All right. Good morning, everyone. Thank you for joining us at the Wells Fargo TMT Summit. Very pleased to be joined by Dave Schaeffer, the founder and CEO of Cogent. Dave, thank you for joining us.
Hey, Eric. Thanks for hosting me. I'd like to thank Wells Fargo again for just a phenomenal venue. It's rare that I get a view of the ocean from the stage where I'm presenting. I want to thank investors for taking time to hear a little bit about Cogent.
Dave, just a couple of days ago, you released an 8-K and a press release about kind of capital allocation and the buyback. Maybe you could just talk about stepping back, the decision to reduce the dividend 98% to kind of help save, call it, $200 million of cash a year. Initially, you suspended the buyback as well. Then you put it back in place a couple of days ago. Maybe you could just talk us through the decision process to go from where you were before earnings to where we are today.
Yeah. Cogent has been fortunate that it's been able to implement a dividend and grow that dividend for 52 sequential consecutive quarters. It returned about $1.9 billion to shareholders. Part of that strategy had been controversial in the sense that we had always dividended out more cash than we were generating, but we were leveraging the incremental EBITDA. If you go back to when Cogent went public, with actually the help of one of your predecessor banks, [audio distortion] , we were EBITDA zero. Over the next 17 years, our EBITDA grew at a compounded growth rate of about 18%, all organically. We decided to first use free cash for buybacks and then supplemented that with a dividend. Most importantly, added leverage. We had no debt when we went public, which was highly unusual for a telecom company.
When we acquired Sprint, our EBITDA took a material step down. It went from basically $60 million,$65 million a quarter and 40% margins to about $4 million in EBITDA because we assumed a business that had a - 80% EBITDA margin. We were subsidized by T-Mobile. That cash came in, and our leverage actually in the short term went down because those payments from T-Mobile were not equally spaced out. When those higher payments rolled to lower payments, our leverage creeped up. It hit a net leverage of 6.6 x. That was beyond management's comfort level with aggregate leverage. We have three uses of cash. We could either use it to give a dividend, to do buybacks, or to delever. We made a conscious decision prior to earnings release that we were serious about delevering and put a very specific target out of 4.0 x net leverage.
To get there, we needed to show both bondholders and equity holders we meant what we said. We were careful in choosing our words. The dividend was reduced to $0.02 a share, a 98% reduction. We kept the dividend in place for any shareholder that had to have a dividend, but it's de minimis. It would give us $200 million of incremental capital to accelerate the delivering. We wanted to also indicate that we weren't just going to take 100% of the dividend payments and divert them to buybacks. We said we would temporarily pause that buyback program. We have $105 million of authorization in place.
With the violent reduction in the share price, which was probably not warranted by either the fundamentals or the dividend reduction, because dividend had already decoupled from the stock, with the dividend yield being almost 11% prior to earnings, it kind of signaled to us that people were not buying the stock for the dividend. With that set off, the board reconvened and said, "Maybe we went a little too far. We could effectively remove the pause immediately." That does not, and I want to be clear to investors, mean that we're necessarily immediately actively buying, but it just gives us the flexibility that we had previously. We are absolutely committed to accelerating the pace of getting from 6.6x net leverage down to 4x .
You've talked about how you will return to revenue growth in the fourth quarter and going forward. Maybe you could talk a little bit about how the legacy Cogent business, the corporate business, the net-centric business, not including the impact from Sprint or the Waves business that you're ramping, how that's performing today. I think you've talked about corporate growing about 3%-4%, net-centric in the higher single digits. How should we expect that to grow over the next couple of years? I know the corporate side has gone through some tougher periods since the pandemic. Are you starting to see any light at the end of the tunnel there?
Yeah. So Cogent's organic business for 17 years grew at a compounded rate of 10.2% a year. When the pandemic hit, Cogent's growth rate decelerated to 5% a year. With that deceleration in growth came a reduction in the rate of margin expansion. So while our margins continued to expand in that legacy business, they were only expanding at 100 basis points a year as opposed to the 200 basis points a year we had demonstrated for the previous 17 years. With the acquisition of Sprint, we did not acquire any net-centric customers. We did acquire a number of large enterprise customers, which were new to Cogent, and some corporate customers. Virtually all of that business was off-net. Much of that revenue was derived from non-core products.
We have worked diligently to groom that revenue, bring as much on-net as possible, to jettison the non-core products that were negative gross margin, and to retain the business that made sense for Cogent. The acquired Sprint revenue, which was 40% of the combined company in May of 2023, so two and a half years ago, over that nine-quarter period has declined at a compounded rate of 24.2%. The reported rate of revenue decline for the combined business has only been 2.4% during that period. The underlying Cogent business actually accelerated somewhat during that period in its growth rate. It's also been helped by the growth of the wavelengths business, which is completely new to both Cogent and Sprint. It is still a small part of our revenues. We only have about 1.5% market share in that addressable market.
The wavelength business today only represents about 4% of Cogent's revenues. It is growing rapidly. The underlying core business should continue to grow at kind of mid-single digits with the blend of corporate and net-centric. For the past two years, it has actually outperformed that. I know a lot of investors, pessimistic investors, think that that business actually is declining. The fundamental arithmetic of the fact that the Sprint business has gone from 40% of revenues to 30% of the combined revenues during that period kind of negates that line of reasoning. Just take your calculator out and do the arithmetic.
Yep. Fair enough. I obviously wanted to touch on the Waves business because that's been very topical. You sound very confident in the $500 million run rate getting there by 2028. Obviously, the timing to get there has maybe been shifted back a little bit. Maybe you could just talk through the decision-making process you're seeing from your sales funnel today. It seems like maybe they're not as willing to take the waves as quickly as you were hoping. You've also had this dynamic of installing some waves that you're not billing for and trying to figure out that balance. As we think about Q4, is that how we should think about the ramp in waves, what you've installed in Q3 but haven't yet billed is what we should expect you to install? Is there any kind of line of reasoning there that we could explore?
Yeah. Let's step back. Sprint was not in the wavelength business. Cogent was not in the wavelength business. The Sprint network was not configured to deliver wavelengths. We started a program immediately upon closing, which was the primary reason for doing a transaction, of wave-enabling that network to connect it to our metro network and eventually serve 800 endpoint data centers. Today, we actually can deliver waves in over 1,000 North American data centers. We can deliver those waves at 10 gig, 100 gig, and 400 gig with any-to-any connectivity in less than 30 days. That is unprecedented in the industry. Now, we initially said that it would take us till the end of 2024 to hit the target of 800 data centers. We actually hit that target. We did not beat it, but we did not miss it.
We also had hoped that we could take a subset of 65 of the largest data centers and generate wave revenue more quickly from them. We hurried up and got those first 65 data centers enabled and w e generated $2 million a month of revenue, about 1,000 waves between those initial data centers in that period while we were still building out the remainder of the network. We had expected to do about 3x that, to probably do something more like $6 million. We are behind in that initial estimate. It was a new market for us, and the market was much more diffuse. It was not concentrated in a handful of data centers but spread out across the whole footprint. Since the beginning of the year and the first nine months, we have almost doubled that wave business off of the broader footprint that we have.
That rate of wave installs is growing. We've built a robust funnel. The customers have not accepted the waves in all cases as quickly as we have delivered them. That has created a backlog. We have waves installed that are not yet billing. We have not had any waves cancel prior to installation. We expect that our growth in the market will continue to accelerate as we demonstrate the five key competitive advantages of Cogent's network over its competitors. It's a five-pillar model. We have more endpoints. We have faster install. We have diverse routes. Those are things you can talk about before you've actually done anything. They're totally in your control. Customers may believe you, may not, but are probably going to test you to see how credible you are. We've actually installed waves now in over 500 of the 1,000 data centers.
Still means we have 500 data centers that are wave-enabled. We haven't sold a single wave or installed one in yet, but we will. There are two more important pillars to this strategy: price. Very different than our historic IP business where there's a commodity price across the market and you could offer a blanket discount. Waves are a bespoke product. They're based on the length, the contract term, and most importantly, the wave size: 10 gig, 100 gig, and 400 gig. We are undercutting the market today about 20%. We may have to become more aggressive. We have not chosen to be more aggressive yet because we have an adequate funnel of opportunities with that 20% discount level. If necessary, we will because we have a de minimis cost basis in this network, basically the $50 million we spent to connect the networks together.
Finally, and maybe the most important of all of the metrics is the reliability of the network. You can speak about reliability, but they're harder words until you actually demonstrate that. What we know is that the Sprint network is older than its competitors but better protected. It was built underneath railroad tracks, and it has had far fewer cuts per meter per year than any of our competitors. On average, our fiber gets cut 1/7th as frequently as the fiber we lease from others, including our competitors. I think we will prove the reliability, and those five factors will allow us to get to that $500 million number, which is 25% of the addressable market that we can serve.
To get to that $500 million eventually over the next few years, do you think you have the adequate sales team in place today? I know your sales force turnover ticked up a little bit in the quarter. Maybe you could touch on that. Just the sales process for Waves, how does that obviously differ from Transit? It's a new product like you talked about. It's probably a slightly longer decision cycle than some of your historical products. Has anything surprised you in that?
Cogent has three distinct sales forces that equate to about 650 quota-bearing reps and about 200 people in sales management and sales support roles. About 850 of our 1,800 employees are in sales. We have very high turnover in our corporate sales force. They are very activity-driven, all outbound telemarketing, and a very large addressable market but short decision cycles. If we're on net, we have a competitive advantage. If we're off net, our competitive advantage is much more diminished. Our enterprise sales force did not exist at Cogent before we acquired Sprint. They are much more farmers. They are very long-term, probably 20-year average tenure of those people. They manage accounts with very long sales cycles but are almost exclusively off-net.
88% of that enterprise business, after going through a concerted effort to bring it on net, is still off-net and will stay there based on its geographic footprint. It's only 15% of Cogent's total revenues. It's revenues that are probably not going to grow and will perpetually decline, but at a much more modest rate. Our wholesale sales force, which is roughly 300 of the 650, much more tenure, much lower turnover, very defined universe of about 15,000 net-centric customers, of which a few thousand are potential wave buyers. Those reps have a long relationship, have earned a lot of credibility in delivering IP services in more data centers, 1,900 data centers, delivering it faster and at lower prices. We are new to the wave market, but the reps selling it are not new to the customer.
The customers are giving us a portion of their wave business to bid on. We're actually winning a majority of the things we are bidding on. Over 50% are being converted. We also need to demonstrate that reliability and the price on a route-by-route basis. To get to that $500 million number, we're going to need to be able to see 90%-100% of the opportunities in our addressable market. Today, we're seeing less than 20%. We will get there. The sales force we have in a consultative process is the right one. For the wholesale or net-centric sales reps, the turnover rate is actually very low, sub 2% per month.
Do you have any sense for what you've installed so far, what is in your sales funnel? Do you think that's Waves business that's coming out of one of the incumbents? I know the switching costs are relatively low, just switching out cross-connects in data center. Do you think it's actually net new demand to the market? I think you've talked about the Waves market growing maybe 5%-10% a year. There's been a lot of new hyperscale and AI activity for transporting data from AI training hubs to inferencing nodes. Maybe you could talk about that dynamic a little too.
Yeah. When we initially contemplated the acquisition of Sprint and the idea of converting the network into a wave network, we took a pessimistic view and assumed the market would not grow. It was going to be like the transit market. Bit volume will grow, but price declines will keep the market flat. The implementation of AI training has grown the market. The market probably is growing at about 5% a year. The hyperscale segment of the market, which includes both training and traditional hyperscale content distribution applications, probably represents about 20% of the market. What will drive customers to make a buy decision is the fact that 55% of the market today is 10 gig waves, and less than 10% of new waves are 10 gig. Because there is no in-service path to upgrade, it does require a new wave. Our network can support all three speeds.
79% of our sales to date have been at the 100 gig level. The market today is between 4%-5% 400 gig. About 10% of our sales have been 400 gig, and we are only 11% of our sales at 10. I think we will capture existing customers. We will capture the need for route diversity. We will capture the market for connectivity to a broader set of data centers, and we will capture a disproportionate share of the growth in the market. This is exactly the playbook we have used in the past. Our 25% market share expectation by mid-2025 was calculated both on a top-down basis, looking at four independent market studies.
More importantly, we went to the 297 net-centric reps, asked them to go out to their deck of accounts and build an account-by-account forecast of wave opportunities over a five-year period, which got us to that number of about $500 million in revenue.
Just to confirm, can you talk at all about how many waves you have installed but are not yet billing? Just thinking about, I know over time you think that gap should materially shrink, but it still exists today. I think last quarter you said there were a few hundred, but I wasn't sure what that number was heading into the fourth quarter.
It was over 300 at the end of the quarter. That install-to-billing period is compressing as customers are getting more comfortable with the fact that we do what we say we're going to do, and the test results indicate the waves work. We have not given a number intra-quarter. All I will say is it's lower than it was at the beginning of the quarter.
Gotcha. Okay. I wanted to touch on some of the other, I don't want to call them hidden assets, but assets that are not as apparent by looking at your financials. The IPv4 addresses, you announced some more material wholesale leasing of those this quarter. As we think forward, you have, I believe, about 23 million unleased addresses left. Maybe you could talk about the path towards monetizing those, either through a lease arrangement on a wholesale basis where you can then securitize those cash flows versus outright sales. I think you've talked about how the large buyers in the IPv4 market are not that active. It's not a very liquid market to sell right now.
Yeah. I'll start with the sale market. We have chosen not to be a seller because market prices temporarily have come down, and the largest buyers are not actively purchasing. While there is a broad market at reasonable prices, that market is not deep enough to absorb the kind of volumes we would bring to market. That market maybe absorbs in total less than 1 million addresses a quarter. If you brought 23 million addresses to market, you would decimate the market. It just could not absorb them at even the reduced pricing level. Two, our leasing business has grown very rapidly. Five years ago, it was probably a $10 million annual run rate business. Today, it is almost a $70 million run rate business.
We have got there by increasing the number of leased addresses and raising prices on the addresses as the value of those addresses to the users have increased. We had been reluctant to lease addresses to brokers who would in turn sublease them. Because in many cases they reach markets that we do not have a direct relationship with, we felt comfortable last quarter reducing that restriction and allowing that and leased out a meaningful amount of addresses. There will be some additional of these large wholesale at a lower price. There is also a disadvantage to that in that we are now one step further removed from the end user, which is important for managing address abuse infractions. I think we have put in place safeguards to mitigate that damage and feel comfortable that both our direct retail growth will continue and some of these wholesale activities.
The final thing that we did with our addresses in Q3 was we ended up establishing an agreement with the other registry services to put RPKI in place on those addresses. That is a security protocol that is strongly recommended by the U.S. government. It is not yet required. That additional security feature will make all of our 38 million addresses more leasable and valuable. Yet we preserve our titular rights, which predated the establishment of any of those registries.
I also wanted to touch on the data center asset sales. You have an LOI for two of the assets of the, I believe, 24 that you've marked as non-core. Do you have any sense for kind of timing in terms of the technical due diligence that these buyers will have to do when we could convert that into a signed customer contract? As you look at the rest of your data center footprint, I know you've talked about evaluating either leasing it or potentially outright sales. Based on the activity you see from talking to counterparties, do you think you'll sell most of it? Will you lease more of it? What is your best guess today?
Let's maybe play the tape back. Prior to the acquisition of Sprint, Cogent had 54 data centers, 52 were in leaseholds, two were fee simple-owned, and there was 634,000 sq ft of data center space with 69 MW of power. When we acquired Sprint, we acquired 482 fee simple-owned telephone switch sites. We decided initially to convert 45 of them at no real capital cost into retail colocation facilities with basically 10,000 sq ft and 1 MW of AC power. After about a year and a half, it became obvious to us that the data center market was short of power and was a bigger opportunity. We increased the number of facilities that we converted from 45 to 125. We have now 186 data centers with 212 MW of power. Almost all of those are owned facilities, former Sprint facilities.
Of those, we identified 24 of the biggest facilities that we would not be able to fill up with our sales and distribution model. We then went to market with one of three models. You can buy the data center outright from us. You just own the building and the land for $10 million a MW. In these 24 facilities, there is 109 MW of existing inbound power. Second model would be you can lease them at $1 million a megawatt a year, triple net. And third, we would consider a JV if you made a capital contribution to us, and we worked in partnership to fill them up. You know, I think the majority, if not all of the 24, will eventually transact. We spent a full year from June of 2024 to June of 2025 converting these facilities.
We had a number of tours and a large amount of due diligence, but the facilities were still under construction. Since the beginning of the third quarter, these facilities are complete. The party that we entered into an LOI with basically had done its preliminary due diligence, demonstrated proof of funds, and negotiated a price that was acceptable to us for the two facilities. They are continuing their diligence. That LOI is being converted to a binding contract, and there are additional diligences. The most, I think, significant to the buyer is verification of power availability. Prior to our decision to convert these facilities, we went to the serving utility, and we said, "For a decade, these facilities have sat basically dormant. We see the transformers. We see the substations feeding them.
Can you verify to us that when we take the load, you'll be able to deliver it? We got that in writing from each of the serving utilities in these 24 facilities. At that point, we compiled the requirements of dozens of potential counterparties who had toured the facilities and made the decision to invest $100 million to bring them up to marketable standards. Some counterparties wanted us to do more. Some wanted us to do less. We were able to get that work done in a year, spend the $100 million, and now we're in negotiation with multiple other parties for multiple other facilities. I think the majority will sell. This is something we've never done before.
Does the emergence of AI inferencing edge use cases potentially make these assets more attractive? I know these are not big centralized facilities with very high power densities that are something that AI training would really be a great fit for. As we see more of these applications proliferate to the edge, does that potentially make them even more attractive in today's market than when you initially underwrote them?
I think it would. They're actually ideally suited for that. Sprint built these facilities for long-distance interconnection. They're typically on 6 or 7 acres, about 15 miles from downtown, along a train track and in an industrial park. They typically are a 40,000-50,000 sq ft building with dual substation feeds and about 5 MW of inbound power. There were full-voltage telephone switches that have been removed. The DC power plants have been converted to AC, and they have adequate connectivity due to our roadway business to all of the data centers in that metropolitan footprint. As a result, as inference moves to the edge, as large language models are developed to core and then replicated at the edge, these facilities are actually optimal for that use case.
One other potential hidden asset or hidden value within Cogent that you have not talked about as much recently was around dark fiber, right? You have this largely vacant Sprint network. We have seen some large dark fiber IRU deals announced with Lumen and Verizon and Zayo and others. Would you consider, in certain circumstances, IRUing out some of that fiber to one of the hyper-scalers where you could pull forward some cash flow through a big prepayment that could potentially help with deleveraging? Is that something that looks more interesting to you today, or is that something that is potentially more in the future?
It is something we have considered. We have done three of those deals to date. We are doing them on a limited basis. For the initial two and a half years, we were focused on getting the wavelength network optimized. We are open to doing those on a limited basis. What we are not open to doing is becoming a constructor of new routes. One of the issues with competitors selling dark fiber, which generates effectively an infinite IRR on an asset that is already in place, is that they are being required to do new de novo builds with very low IRRs. In our discussions with some of the hyperscalers, their request for new builds has dwarfed the opportunity in selling. We are open to it. There have been a few. We will probably do a few more. It is a valuable asset, and it's one that I think will unlock some value.
All right. Great. We have run out of time. As usual, I only got through about half my questions with Dave, so we will have to ask them later. Thank you, everyone, and appreciate you taking the time today, Dave.
Hey, thanks for hosting me, Eric. Hopefully my answer for you was.
Oh, no, that was great. Yeah. No, appreciate it.