Okay, good morning. Sorry, we're a couple minutes late there, but great to have Dave Schaeffer, CEO of Cogent, here again at our TMT conference. My name is Frank Louthan, and I'm the Senior Wireline Analyst here at Raymond James. And Dave, you're the one CEO I almost didn't send a list of questions and topics to, because you don't need it.
I didn't read them anyhow.
Of course you didn't, because you can spout chapter and verse on all this stuff. But let's talk a little bit, start talking a little bit about Cogent. Tell us a little bit about who you are, how you fit into the space, and who you compete with, just kind of level set for folks.
Yeah, so first of all, thanks for inviting me, Frank. Thank the investors in the room for their attention, and Raymond James for a great forum. Cogent's a 25-year-old company. It is the largest carrier of internet traffic in the world. We operate a global network in 56 countries that carries a quarter of all internet traffic. We connect to 1,750 data centers around the world and a billion sq ft of office space in North America where we sell to end users. We operate a fiber network that was built out of IRUs, where we bought fiber from other providers, nearly 300 providers, to piece together that global footprint. And then we have supplemented that network by acquiring the Sprint Global Markets business from T-Mobile in May of 2023. In buying that business, Cogent changed three ways.
One, we acquired a large enterprise customer base that we had never had previously. These are Fortune 500 companies that operate global networks, and we now service those customers. But we've been in the process of grooming the product set and the customer locations to ensure profitability. We were paid $700 million by T-Mobile over a 54-month period to acquire that business. The second way we changed was we acquired a physical fiber network that was built between 1982 and 1991 at a capital cost of $20 billion, $500 million, and was sitting basically dormant. That was the world's first transcontinental fiber optic network, 19,000 route miles of intercity fiber, 1,200 miles of metropolitan fiber that was designed to carry TDM voice. That network was sitting idle for a decade, and we are repurposing that network to sell optical transport or wavelength services.
And then third, we acquired 482 physical buildings, fee simple-owned real estate, that were central offices for the Sprint network. That was 1.9 million sq ft of real estate buildings and 230 megawatts of inbound power. And we are repurposing 48 of those as data centers. So Cogent has evolved as its business got larger, its EBITDA grew, and as we have had the concurrent efforts of repurposing assets and reducing burn of an operating business.
All right, great. So let's talk about the wavelength business a little bit that you mentioned. Talk to us about what is a typical profile of a wavelength customer, what are they buying, what are applications that they're using for wavelengths that maybe they aren't using for some of your other more traditional products, or that they would be buying dark fiber for, that sort of thing. Give us an idea of insight into that market.
So first of all, the cheapest, most ubiquitous, and easiest way to move bits is the public internet. The internet is easy to use, but it has three structural limitations. One, it does not have defined latency. You don't know how long it's going to take for a packet to get from one location to another. Two, it is not as secure because it is a shared medium. And third, it has packet size limitations. So companies who need one or more of those characteristics buy a private network. That private network can be bought in three different flavors. You could buy wavelengths, you can buy dark fiber, or you can go out and build fiber. The wavelength market globally is about $7 billion. $3.5 billion of that is in North America. $1.5 billion of that $3.5 billion are metropolitan waves. $2 billion are intercity waves.
200 customers account for 80% of the market. And the customers buy these wavelengths for four significant purposes. One, they need to move content between locations. Two, they need to extend networks to new locations. Three, they need to interconnect multiple islands of networks into a more holistic network. And then recently, there's been a fourth application, which is AI training, that is taking data that is resident in one location, moving it to another location, often where power exists, using that to create AI large language models and then returning that traffic. So those four types of customers can either buy wavelengths, buy dark fiber, or build fiber on their own. Dark fiber is more expensive than wavelengths. Building fiber is more expensive than buying dark fiber. So there's a pecking order. Some companies value that security and control, so they will build their own or buy dark fiber.
But on a pure economic basis, if you can move the bits over the internet, you should. If the application requires these characteristics of latency and packet size limitations, you then would use wavelengths.
Great. Okay, so that's helpful. So what are some of the challenges? One of the goals you're having to get your locations set up to be able to fully offer this, get that set up by the end of this year, which we're coming down to the wire here. What have been some of the challenges, and where are we on that schedule to be fully ramped to sell the wavelength products across the full breadth of your network?
We will next month be in a position to sell wavelengths between 800 data centers in North America and any-to-any data center connectivity and provision that wavelength in two weeks. That is substantially different than how the industry sells wavelengths today. There's approximately 140,000 installed intercity wavelengths in the market today, accounting for that $2 billion. The market is dominated by companies such as Lumen and Zayo. The wavelengths are deployed on heterogeneous networks selling multiple products, and they are built on a custom basis. A typical wavelength install takes three to four months and requires six field dispatches to complete. What we have done at Cogent is taken an empty voice network and repurposed it and optimized it for wavelengths. The four steps that we had to initiate were first, the physical connection of the long-haul network to the metro network.
The Sprint network typically stopped 15 miles outside of a city in a rail yard where they built a central office and then bought lit interconnection to the ILEC in that market. We extended that network in 110 locations around the country to intersect the pre-existing Cogent metropolitan network. We have roughly 17,000 route miles of metropolitan fiber in North America. We then had to reconfigure those metropolitan networks to optimize them for wavelength delivery. We had to deploy reconfigurable add/drop multiplexers at the intersection of the long-haul and the metro network to direct those wavelengths. And then finally, we had to deploy transponder shelves in those 800 targeted data centers. There are 4 times 10 to the 1980th power number of permutations of wavelengths possible on 800 data centers. So there's no way you can pre-provision that capacity economically.
If you can tighten that number up a little bit more for us, that'd be helpful.
That's a really big number. Considering Google is 10 to the 100th power, this is 20 times more exponential power. So don't ask me to write it out. I'll be here for a couple of days writing on a blackboard.
Of great faith, you could do it to every decimal point, but anyway, go ahead.
So what we have done is built a network out of a former voice network that can deliver these wavelengths more quickly along routes that are unique and be able to provision to more data centers in less time than anyone else. And maybe most importantly, because we acquired this $20 billion asset for $1, we have a zero cost basis, which gives us a great deal of pricing flexibility.
All right, great. So one of the controversies we hear about this is the TAM, so the addressable market and what the real value is. Talk to us about that and your opportunity, what you see as your opportunity, the amount of market share you think you can get. I think you put some numbers out, but what you think you can ultimately, your revenue can be in this. But talk to us about the TAM and how realistic it is to get to your ratable share of the market.
The market, as I said, is $2 billion. It's measured by four independent research firms that all size it relatively equally with 140,000 discrete wavelengths. The wavelengths are priced three ways based on size. They come in three sizes: 10 gig, 100 gig, and 400 gig. That'll evolve over time to 800 and 1 terabit. Secondly, you price a wavelength based on distance. And third, the longer the term, the lower the price. Wavelengths are sold on contracts between one and five years in duration. We looked at that market top down and said, we have 25% market share in the transit market. We can replicate that in the wavelength market. That's a top down view saying we can get to $500 million. A much more precise way to look at the market was bottoms up. We have 287 wholesale sales reps that sell transit services.
80% of the wavelength market is bought by 200 buyers. 150 of those 200 today buy from Cogent. We looked in the sales decks of each of those reps, asked them to project the number of wavelengths that they would sell over a four-year period. We discounted that for the kind of inherent optimism that salespeople have in their ability to succeed. And we're comfortable that we were still exceeding the 25% market share that we looked at from a top down perspective. So our target is to go from 1,100 waves today and a $17 million annualized run rate, looking at Q3 numbers, to mid-year 2028 doing $500 million in wavelength revenue with about 35,000 wavelengths sold.
Okay, great. So you talked about your cost basis here. You kind of famously used your cost basis before to drive pricing down. Does hitting that? How does that goal hit on a unit basis for wavelengths relative to where the market is today? To what extent do you think you're going to use your pricing power to buy that market share to get there? Or do you think you get that just naturally by it being a two-player market that customers are going to want a three-player market and you'll just naturally steal share? How do you think about that?
Yes, so there are some niche regional players in the market beyond the two national players. We'll be the third national player. Second, we are very focused on this market with a larger sales force. We do have qualitative attributes that make our product superior, whether it be the uniqueness of the routes or the universality of the endpoints. But at the end of the day, as in any commodity business, price matters. In the IP market, we became dominant by undercutting the market by 50%. We continue to do that and are today the largest carrier of internet traffic. I don't believe we're going to have to be quite that aggressive on the wavelength market. My guess is we'll be undercutting the market 20%-30%, but it is a more custom priced product as each route is different.
We'll do whatever it takes to win that market share. We are committed to gaining share. We hope we don't have to go to a 50% discount, but if necessary, that's what we'll do.
Okay. All right. So yeah, go ahead.
Hey, David. Could you clarify the integration of the assets and what to think about in terms of when you get to a certain milestone and what we should think about as far as the cost to upgrade everything? And then lastly, you talk about unique routes and differentiated routes. Are they tiered? Are they tier one, two, three, or somehow? I mean, the routes vary. And what sort of pricing do you get on those? And how do we think about that in the big picture?
All right. Let me take those in reverse order. So the Sprint network was built along railroad right of way. 90% of the 19,000 miles of owned routes Sprint are the only party in that right of way. Most of the industry built its networks along public highway right of way. So that gives us a unique advantage for diversity, and a wavelength is an unprotected product. So buying diversity is extremely important. In terms of the market served, it's 110 markets. That's both tier one, tier two, and tier three cities, whether it be New York or Los Angeles, whether it be Indianapolis or New Orleans, or further down to places like Huntsville, Alabama, or Cheyenne, Wyoming. Basically, the Sprint network went anywhere that were long distance voice calls. In terms of the integration, there are actually two integrations happening concurrently.
The first integration is the customer base that we acquired that was burning $300 million a year. We were paid $700 million to take it. If we didn't fix anything in two years, the subsidy would be gone and we'd be left with the losses. What we had to do was change the cost structure of that business. We cut G&A. We moved traffic from off-net to on-net where possible, and we eliminated gross margin negative products. We eliminated service on low capacity circuits, and then finally we eliminated service in countries where there were low volumes and without proper licensure, and we're still in the process of extracting those cost savings. We have been able to get roughly $165 million of the $220 million of targeted synergies in the first 18 months. We said it would take 36 months. We're ahead of plan.
And we've also said we'll take that 220 up. We have not quantified what the new target is. We have got the burn rate on that acquired business down from roughly $300 million annually of negative EBITDA to negative $40. It'll be zero by this spring, and it will turn into a profitable business, albeit on a smaller total revenue scale in 2025. The second integration is the network integration, the physical layer network that had to be integrated, and that has two parts to it. On the wavelength network, we had to do the physical extension to tie our network and their network together. That is complete. We had to do the reconfiguration of our metro networks and then deployments of bays, ROADMs, and transponder shelves. That will be complete at the end of this month, allowing us this ubiquitous coverage.
The second integration is taking a subset of those buildings that we acquired and turning them into data centers. We had initially targeted 45 of the 482 buildings to house a Cogent retail data center. That was going to be done at virtually no capital cost, and it was going to take one megawatt and roughly 10,000 sq ft in each of those buildings and add it to our existing 54-building footprint, so we would have a total of 99 data centers, and we would have increased our total power to about 105 megawatts, or excuse me, 115 from the 69 that we had previously.
As it became obvious that there was an acute shortage of power and data center space, we changed our thinking at the beginning of this year and have targeted three more facilities to be converted and then to take one million sq ft and 100 megawatts of excess power and market that off on a wholesale basis. So that work will continue through mid-2025 and be complete.
All right, great. And that kind of gets to another question. So maybe talk to us, so you got a total investment there. What do you think you can get from a rent perspective? What price per kW do you think that you'll be able to lease that data center space at? It's clear. We cover the other data centers. There's a lot of demand here, but your space is a little chopped up, so the demand is a little bit different. Talk to us about where you are with pricing and how you think you can move that space.
Yeah, so on the network side, we said we were going to spend $50 million tying the networks together. That money was spent in 2023 and the first half of 2024 and is fully expended. On the data center side, we are going to spend $100 million refurbishing this 100 megawatts of power and 1 million sq ft. That work began mid-year 2024 and will continue to mid-year 2025. It will be spread across 21 facilities. We are going to market offering that space under two different business models. Either you can lease it at $1 million a megawatt per year, triple net, and you would have to sign a long-term lease with CPI adjustments, or you can buy the space outright, fee simple, for $10 million a megawatt. Both of these price points represent about a 40% discount to the current market price. Now, you're absolutely correct, Frank.
These facilities were not built as data centers. They were built as telephone switches. What do they look like? They're typically on six or seven acres, 15 miles from downtown, initially only connected to the Sprint network near a train track with a brick-and-mortar building of about 50,000 sq ft that housed tens of thousands of telephone switch cabinets. We've been depopulating this gear. We've been converting the negative 48 DC plant to AC, refurbishing the fire systems, the generators, the UPSs, and we went out starting in April of 2024 to 115 counterparties. We've had 18 inbounds, so 133 interested parties. Roughly half of them are still talking to us. We are literally daily having multiple tours across the footprint, and we're in negotiation with those parties, both for purchase or for lease. There are some parties who want the whole footprint.
Some just want a subset of the footprint.
All right, great. Well, the fact that you can sell some, you have the daily leases kind of tells you where we are in the demand cycle. A couple of years ago, I think those facilities wouldn't be as attractive, but I think that'll be an interesting reflection.
I think that's right.
So talk to us as we kind of have running up on the time here. Maybe we'll go a minute over on IPv4 leases. Give us the outlook for that. And then the question we get a lot from investors is, what's the longevity of those assets? The potential they get displaced by something else. How should we think about that as a long-term revenue source?
I'll try to keep my answer short, but I think some foundation is necessary. One, when the internet was started, there were three basic technologies: TCP/IP to connect computers together, BGP to connect networks together, and a unique addressing scheme, IPv4, which created 2 to the 32nd power addresses, which is 4.3 billion addresses. Those addresses became scarce as the internet succeeded. The US government retained 800 million addresses. That left 3.5 billion addresses for the world to use. A new protocol was introduced in 1998, IPv6. It has 2 to the 128th power, but only today accounts for 7% of global traffic. So V4 is still the dominant way to connect. The world ran out of addresses in 2017. We began leasing those addresses out. Initially at $0.30 an address, we've subsequently raised prices to $0.50. We are leasing about 30% of our 38 million address inventory.
We are competing with Amazon and Microsoft, which spent 11 years building an inventory by acquiring addresses and leased their addresses out at an average price of $3.60 per address per month. So nearly a 7x our pricing. Will IPv6 eventually replace IPv4? Probably, but it's been 26 years since it was introduced, and it's got 7% market share. I think there's a long tail here that we'll benefit from. Today, there's an active market to buy and sell addresses at between $40 and $50 an address. That market peaked at $60 an address. We have 38 million addresses that really don't show on our balance sheet. So the three assets that are latent to our business are the dark fiber and our Sprint network that we're not using, the excess data center space that I've just discussed, and now these incremental addresses.
Great. Well, the rate of change in this industry is always a lot slower than people think. We're still talking about copper phone lines 25 years after I started covering this sector.
You probably have the Princess phone at home, Frank.
I do have one that I picked up somewhere. It's not plugged in, but I've got one for the museum. All right, Dave, really appreciate the time today. Thanks, everybody, for being here and for being on the webcast. And everybody, have a great conference.
Thank you very much, Frank.