. When we got started, there were 200 global backbones, nearly 1,400 metropolitan competitive carriers broadly defined. Virtually all of those companies have disappeared. We found that the only construction that made sense for us were very limited extensions from manholes in the street into a building or up the riser to serve those buildings. We chose not to build to small cells, not to build new long-haul routes, but rather to go out and buy.
That model still is a great model for Cogent. We continue to expand our network through that model. We're doing an expansion right now in India. I couldn't even tell you how complicated it would be if we tried to build in India. It's complicated enough to just be able to purchase dark fiber in that market. When we acquired Sprint, we acquired a fallow asset. We chose not to build. In the places that Sprint was deficient, we bought dark fiber to supplement it for a different use case, wavelengths. We have not been able to justify an adequate ROIC. Kenny and others have different core competencies, different market focuses, and can generate that return because of the large number of customers that they sell that fiber to. Whereas if we built it, we would be the primary user.
There's generally just not enough return to justify the construction cost.
Kenny, would you agree with that characterization?
Yeah. I think everything Dave said is right, of course. I think, first of all, there's room for different business models in the industry. We're a wholesale fiber provider. We want people who use our fiber to be successful. Dave's a customer. He's obviously very successful. We want our other customers who theoretically are using off-net fiber for their businesses. We want them to be successful. We want the hyperscalers to be successful. I think that at the end of the day, Dave's also right that there are many parts of the country that do have multiple networks, do have multiple fiber providers. Being a fiber owner in those markets is not appealing to us. We want to be where there's not another fiber provider or where there are fewer other fiber providers as a result.
Back in 2017, 2018, we went on a very large build. We entered into a very large build phase. We built into about 10 different metros, largely in the Southeast, that we identified as great areas to build network where there were no other fiber providers and had good competitive dynamics with really large carriers. We used some of the things that Dave mentioned, like fiber to the tower or small cell deals or building for a large E-Rate school, building for a couple of large healthcare campuses. We used those as anchor customers. The thesis was we're going to build that network in a range of a 5%-10% initial yield, which is not particularly exciting. We had the thesis that we could build with that initial cash flow yield.
Every dollar that went into the ground, there was an immediate 5-10% cash flow generating customer that was locked in for a 10- or 20-year contract. We would come back later and sell that network to other customers, the second, the third, the fifth tenant, very analogous to the tower model in that regard. We track ourselves with that every quarter. We tell investors, going back to that build phase, those initial 5-10% yields are now yielding close to 30% cash flow because we've leased them up effectively. Our capital intensity, while back during that period of time, was over 50%, it's now down to 20-25%, which is a comfortable area for us to be in.
That shows that if you're smart about picking the right markets, picking the right demographics, the right competitive demographic, and the right anchor customers—by the way, all those anchor customers are still on those networks and in some cases buying more capacity over time—again, those are customers who are using someone else's fiber. They are successful in those markets. We are successful in those markets. It is just a strategy difference versus others who have built fiber in larger markets and more trafficked routes. For us, that strategy works. We are going to continue that on a go-forward basis.
Okay. Great. Dave, some of your larger peers are spending a lot of time and effort trying to modernize their networks to offer novel services, automate provisioning, be able to offer services on a consumption-based model instead of a flat price model, and so on. Do you think that sort of approach has merit? Do you think it's likely to appeal to customers or any risks if you don't follow suit?
Yeah. So I often hear the term network as a service. I'll be honest, I have no idea what that means. Every network.
That's more than you would tell me because I don't know.
Every network is a service. I also hear the concept of variable bandwidth or bandwidth on demand. The internet provides that. You get a connection. You use it when you want to use it. When you don't , you are still paying for it, but it is available. The real cost is in the physical layer connectivity to the customer. That is what Kenny said about constructing and getting an adequate return on fiber. The services you run on that network fall into two primary categories: internet, which is now then available for every application you can imagine, or a dedicated transport network, i.e., wavelengths or dark fiber. Those are the two primary customer use cases. In delivering either of those, we as operators benefit from the advances in technology.
There are two basic technologies that drive innovation, the first being improvement in wave division multiplexing, i.e., getting more bits, move more miles for a lower cost. There we are very far away from hitting the limits of material science or the capabilities of fiber. The first system that Sprint deployed as a telephone system carried 565 Mb, i.e., a half a Gb on a single pair of fibers with a repeater every 12 mi. That was revolutionary compared to a microwave system that actually could carry only half of that traffic. It was then scalable. Today, that same fiber, the exact same physical fiber, can carry 160 wavelengths with roughly 40 Tb of capacity per wavelength by modulating more effectively using more colors. There are still further advances.
The cost of moving that bit is coming down at about 80% a year, faster than Moore's Law. The second advance is in routing. That is a more specialized technology, fewer vendors, and only about a 40% per year price performance improvement. In running a network and selling services, it ultimately comes down to what is your cost of manufacturing the underlying service you are selling. If it is a network, can I move more channels? I remember in the early days of cable going from a half a dozen linear channels to Gerald Finn saying that we are going to have a world and John Malone of 500 channels and people believing that was impossible. Now we are in a world where there are tens of millions of channels. It is that innovation that you have got to be able to adapt to.
Okay. Great. In the couple of minutes we have left, I thought I'd ask a wavelength question for both of you because that's an important part of your strategies going forward, Sprint for you, Dave, and Kenny with the Windstream acquisition, accelerating your plans there. I guess as insurgents, you guys have nothing to lose. Incumbent providers have everything to lose. It raises some interesting questions regarding pricing and the manners in which the incumbent providers might respond to your attacks. If we look out a few years, how do you guys anticipate the market changing in terms of whether it's product attributes, pricing, common capacities, and so on?
I'll comment really quickly and love to hear Dave's comments on this point because Dave's a bigger waves provider than we are. I think the waves market's going to continue to grow. I think it's going to grow in capacity. It's going to grow in dollar size. I think there's going to continue to be three or four or five large waves provider. As you said, Dave and I are both insurgents. We're share takers, whereas the other three or four have embedded bases of lit transport, really, in some cases, not modern technology lit transport. It's older technologies that have to be transitioned. Therefore, that creates opportunities for people like Uniti and certainly for Dave to take share.
I think we're going to continue to compete on the same characteristics going forward that we're competing on today, which is network quality, uptime, provisioning intervals, and price. There are other things, of course. Those things tend to be important when we—and I did not mention probably the most important one to Uniti, which is unique routes. I think we try to compete on those terms. I think that's really roughly the order. A lot of people worry about price compression and price pressure. I think those first three or four characteristics are super important to customers. I think they're going to continue to be important. I do not see a race to the bottom on price. I think there's definitely competition. There's certainly competition on certain routes.
If you've got more unique routes and you've got network quality, you've got good customer relationships, and you can turn up waves in a timely fashion, I think you're going to be able to compete.
Okay. Dave?
I got a flashing red light, but I'll try to go quickly. I think eventually in the developed world, there will be fiber ubiquity everywhere. We're not there yet, but we are getting there quickly. It comes down to the use cases, internet being the largest of those, private networks, wavelengths, dark fiber being the second. The advances in technology are going to continue to drive down price per bit mile. Carriers have to look at other parts of their cost structure, their cost of revenue acquisition, their billing systems, their provisioning systems, their customer care systems. All of the legacy providers struggle with rigid fixed cost structures that have not been able to flex to this lower revenue per bit model.
The dirty little secret that people do not want to talk about is, yes, the unit volume is growing, but the actual dollar spent for services is still a shrinking buy. That does not mean there are not insurgent niche opportunities. If you are a legacy provider that has a very rigid cost structure and your total addressable market for the services you sell are shrinking, you need to be able to flex your costs faster than your revenues are declining. That has not happened well for the legacy providers.
Okay. Great. Dave, Kenny, thank you so much for joining us. That was terrific.
Thank you, Nick, for hosting. Thank you.