...So our next session is with CEO and founder of Cogent Communications, Dave Schaeffer. Thank you for joining us, Dave.
Hey, Dave, thank you for hosting me. Thank BofA for a great venue, and thank investors for taking the time to hear about us.
Yeah, so I mean, there's a lot going on in the world. Some of it has to do with fiber to the home, some of it has to do with fiber to AI, and everything has to do with fiber, and that seems to be your wheelhouse, so I wanted to, let's start with, Lumen was here earlier. We spent some time talking about the $5 billion contract that they signed, part with Microsoft and a handful of other players.
And they're attributing the win and really the emergence of an entirely new business model to the conduits that Jim Crowe put in the ground twenty years ago, giving them a huge time-to-market advantage in a what is a foot race between all the hyperscalers to kind of build the AI engines of the future. And the question has arisen, you know, why not AT&T? Why not Verizon? Why not Zayo? Why not Cogent? Is this a piece of business that you looked at, or was it or were you passed over because Lumen's fiber, you know, conduit network was just so clearly advantaged in delivering what the hyperscalers need to get?
So it was actually an opportunity that we did look at and engage with Microsoft, who is a transit customer of Cogent and a customer that we've had a very long relationship with as their primary transit provider, and respectfully passed on the business. So in your opening statement/question, there's probably at least three different questions embedded. The first one is, we are a strong believer in fiber as a technologically superior medium. You know, the Internet was conceived really before fiber optic networks were deployed, and the Internet was agnostic to the underlying transmission media. But as the bit volumes increased on the public Internet, fiber became the only medium that would, in fact, support the volumes.
And Cogent built its original network of sixty-one thousand route miles of inter-city fiber and eighteen thousand route miles of metropolitan fiber by buying dark fiber from three hundred and twenty-eight different suppliers, pre-acquisition of Sprint, and Lumen was one of those suppliers. In fact, we bought twelve thousand nine hundred miles from them, and we bought a single pair of fibers. We then deployed equipment that resembled a corporate LAN on a global architecture, and then interconnected that to the highest traffic locations and productized that network solely to deliver internet service. That level of specialization allowed Cogent faster provisioning times and a much lower cost structure.
We survived when nearly 1600 other institutionally funded telecom service companies failed around us, and it was because of that laser-focused strategy on the internet as the only network that would matter, and the belief that you had to be the lowest cost producer per interface routed bit mile, and then you had to have the lowest cost of revenue acquisition, but maybe most importantly, with the advances in technology, you need to capture more of that technology efficiency than your competitor, and those guiding principles were at the heart of Cogent's model. We had the opportunity to actually announce two years ago at this conference, to acquire the Sprint network. The Sprint network was built in another time and for another purpose. It was the first nationwide fiber network.
It was built entirely to carry long-distance voice, and that initial network had a capacity of half a gigabit per pair of fibers. Today, on a single pair of fibers, we can support over 10 terabits with existing technology, and that's continuing to improve. The cost to move a bit-mile has fallen at a compounded rate of 80% per year. On the Sprint network, it had basically been dormant for a decade as Sprint exited the wireline telecom business, and in their enterprise business, only 7% of their customers used the Sprint network. 93% were completely off-net to the network. What we realized was we could solve a problem for T-Mobile in taking over that money-losing enterprise business, shrink that business, exit certain non-core products, rationalize the cost structure, stabilize that business, and make it marginally profitable.
Take a business that was burning $1 million a day and turn it into a marginally profitable business, producing probably about $80 million-$90 million a year of positive EBITDA. But the real upside for Cogent was taking over that fallow fiber optic network. Now, that network turned out to be a little different than the network Lumen has in three respects. One, it's a long, unique right of way. 90% of the network that Sprint has, no one else is on those right of ways. Two, because it was the first fiber optic network to be built, it was built differently than the way Lumen's empty conduit network that Jim Crowe designed was built.
It was, in fact, an armored network, put in this bronze jacket, buried six feet under the middle of the railroad track, versus most of the conduit networks that are two feet deep along public highway right of way. The third attribute that is different is the type of fiber so the fiber that was deployed was SMF-28. It was first generation of fiber, and it was thought in the late 1990s that that fiber would be obsolete, as everybody needed to move to higher transmission speeds and that, in fact, would have been correct if all of the transmission technology remained non-coherent so you could talk to Ciena or to Corning or Nokia, any of Sumitomo, any of the fiber vendors or the equipment vendors.
The movement from half a gig to 40 gigs per wave, and the move from one wavelength to roughly 80 wavelengths, was all done with non-coherent optics. At the time, it was believed that you had to switch to a different type of fiber, non-zero dispersion shifted fiber. The most prevalent of those was Corning E-LEAF. The second most prevalent of those was Lucent TrueWave. Those were the two brands that were most widely deployed in North America, and that fiber turned out to actually be not very good for coherent transmission technology. When the marketplace demanded 100 gigs per wavelength, all of the equipment vendors globally switched their optics from non-coherent to coherent. When that happened, the non-zero dispersion shifted fiber actually performed worse than the original SMF-28. Jim Crowe's thesis was correct, but it was correct for a different reason.
The fiber he put in turned out to be obsolete in a couple of years versus fiber that was a decade earlier, continuing to still be manufactured and employed today. Now, it is true that the brand-new SMF-28 that's bought today, you can go still buy that at Corning, is slightly different. It has better optical purity and slightly lower loss, so you'll gain about one-tenth of a dB per kilometer on the newer fiber versus the older. But the number one reason why the fiber underperforms is actually splice cuts. So each time the fiber is cut and spliced back together, there is loss in that fiber.
We, as a buyer of fiber from Lumen, can absolutely attest that on a per mile basis, there are over twice as many splices on the Lumen fiber today than there are on the Sprint fiber, just because of how it was physically deployed. I do think there is a need for SMF-28. I also believe that AI will generate more demand, but I also think there's an arithmetic problem, and that is how many fibers you actually need to do what you want to do. We are the largest carrier of internet traffic in the world. We carry a quarter of the world's traffic. We do that on one pair of fibers. We are utilizing roughly 30% of the lit capacity in that fiber and have the capability of increasing that lit capacity by at least a factor of ten.
So why would I need hundreds of fibers? Now, AI is a different use case. It's moving large data sets from location A to location B, and that can require very large amounts of transport, but probably not more than the global internet. So a long answer to your question, but, I think the ability to have an extra conduit is really an insurance policy against having the wrong type of fiber. You know, and for that, to that point, actually, Verizon owns an empty conduit on the Lumen network.
Mm-hmm.
It got that from XO. Carl Icahn litigated that, and the courts ruled that they have the right to populate that. Verizon chose not to bid on this business, and AT&T chose not to bid on this business, and the primary reason for that is, as part of the deal with Microsoft, you had to build new fiber, brand new de novo builds, at very low IRRs, and when we looked at it in totality, we just said, "Thanks, but no thanks.
Low IRR for you because you don't have a conduit and you don't have a deal with Corning to get all this fiber, and perhaps reasonable IRRs for Lumen, who's in a different asset position?
So IRRs are IRRs. We have plenty of access to fiber if we need it. There's no shortage of fiber in the world today. Corning, the old Alcatel, now Nokia Fiber or Sumitomo, you can call any of them and get as much as you want.
Mm-hmm.
There's no shortage here. It's just like coffee in the Bank of America coffee pot. It doesn't run out. It's bottomless.
Good analogy. I like that one, Dave. Yeah.
Well, there's plenty of it there. In theory, it could run out, but it's not gonna run out during this conference. The issue was that you had to build brand-new build, not where you have conduit, but to new data centers where there is only one customer, Microsoft. And Microsoft has enough monopsony power, along with other hyperscalers, to drive those IRRs down to low single digits. They typically fund about half of the build upfront, and they expect you to fund the other half. This is not all that dissimilar to the business that got Crown Castle in building small cells. There was this huge mania a few years ago.
Everybody's got to build to every small cell with huge bundles of fiber, and because there were only three or four potential customers, they forced the builds to accept these low IRRs that then did not generate a sufficient return. That's not the only issue with that fiber business in Crown Castle, but that was a contributing factor. We have been disciplined for twenty-five years. Sometimes the best deal you do is the deal you say no to.
Well, you went a lot of years saying no to lots of deals, and then before you got to the Sprint deal, and I want to talk a little bit more about that. But to your point, you know, we had Crown Castle here yesterday. You know, they're looking at all options with respect to the fiber business that they have after Elliott got involved and catalyzed a strategic review of all that. Is there anything about that business that interests you?
So there is real value, but that value is very low, in our opinion. You know, we looked at the companies that made up that business, whether it be FiberTower or Lightower, or some of the other businesses that got rolled up into that. It's about a $2 billion business. About half of it looks like a more traditional CLEC-
Mm-hmm.
-and other half is, in fact, a fiber business. I think their basis is such that there is not a market-clearing price that we would be interested in. Is there some value? Yes. Is there a value that would be all acceptable to the seller that we could get comfortable with? I don't think so. So again, we're gonna spend our time elsewhere.
Uniti was here yesterday, and they're gonna buy Windstream, and Kenny Gunderman said that they're looking at all options with respect to maximizing share price value for themselves. Presumably, that on its face, they're looking at selling their managed services business, which probably doesn't sound like a Cogent type of business to be interested in, but they have a fiber-based services business in second- and third-tier markets. There's probably some assets at Windstream that could be merged into that. Is there anything about that of interest to a Cogent?
Kenny is an old friend. There are really three businesses, and there have been multiple processes that have been public and disclosed around divestiture of each of these businesses, and at least to date, none have come to fruition.
Are you in that F-4 somewhere?
I doubt we would be in that mix. For Kenny, I think there was a natural reason to bring Windstream and Uniti back together, as the lease structure was not sustainable as constructed. There is, you know, the old PAETEC business, McLeod, that has morphed into a managed services business. Most of the revenue is not connectivity-based, but rather value-add services. That's the exact business that, when we acquired Sprint, we were trying to kill as fast as possible because it's gross margin negative and I don't believe there's value in a telecom service provider marking up third-party services. Second, there is a dark fiber business that's the primary Uniti business, and some of that is inside of Windstream as well.
Some of that is resale of Lumen, IRUs that they had, that they could resell, some that were as part of the settlement when CenturyLink and Level 3 merged, and some are their own builds, particularly, some along the coast in the southeast, South Carolina. And then there is a third business, which is kind of their national wholesale business, and I think they're subscale out of their footprint. I think they've looked to divest in many different flavors, both when Elliott controlled Windstream and Uniti, and I think we're more an observer than a participant.
Okay, let's talk about the deal that you did do, which is a Sprint deal. And I would say that I kind of look at it as having three parts, basically. I'm sure you look at it differently. But you know, one part of the deal was, you know, let's get the cost cutting done. Let's get this thing from loss-making to breakeven to maybe making, as you said, a modest amount of profit, profitable EBITDA. And that's on track. Seems that's on track.
That's deal, that's-
That's part one.
Aspect number one.
Aspect number-
I learned agreement, by the way.
Aspect number two was, you know, the creation of new business opportunities, specifically wavelengths, but also maybe dark fiber.
Data centers.
And then I was gonna get to part three.
Okay.
That's part two is kind of the new core business things. And then part three was kind of the hidden value stuff, whether it's IPv4 or data centers, maybe dark fiber is in there, too. And so it seems like the surfacing of value in the hidden value has been a positive surprise, like the ABS that you did for the IPv4. And I think just making people aware that IPv4 is a thing, which I don't think many people did before we started doing our calls on that. But that the wavelength business has been just way behind schedule. And it was weird to me because you know so much more about this business than anybody, that your expectations for this wavelength business were so much higher than what you've been able to deliver.
The line of sight to when you will be able to deliver on the promises that you thought you were prepared to make a year ago is still fuzzy. Let's start out with that wavelength business. Why did you make the promises you made? Why can't you fulfill those promises, and when will you hit those goals?
Okay. So the network that we acquired was not a wavelength network. It was basically a dormant network. When I announced the transaction at your conference in September of 2022, we said it would take us about a year and a half post-closing to convert that network. So there's no real surprise we closed in May of 2023, and we will complete the conversion of that network to making it a wave-optimized network at the end of this year. We're still on track to do that. That conversion has three key components: one, the physical interconnection of the network.
Two...
Two, meaning extending the Sprint network to the metropolitan networks. The Sprint network typically ended 10 or 15 miles outside of a metropolitan market.
Mm.
And then, at that point, there was a connection to the ILEC in that LATA for voice termination.
Mm.
That was what it was built for. Well, because we want to sell wavelengths, we got to get the network extended into the data centers. That physical effort at 110 locations is complete and was complete in February of this year. Actually, a little ahead of when we thought we would get that component on.
What percentage of people listening, you think, know what a LATA is?
Probably those who are as old as you or I, but it's Local Access Transport Area. In the Bell parlance, the country was divided into about a hundred and sixty-eight LATAs, and within a LATA, you could make a local phone call, and if you crossed a LATA, you had to pay a toll. And when the AT&T was broken up, the LATAs were parsed out to different operating Bell companies. There's one exception to that rule, which was, in a few states, there were intra-LATA tolls allowed in very large LATAs, mostly in Texas, and needless to say, since the voice business is gone, the concept of a LATA is gone.
But-
Back to what we have to do. The second key component is the reconfiguration of our metro networks to support wavelengths. We are busily doing that. We have to reconfigure about 12,500 miles of fiber and about 100 markets around the country to have wavelengths go to data centers and IP go to both data centers and multi-tenant office buildings. Then third, we have to deploy two sets of equipment. We have to deploy transponder shelves in 800 data centers, and we need to deploy ROADMs at the physical intersection of the long haul and the metro network. Think of that ROADM as a traffic cop that directs wavelengths to the correct location, and we're probably about 80% of the way through that implementation. When we got started, we knew that there was a lot of pent-up demand.
There was a lot of frustration with the current wavelength supplier base. We hurried up and got about 65 large data centers ready to accept wavelengths. We would still be forced to provision those wavelengths the same way that our competitors would on a bespoke, one-off custom basis. It would typically require six field visits to turn up a given wave. But we thought we would have enough demand, large data center to large data center, to do that. What turned out to be the case is we had plenty of demand, but the demand was at one end in those large data centers, but most of the demand had the second end in a smaller data center.Y
Mm.
The amount of manual effort to go in and custom-build those wavelengths to each of those smaller data centers was just too daunting, and it would have put at risk our entire network optimization program. We made a conscious decision to build a backlog, allow some of that business to fall away, and to focus on the network optimization. Today, if you buy a wavelength from any competitor, it will generally take 90-120 days to be installed, and it will require those six full site visits and generally somewhere between 30 and 60 days of custom engineering to be designed at a total installation window of 3-4 months. Transit services, which is how Cogent had defined its original business, worked very similarly.
Twenty-five years ago, if you went to a data center to buy transit, the biggest port you could buy was a DS3. That was a 45 megabit port. The cost was $300 a megabit at that time.
Mm-hmm.
It would take you about 90 days to get it installed. Cogent came to that market, lowered the price to $10 a megabit, 97% discount to the market, and we cut the provisioning time from 90 days to nine. No, we're not magicians. The way we did that is we designed a network that was optimized to sell high-capacity internet services, and we walked away from every other product and service, including wavelengths, including MPLS. That turned out to be the right decision. When we looked at the Sprint network, we said we can do the exact same thing by taking that dark fiber and optimizing it for wavelengths. We will be able to provision a wavelength from any data center in North America to any data center in two weeks.
The number of permutations are seven hundred and ninety-nine, so it's eight hundred minus one, divided by two, factorial. That's gazillions of potential permutations where those waves can go.
Mm.
No one can pre-provision all of that capacity, but what we could do is build a base network that will allow us to turn up a wave with two field deployments, one at each end, and then automate all of those internal steps. We will win market share based on the ubiquity of our coverage, the uniqueness of our routes, the speed of our provisioning, and at the end of the day, we'll do what it takes on price. I have no doubt we will hit the forecast that I gave at your conference two years ago that said, by May of 2028, in the wavelength business, we will be at a run rate of $500 million. Yes, we are behind getting there because we did not sell big data center to big data center, 'cause that's not where the demand was.
But in aggregate, we know that we will serve the entire market.
That's super helpful. The answer to that question is then, you kinda misunderstood where the demand was gonna be. Why is the demand looking like it does and not the way you thought it would look?
So I think most companies use the internet for large data center to large data center because there's plenty of internet capacity, and the cost per bit mile is two and a half times cheaper than using a wavelength, where if you're at a small data center in a secondary market, there may not be enough internet connectivity with enough reliability to get you to that mega aggregation point. So if you're in Racine, Wisconsin, at a data center, and you need to get to 350 East Cermak in downtown Chicago, which is the largest aggregation point in Chicago, you will probably use wavelengths to do that because there's just not enough internet connectivity in Racine.
If you were at 350 Cermak and you want to go to 1950 Stemmons in Dallas, the largest aggregation point in the south, south central country, you would end up using internet to do that. But if you were in Lubbock, Texas, wanting to go to either Dallas at 350 Cermak or 56 Marietta in Atlanta, you're probably gonna want a wavelength for that. That's just where the market is. Remember, we had no market experience. So just like you mentioned, the hidden value parts of the business, we baked into our financial guidance. And again, it's multi-year. It's not specific annual guidance. We said we will be doing $1.5 billion in revenue. We're doing $1.1 billion now. That'll be in May of 2025. We're doing $350 million run rate of EBITDA.
Today, we'll be doing $500 million of EBITDA, and we will be on a path to continue to grow at between 5% and 7%, with about 100 basis points a year of margin expansion. That's what we said two years ago. That's where we are today, and we will do that through growth in all of these segments. But not included in that is dark fiber, IP address sale, or wholesale data centers. I do want to touch on something you said about IP address space, and I think that's a misconception. Prior to the acquisition of Sprint, Cogent already was the third largest owner of address space in the world and had 28 million.
All of the addresses that we have leased came out of Cogent inventory, not out of hidden inventory that came out of the Sprint acquisition. We did get an additional 9.9 million addresses, all unleased in that acquisition. It's also public knowledge that T-Mobile sold 2 million addresses prior to the acquisition closing. It was simple calculus. They could have sold all 12 million that they had, and they would've just had to write a bigger subsidy check to us. We knew what the addresses were worth. There was a trade-off in the negotiation that determined the ultimate $700 million subsidy payment, so it's not like there was some hidden gem that we didn't know about or T-Mobile didn't know about.
Mm.
Now, what was surprising to us and was an upside surprise is the demand for data center space. We went into this... and again, go read the transcript of the initial announcement. We said we were gonna take 45 of the 482 switch sites that we acquired from Sprint and convert those into data centers, but we said they will look like Cogent data centers. That means they would have 10,000-12,000 square feet at about a megawatt of power, and we would sell one or two racks at a time to corporate customers. Prior to the acquisition, Cogent had 55 of those facilities and 634,000 square feet of raised floor and 69 megawatts of power. We've actually uncovered three more that make sense. We're actually converting 48, now 45, than originally set out.
But what we found was there was an extra million sq ft and an extra 100 megawatts of power that we can't use. There's a lot of demand in the market. We're rapidly converting that, enabling that to be sold, and we've started conversations with people to buy it. That is hidden value we had not anticipated. And then on the dark fiber piece, we knew how many fibers there were along the 19,000 route miles of intercity and 1,200 miles of metro. The fiber count ranges from 24 strands to 144, depending on the segment. We will generally use 6 strands to run our business. That means there's a lot of extra fiber on every route. We have not started to sell that because we need the resources that would provision that dark fiber to focus on getting this wavelength project complete.
Once that's done, we're gonna monetize it, so we will be willing to sell that dark fiber when we've got all of this foundational wavelength work done. Did that fully answer?
No, that was, that was good. Thank you. I think I've gotten four questions, and we're almost done. So, just quickly on the data centers, can you tell us a little bit, what is it gonna cost from an upfront perspective to do this conversion of these legacy central offices into data center space that can actually be used by people who need real data center space? And can you give us any concrete examples of assets that have transacted that we can kind of get a baseline for what this could be worth? I mean, data center by data center?
Okay, so the typical data center is a brick-and-block building of about 40,000 sq ft, sitting on six or seven acres of land in an industrial park that's 10-15 miles outside of downtown. That's what it physically looks like. There will be multiple Sprint long-haul fibers coming into that, and now there are multiple extensions from that facility to the metro footprint in that given market. In those facilities, there were 22,500 bays or cabinets, 3-foot wide, 8-foot tall, full of telephone switches.
Mm-hmm.
DMS-250s, 5ESSs, things. If people don't know what LATAs are, they surely don't know what these are. We are down to 6,000 still left, so we've pulled out the vast majority, 16,500, but we've got to continue to pull them out. We have to go in and test and recertify the battery systems, the fire systems, the alarm systems, and the generators, and all of the backup facilities. The good news is, the serving power is in place. There's 230 megawatts of inbound power coming into these facilities. There's generation and transmission capacity in place. The most important step is, all of these facilities were designed to operate at negative 48 volts DC. When you run a data center, all the power comes in off of the grid. Usually, it comes in high voltage AC.
You have an on-prem transformer that steps that down to 480 three-phase. That then comes into the facility, it gets inverted to DC, it charges a battery plant, and then there's an inverter that turns that back into AC 120. There's a transfer switch sitting between the generator and the batteries, so if there's an interruption to inbound power, the center runs off of the batteries, the generator kicks in and charges the batteries, as opposed to the power grid charging the batteries. That's how pretty much every data center is built. In our facilities, there is... That last step is missing. There is no rectifier to invert that power. We are deploying those now. So it's a combination of housekeeping, general cleanup, testing systems, and getting these inverters in place. We are working on that. We've actually accelerated that.
We thought we would do this over a three- or four-year period. We're now trying to do it in a one-year period, and this is in response to the demand. There is not a single transaction completed today. I don't want to mislead you or misguide you on the likelihood of those. What I will say is, we went out to 115 counterparties at the end of April. We had an additional 18 inbounds, so we've had 133 parties that we've spoken to. We have 58 of those parties in some kind of active negotiation or discussion. Some are interested in just one site, some are interested in multiple sites, and none of the sites are ready to plug a server in. By the end of the year, there will be a subset of sites. This work will not be completed year-end.
It'll bleed into next year. Again, we have finite resources, but we're trying to get these marketable while there is this short-term, strong demand. We priced our space either on a leased basis or a sale basis at about a 40% discount to where the market is. We are pricing these at a rental rate of $1 million a megawatt a year or a sale price, be simple, at $10 million a megawatt, and we'll see what the market says in response to that.
It's a great place to leave it, Dave. Good conversation. Thank you so much as always.
Thanks, Dave.
Appreciate it.
Sorry we didn't get to more questions.
Oh, you got the... You know, this would take three days to do. Appreciate it, as always. Thank you so much.
All right. Thanks, Dave.
All right, good.
Did I start to wow you yet?