Well, good morning, everyone. Thanks for joining us. I'm Nick Del Deo with MoffettNathanson, and I'm thrilled to be joined by Dave Schaeffer, the founder, chairman, and CEO of Cogent Communications. Dave's been kind enough to attend all 11 of our conferences. Appreciate your consistent support over the years.
Well, thank you for inviting me. I think I've usually had the dubious honor of being the smallest company in your coverage universe. So at least I make it consistent, maybe not the largest. Thank investors for their time, and MoffettNathanson for a great venue.
Okay. Well, great. Well, listen, Dave, you know, I could, I could literally interview you for hours, given the amount that's going on in Cogent. But I wanna spend the time we have focused on a, on a few particular topics: waves, synergies, hidden assets, and capital allocation. You know, so maybe we'll start with waves, 'cause that's really the number one opportunity ahead of Cogent. The demand seems to be there, installation seems a bit of a challenge. The demand seems to be a bit more atomized than you expected. I think the way that you explained it to me a little earlier was sort of more large to small data centers rather than large to large, so it's more diffuse. You know, having all those data centers enabled is really gonna be a key to getting this, this business humming.
So can you give us confidence that by year-end, you're gonna have the network configured, the data centers enabled, the sales processes, the field processes in place to really, to really press the gas on this opportunity?
Yeah. So I'm gonna start by looking back at Cogent's historic transit business, which had many of the same challenges when we got started. We needed to be in a ubiquitous number of data centers, and we needed to be able to provision quickly. We've identified 800 data centers in North America that we believe generate virtually 100% of the addressable market for North American transport wave services. We currently have equipment in 420 of those. We will complete the installation by year-end in all 800. It is the second part that's probably the more challenging and more important, which is to be able to provision those wavelengths quickly. One of the key differentiators for Cogent is going to be the ability to turn those waves up in time frames similar to we do with transit services.
In order to accomplish that, we need a standardized network architecture. We have deployed a series of physical connections from the Sprint network to our metro networks. Those are complete. That was probably the one part of this reconfiguration that could have potentially been out of our control. We now have completed 100% of those in our connections. The second is the deployment of those transponder shelves. We're about 55% of the way done, and that's the easiest of the tasks. The third task is the deployment of ROADMs to be able to direct the long-haul waves into the appropriate metropolitan ring. That is about 50% complete and is a relatively simple task. It is the final task that's the challenging one, which is reconfiguring 800 rings that touch 2,660 buildings, segregating multi-tenant office buildings from data centers.
We are working diligently on that. We have hundreds of those rings already reconfigured, but we definitely are not complete. We have a clear path to do that, and we should be able to then provision a wavelength, any to any, any data center to any data center, with a two-week provisioning window. You know, we've got nearly 1,000 people working on it. We've got a series of processes in place. It is all under our control. It is a large number of field exercises, but we are on track to hit that year-end goal. We can definitely sell the wavelengths in all the locations, but what we have to do is be able to provision them much more expeditiously.
When did it become more apparent that demand was spread across a broader base of data centers than what you expected? And, you know, the reason I ask is because I feel like you're kind of guiding the street to an opportunity that people can't quantify, right? I think they rely a lot on your judgment here. So to the extent that the demand was a bit different than what you anticipated, and that had implications for the pace of revenue growth and installations, you know, it's probably not great from a investor perception perspective, so maybe help us out there a bit.
So let's maybe attack the question from two different perspectives. First of all, there are independent third parties that size the aggregate market. Our independent channel checking and customer conversations kind of say the market is about $2 billion, which triangulates to the three third-party studies that have sized the North American transport market: Atlantic-ACM, TeleGeography, and IDC.
Mm-hmm. A second way to look at it from a total market size is what our competitors are doing in wavelength. They are not as transparent in breaking out their products, but it appears that they are today capturing 100% of that $2 billion market. We always knew that we needed to have a ubiquitous solution to be able to maximize the value of the Sprint network. We took our experience in Transit and projected it. Within Transit, we served 1,680 data centers around the world, 800 in North America, and there is a roughly 80/20 mix, where 80% of Transit volumes come out of the largest 20% of the data centers. We said: Let's target those very large data centers.
These are aggregation points like 1 11 Eighth or 60 Hudson right in New York, 165 Halsey in New Jersey, or, you know, 350 Cermak in Chicago, 1950 Stemmons. These are major carrier hotels where multiple data centers are resident in those buildings. We had a finite amount of field resources. We began focusing on those facilities before we had any demand. We continued to collect demand with, you know, about 400 waves provisioned and about 2,400 waves in the funnel, and that demand has come in with a large data center, in the vast majority of the cases, connected to a smaller data center, rather than large to large. We had no way to know that in advance till we had specific orders. We built a heat map by route, by data center, to try to help direct our prioritization.
Eventually, we have to do all of the facilities. We have to standardize the architecture to hit our objectives. So we knew that the initial data centers would still require longer provisioning cycles. We were able to provision some initial waves in about 60 days. Our wave provisioning time is today kind of normalized at around 120. It's actually gotten worse, because we're manually provisioning waves across this more atomized universe, as you described it. With the architecture we're deploying, we will be able to do a much more simple provisioning. That provisioning will require three work efforts: a physical site visit at the two endpoints to install a pluggable optic and a virtual configuration through our network operation center to build the path between those two endpoints. Our competitors today provision waves much the way we do. They take a customer demand in two locations.
They go out and survey their network to figure out what kind of field work is necessary. They typically do multiple, 3, 4, 5 field deployments in order to put equipment in, to be able to have a contiguous path for that wave, and then they provision it. It is why the industry averages 3-4 months to provision a wave. Our design will allow us to do that in 2 weeks. You know, why we got it wrong? We just didn't have enough actual customer orders. You could do all of the third-party market research, but ultimately, customers buy where they need services.
Okay. You know, so, you know, while revenue may be ramping at a slower pace than you would hope for, you've noted that demand has been stronger. So, you know, longer term, that's probably the more important takeaway. What do you think has been behind the, you know, the stronger-than-anticipated demand?
So we anticipated three pools of demand: content replicators being the largest group. This includes both hyperscalers and smaller content-generating companies, from CDNs to hosters that have multiple sources of data that they're trying to synchronize. That is the number one use case for wavelengths. The second use case is regional access networks that operate as disjoint islands that are then connected. Cable companies, small telcos are the bulk of this group, some international carrier demand. The third group is by far and away the smallest, which are organizations that value the security of a wavelength over the internet. A wavelength, on a per-bit-mile basis, is more expensive than the internet. So you have to have a reason to need a wavelength. For the two previous use cases I described, the latency and packet size variability justify the premium.
In the third group, it's the security, and that is typically very large enterprises and banks. That's probably less than 10% of the aggregate market. There has emerged a fourth group, and I'm almost afraid to use the term-in front of investors, of AI. We were joking about it before I got up here on the podium, and it's like all you have to do is say AI, AI, and your stock is great. Well, the reality is, AI is very distributed today for two reasons: the data for the training models and the compute sit in different locations, and secondly, the compute is being distributed to multiple locations because of power limitations. Both of those facts have driven extraordinary demand from AI. It is incremental. It was not anticipated when we acquired the Sprint backbone.
Okay, that's great. Let's switch to synergies, right? So, you know, obviously, a big part of making the Sprint deal work is putting a tourniquet on Sprint and stemming those losses. I think you've laid out a $220 million synergy target. I think you've shared that you've reached about 40% of that goal. I think, you know, we could spend a lot of time trying to match what you've realized to date versus the reported financials. I know there are a lot of moving pieces there. Let's, you know, let's focus on what's gonna happen prospectively. Can you help us understand some of the specific sources of synergies that you're looking at going forward, the cadence at which you think you can realize them, and kind of the split between OpEx and savings on principal payments?
Yeah. So, we had the good fortune of being able to schedule out, with a fair amount of detail, places where we could achieve cost savings in our due diligence. 93% of Sprint's revenues were derived from off-net services. We could map those customers and see where we can move them to on-net. Reducing that telco cost is the largest single area of cost synergies. The second area was headcount. We knew how many people were in the business when we signed the deal. There were 1,320 employees in the Sprint wireline business inside of T-Mobile. Based on our interaction with T-Mobile, they were able to reduce that headcount to 942 at closing. We have taken out roughly 225 people.
We're down to about 718, and there are some additional individuals who cannot function in the new organization and will be taken out. A third area is redundant facilities elimination. We completely eliminated the Sprint international network that was costing $25 million a year. We migrated that traffic onto the Cogent network. It originally had about 60 locations on that footprint. There are only 3 of those 60 that still exist and there is no traffic in those remaining three locations. We were able to eliminate the connectivity between them by migrating onto our fiber network. So a $25 million bucket achieved and clearly identified as international.
There is a $15 million bucket of cost of goods savings that is yet to be achieved by migrating the Cogent IP backbone over 12,900 miles in North America onto the Sprint backbone, eliminating a $15.5 million O&M expense that we're paying to an IRU vendor, where we've already secured that IRU through 2031, but have the ability to exit the maintenance expense in mid-year 2025. So that'll be a cliff fall. The other $205 million, well, minus the $25 million, so $180 million, will come much more ratably. It will probably be roughly 70%-80% COGS savings and 20%-30% SG&A savings. There are some office facilities that are being eliminated. That's an area of SG&A savings. There are technical leases that are being consolidated. Sprint had previously had a presence in 23 carrier-neutral data centers. All of those are being eliminated and migrated into Cogent's footprint.
There are leased technical facilities, a large presence at NAP of the Americas, 111 Eighth Avenue, and the Westin Building Exchange in Seattle, which are major carrier hotels where Sprint had a significant presence that's no longer needed. There are a number of operating fiber leases that Sprint had that go away. There's an additional uneconomic capital lease that we will be able to eliminate and have go away. So there are very quantifiable, specific routes or specific locations to tie to those COG savings.
Okay, and, and just to be entirely clear, the $50 million uneconomic lease you cited, that's a, that's a finance lease or capital lease?
That is a finance.
Is that, is that savings in the $220 million, or is that in addition?
That's above and beyond.
That's above and beyond the $220 million.
Correct.
Okay.
Because the $220 million includes operating expenses. It does include a significant number of short-term operating fiber leases. So Sprint's network was comprised domestically of 19,000 route miles of owned inter-city fiber, 1,200 route miles of owned metropolitan fiber, and approximately 6,000 route miles of leased fiber. Of that leased fiber, only one of those 2,800 miles of fiber were this capital lease. The other, roughly 3,200 miles, were operating leases that will all go away.
Okay. Okay. I wanna, I wanna reconcile some of these targets with longer-term guidance that you've laid out. Okay, so if I look at 2022, which was your last full year before the Sprint deal closed, Cogent produced about $230 million in EBITDA.
$233 million .
$233 million. I think you've argued that post-5, about 5 years post-close, you should be doing about 500 total new business. You've also said that you'll be doing about $500 million in waves revenue around that time at a very high incremental margin. You just laid out the synergy targets. You'll have some underlying growth in the-
Core business
... core Cogent business. I guess if I add all that up, it seems like you'd get to a number well in excess of $500 million.
I think that's correct, Nick.
Okay.
What we did is kind of two things in laying out that guidance. One, it did not make sense to put a more aggressive number out because it's five years out, and I think people need to see incremental proof points. They needed to see a general high-level rationalization and a base case, but something that we could do much better on. Two, we did not have perfect clarity to all of the customer relationships. We saw five names out of 1,396 customers that Sprint had when we signed the deal. Some of those customers are gonna go away. Some of that is good, and that we want it to go away, and it's baked into our synergy numbers. Some may go away that do contribute EBITDA margin. So we built in some additional cushion into our numbers.
I think as we think about our multi-year performance on a blended basis, the kind of three businesses, that being the Cogent classic business, the incremental Wave business, and the inherited no-growth or very low-growth enterprise business, should grow at a blended rate that is below where Cogent has historically grown at 5%-7%, and we outlined about 100 basis points a year of margin expansion. If you just extrapolate last quarter, we had 260 basis points of sequential improvement in one quarter. Well, obviously, the investor looked at it and said: "Well, you could just keep that up, and you'll be back up to better-than-Cogent margins within, you know, 3 years." That's not realistic. Those synergies are not gonna come to that magnitude, and they're not gonna be perfectly linear.
What we tried to do in laying out that multi-year high-level guidance is give ourselves a little bit of flexibility on the run-off rate in that enterprise business.
Okay. We appreciate you sharing that thought process. Let's turn to your hidden assets. I mean, it seems like you have several hidden assets in the business, I think most notably IPv4 addresses, the data center space you're converting, as well as dark fiber, which seem to be worth a lot of money, and you're not getting credit for it in the marketplace. Let's start with the IPv4 addresses, right? So you own nearly 40 million of them in total, and the larger blocks that you own seem to trade for, we'll call it $50 or so per address. So simple math would say, in total, $2 billion pre-tax value. And obviously, you're leasing some of them out, so they're in your, in your results.
If you look at the unleased ones and apply that sort of value, you're probably in the $1.2-$1.3 range, something like that. Is that fair, or are there other considerations we need to take into account?
So our hidden assets hide in plain sight. It's not like we somehow camouflaged these or did not make them available. Now, we didn't highlight them, we didn't focus on them, and they were a relatively small part of Cogent's business until recently. Those addresses have consistently increased in value. Like with any asset, it's not a perfect straight line. It goes up and down quarterly, but the trend line since 2011, when the first addresses were bought or sold, has been about a 12x appreciation in those addresses. We began leasing them. We, by year-end 2023, we're generating a $35 million run rate on those addresses with virtually no costs associated with them. If we sold those assets because we have no basis in them, we would pay taxes. We also had a couple of lucky breaks that we couldn't have planned for.
One, Amazon and Microsoft started leasing at a much higher price, and two, the transition to a competing technology, IPv6, has been much slower than anyone had predicted. You could go back and look at dozens of research reports for the transition from IPv4 to IPv6, going back to 1998, predicting within two years, we're gonna be transitioned. Today, it's only 7% of the internet. So I think these addresses have a long-term value, as well as short-term ability to produce cash flow. We chose to initially securitize those assets. That securitization had two purposes: one, to raise money. That was simple. I get that, that's why you borrow against it. But two, to show that there was intrinsic value by bringing new investors in without paying taxes. That does not preclude us from selling addresses or doing another securitization. We're gonna look at both of those.
We get it. Investors want the money now, not in the future. Now, the reality is, if we sold those addresses, we would probably just keep the cash on the balance sheet. Cogent has been pretty good at returning capital. In fact, we get criticized sometimes because for 14 years, we've paid out more than 100% of free cash flow. That's not by accident, that's by design. We had an unlevered balance sheet and the ability to grow EBITDA and disgorge cash with incremental borrowings. Our addresses have played a component in that and may play a larger piece in that. The final point I'll make on addresses is, the two largest buyers of addresses over the past decade have stepped back recently. Prices have softened in the short term.
Yes, we may sell into that softer market, but there are other assets on our balance sheet that the demand is at a peak level now and may make more sense to monetize.
Okay. You know, so I think you laid out, you know, some of the logic for not selling them, right? Appreciated assets, you've got the tax, potential tax bill, the bids at any given time-
And leasing-
And l easing revenue.
I guess on the flip side, what do you view as the greatest risks to hanging on to these addresses? So, for example, like, the Department of Defense owns a lot of these things.
They become worthless. You know, it's like you buy Bitcoin, and then tomorrow, nobody wants Bitcoin. What's the intrinsic value of a Bitcoin? It's zero. Can't touch it, feel it, or see it. Can't even use it. At least the address has a current use, and there's a real cost in renumbering. So there is a date and time when these addresses will not have value. That date and time, however, is pretty far out, and there is probably more ability to harvest cash for a longer period of time than I think a sale would recognize today. That may change. You know, we have improved our commission payouts to accelerate leasing. We have removed restrictions associated with who we lease to, and then third, we have begun raising prices.
The beauty of having two large competitors set a pricing umbrella that's 12x your price, we don't have to get to their price, but we have a lot of headroom to increase revenue per address. So as I look at the unused inventory, we have three levers to pull: more leasing volume, higher price per lease or potential sale.
Okay, let's turn to the data center conversions you have underway. I think kind of long story short, you've talked about there being about 170 MW in total in the facilities that you're converting, of which maybe 100 MW or so you're making available for sale or for lease. How are you thinking about valuing that capacity?
So when we acquired Sprint, and in our projections, there are no numbers associated with the data center opportunity. We have, again, been lucky in that the demand for data center space has increased rapidly since this deal was announced, and there is an acute shortage of power, driven mostly by that AI training application that we talked about earlier with wavelengths. As a result, what we are doing with these data centers has changed in our thinking. So our initial model was to look across the 482 facilities that we acquired that are fee simple owned, that actually have 230 MW of power. Take those centers that are in bad locations or too small, and just ignore them. Take the larger of the facilities, depopulate them from telephone gear, and convert them to data centers.
We had initially thought these data centers would look much more like Cogent's existing data centers, where we would sell one and two racks at a time. In these facilities, we've identified 45 locations to be converted to data centers that have 170 MW of the 230 MW in the entire acquired footprint. Of those 170 MW, are spread across 45 facilities, we went a step further and identified 21 of those facilities where there is a surplus of 1 million sq ft that we don't think we can lease out, and 100 MW of power that should be monetized by either selling the facilities or by leasing them out to other operators. Within each of these 21 facilities, we're dividing it into three discrete areas: 0.5 MW and 1,000 sq ft to house our technical equipment, to run our network.
An 8,000 sq ft-10,000 sq ft space that would have 1 MW of power, where we would run a retail colo service, much like Cogent has done across the 55 facilities we had pre-acquisition. Then 1 million sq ft and 100 MW that we should either sell or lease to third parties. We've just gone to market. These facilities are not completely ready to market. Actually, I forwarded you a sales brochure earlier today on one of the facilities. As we complete this conversion over the next 6-8 months, our expectation is some will be sold, some will be leased. Our model is simple: We will lease facilities on a triple net basis at $1 million per megawatt per year, with CPI adjusters, with lease terms from 3 to 20 years.
We'll accept any of those terms, or we will sell the facility at a capital cost of $10 million per megawatt. Now, if we sell the facility, we will become a tenant for that half a megawatt that I described for our network and a retail colo, but these are not sale leasebacks. We are taking only a very de minimis portion of the total capacity of the facility.
Okay. And, you know, when you had mentioned $10 million as the sale price you're looking for, my initial reaction was that seemed elevated relative to what we saw for other comps in the market, given the size of the facilities, the locations, their age, and so on. I think you made some interesting points earlier about, say, land or expansion options that might help to bridge that gap. So maybe briefly talk about that angle.
So let's talk about what these facilities are. These larger facilities were tandem switch sites. They were built generally 10 mi-15 mi outside of major metropolitan areas. They're in an industrial park. They sit on 6 or 7 acres of land, and they have a 50,000 sq ft building with, call it, 6 MW of power. There's variance. The largest one has 13 MW, the smallest has 3.5 MW, but 6 MW is pretty standard.... It's a brick-and-block building that, when it was initially built, used 100% of the power that was brought to it to run a bunch of telephone switches. For the past 10 years, these have sat dormant. They're still full of old telephone switches, and they've been maintained.
You know, the buildings have had air conditioning and battery systems checked and replaced, generators have been exercised, but there's been no active commercial activity in the facility. We're going through and ripping out the old telephone equipment. We're connecting these facilities to metropolitan networks that make them much more marketable in terms of having greater connectivity, and then at these facilities, there are two expansion possibilities just beyond what is initially there. One is the ability to potentially bring more power into the facility because the transmission capacity already exists. The answer is yes in some of the facilities and no in others. The second is to be able to build additional land on the land that's on those footprints. In most of the cases, the zoning allows for incremental data center expansion at this facility.
So if we look at the comps, I think that these actually show pretty well because there is this incremental opportunity.
Okay. That's helpful and something I hadn't appreciated previously. You know, the last, you know, real kind of fallow asset or hidden asset that you've talked about is dark fiber, which is probably the toughest for those of us on the outside to conceptualize in terms of value. A lot more nebulous, it seems. Maybe talk about how you think about the value of that latent asset.
So, couple things. One, Cogent's been a fairly active purchaser of dark fiber from 328 different sources. We have roughly $520 million of capital lease obligation on our balance sheet that we still owe on that fiber, and we have probably spent, either us or predecessor companies, several billion dollars to acquire that fiber. So that's us as a buyer. Two, there are some companies whose primary business model was to sell dark fiber. At the very beginning, that was Level 3's initial business plan. It pivoted away from that. At Zayo, that remains a significant part of their business. At Windstream, at Uniti, it's a significant part of their business, and we have a nationwide footprint of 20,000 mi, you know, 20,200 mi.
We're gonna typically use a few pairs to run our business, meaning our IP business and our wavelength business. All of the excess fiber is surplus. Now, the reason we have not started to monetize that is the same people that would provision that are the people that are entirely focused on wave enabling the former Sprint network. Remember, we took over a network that was not designed to sell wavelengths. It was built and designed to carry long-distance telephone calls. That doesn't mean it can't be repurposed efficiently. It absolutely can, but it takes thousands of discrete projects to make this happen. We need to complete the wave enablement of the network, then we'll turn to selling dark fiber. None of that is in our projections. We don't know what it's worth because there's a great deal of price variability depending on the specific routes.
To complicate that, the cross-sectional densities are different. We have some routes that have 24 fibers. We have other routes that have 144 fibers. Probably on average, 60 fibers or 70 fibers per cross-section, but we got to test the market to make sure that where the demand is hopefully matches up where we've got the inventory. I am certain there's gonna be some stranded inventory that'll never be sellable.
Okay. So, not totally clear from the inside as well in terms of what-
No.
Yeah.
As I think about these three buckets of assets that you've identified as hidden, I would say we've been transparent they exist. They're just not being monetized by Cogent today effectively. There is probably about $3 billion of value that obviously is in addition to whatever you value the operating business of Cogent.
Yeah. Well, you know, in the time we have left, I kinda wanna ask a question that revolves around that. You know, I think given where the stock price is, I mean, I won't ask you for a number, but I believe you think it is trading well below its intrinsic value. I would certainly argue that's the case. I think a number of the people in this room would probably say, against that sort of backdrop, even if there are leakages or inefficiencies involved with selling some of these fallow assets or surplus assets, wouldn't the best use of your capital today be basically selling what you can and buying back stock? Like, is that not the most accretive capital allocation option you could take today?
No, I'm not sure it is. So, first of all, there, as I mentioned, there's a tax inefficiency, there's a transactional inefficiency. But Cogent has been, I think, among telecom companies, one of the most astute allocators of capital. We have invested in network assets that generate higher ROIC than any of our competitors, and we have taken excess capital and returned it to shareholders. The method of return has varied over time. We've bought back roughly 22% of our outstanding float, but we have 47 sequential quarters of issuing a dividend and growing the dividend quarter- on- quarter. The right answer may be to just give it to the shareholders and let them make a decision. Whether it's a one-time special or a recurring dividend or a buyback is to be determined.
The final point is, if assets are appreciating at a faster rate than your stock is appreciating, holding those assets could make some sense over time. There is, no one has a perfect crystal ball into the future to know when the IP addresses will peak in value, when the rapacious demand for power will be met, when everybody who wants dark fiber has enough, and there's no more demand. Right now, all three of those things are seeing increased demand. We have an immediate problem, which is we have finite resources to get these ready to market. You know, I've heard the question raised: Are you worried about an activist? And listen, my goal is to maximize value creation for shareholders.
Not every shareholder will agree with the strategy we've taken, but I think over Cogent's life cycle, we have utilized capital and generated better operating returns than any other company in telecom. And I know that's a pretty bold statement, but I think, you know, 20 years of being a public company actually plays that out. And equity is permanent, so you can satisfy short-term owners by financial engineering, but you create long-term value by allocating capital at a substantially higher return than your cost of capital. And it's not clear to me that selling assets today and buying back stock or issuing a big dividend is the best way to get that maximum ROIC.
Okay. Well, unfortunately, we're out of time, Dave. This is a super informative discussion. I appreciate you sharing the time with us.
I want to thank everyone for listening. Thanks, Nick.
Okay. Thanks, Dave.