All right, great. Thank you, everybody, for listening on the webcast. I'm really pleased to have Dave Schaeffer, CEO of Cogent, here with us today. We'll kind of go through a few questions and go from there. Dave, why don't we start out? Recent quarter had a couple of surprises. You got the dividend, the ramp to the wavelengths, I think was a little disappointing. Maybe walk us through a couple of those points. What should investors take away from those things and what happened and what from the messaging from the previous quarter?
So first of all, thanks for hosting me, Frank, and thanks, Raymond James, for a great venue. And thank all the investors for taking time to hear a little bit about Cogent. You know, I think in looking at this most recent quarter, there were three big events. The first one being the reduction in the dividend. And we had grown our dividend sequentially for 52 consecutive quarters. Because of the acquisition of the Sprint Wireline business, the negative revenue growth and cash burn of that business, we were paid $700 million by T-Mobile. Those payments were front-end loaded and actually reduced our leverage initially. But after the first year, those payments stepped down. And on an LTM basis, our leverage increased to 6.6 times levered. And we felt uncomfortable with that level of aggregate leverage.
For that reason, we chose to reduce the dividend by 98% to $0.02 per share and keep it at that lower level until such time as we reach four times net leverage. We also temporarily paused our buyback program just so we weren't sending the signal that we were taking every dollar of dividend and using it for buyback. We still have $105 million available under that buyback program. So that reduction in return of capital was event one. Event number two, we have grown our EBITDA for the past nine quarters post-acquiring Sprint by reducing costs faster than we reduced revenue. Even though our top-line revenue has declined at an average of 2.4% over that nine-quarter period, an annualized rate, we've been able to grow EBITDA over that period for nine consecutive quarters.
We've been able to grow the EBITDA margin from about a 1% margin the first quarter after the acquisition back to a 20% margin without adding back the subsidy payment from T-Mobile, which brings margins to about 30%, still below the historical rate for Cogent pre-acquisition, where our EBITDA margins were about 40.5% in the quarter immediately preceding the acquisition. So I think that ability to continue to grow our EBITDA with moderating capital expenditures, our CapEx fell sequentially by $20 million and is returning to the kind of stabilized rate of about $100 million of CapEx coupled with $40 million in principal payments on capital leases annually. And then I think the third thing that investors focused on in the quarter is the growth rate in wavelength revenue. Wavelength revenue represents about 4% of Cogent's total revenues. Those revenues grew 93% year- over- year, 14% sequentially.
While those are respectful numbers, they're off of a small base, and I think investors were looking for a faster pace of wavelength growth.
All right. Well, why don't we talk a little bit about, so a couple of things. Part of the rationale on the dividend side is deleveraging. What are some things you can do to delever? And in particular, selling any other assets like the IPv4 leases, things like that. How should we think about that?
Yeah. So there are two paths to us improving our leverage profile. The first, most obvious and most effective of those paths is growing EBITDA.
The old-fashioned way.
We've done that now for nine quarters with negative top-line trajectory. We are now inflecting to positive revenue growth. And while we still have cost-cutting opportunities to help grow EBITDA, we will also get the high contribution margin from those incremental on-net sales, whether it be IPv4 addresses, wavelengths, or our core on-net internet and VPN services. Secondly, we have surplus assets that we could divest of. That's, I think, really the heart of your question. We identified a total of 24 data centers that comprise 1 million sq ft and 109 MW of inbound power as surplus. We have a total inventory of 186 data centers, 213 MW of power in about 2.1 million sq ft. Those largest facilities sit empty. If we are able to sell them, we benefit two ways. One, we get the proceeds to use to deliver.
Two, there is a carry cost associated with those facilities that would be eliminated. We have targeted $10 million a megawatt. So 109 MW would be $1 billion, $90 million at risk. There is about a $20 million annual cost to carrying those empty facilities that have been converted. It took us over a year to remove the old telephone switchgear and convert those - 48 volt DC facilities into AC 120. We announced a letter of intent for two of them for $144 million. We have a number of other LOIs in negotiation. The second thing that we can look to monetize is excess fiber in our network. We have done three small dark fiber IRUs. We will continue to evaluate those. When we acquired Sprint, we acquired 19,000 route miles of intercity fiber that is owned and 1,200 route miles of metropolitan fiber.
We typically are using anywhere from two to eight strands of that for our own purpose. The routes themselves have anywhere from 24 to 144 strands available. So there's substantial excess dark fiber, and we will continue to look at ways to monetize that. And then the final point that we have excess inventory of is IP address space. We are the third largest owner of address space in the world with about 38 million IPv4 addresses out of a total universe of 4.3 billion. 800 million are owned by the U.S. government, about 132 million are owned by Amazon, and then ourselves, and then some other substantial players. We have leased out a significant portion of those addresses. Today, we have about 15 million addresses leased.
We have about another 1.5 million addresses that we have allowed customers to use free of charge, kind of grandfathered in for very old customers before we started leasing out addresses. That means we still have a surplus inventory of over 22 million addresses available to monetize. We are continuing to lease them out. We have seen the leasing income in that business go from about $12 million a year in 2022 to a run rate exiting 2025 of over $70 million. We also would consider selling addresses, but since the two largest buyers of those addresses, Amazon and Microsoft, have not been buying in the market, I don't think that we would be a likely seller until we saw prices rebound. What we did do in the most recent quarter is, for the first time, entered into a wholesale agreement in Q3.
We've done a second one in Q2 where we've done large blocks of addresses at a discount via lease to third-party resellers who then, in turn, sublease those out.
And so what is the opportunity there? I mean, is there more opportunity on the pricing side? What is the advantage of that wholesale agreement versus going out and just marketing it? Do you think you could sell all 22 million that you have left if you wanted to, or would that break the market?
Sale and lease are two different things, Frank. So we could definitely sell the inventory we have and raise cash. I think that would be a non-optimal approach.
I meant either besides the wholesaler, you lease them yourselves, like keep on your own book. Could you lease those other 22 million yourself?
We could, but I think our pace of leasing has been at a couple of million addresses a year since we began leasing. So at that current pace, it would take us nearly a decade to lease out the addresses. If we decided to go with some of these wholesale opportunities, they tend to come in much chunkier, bigger blocks of addresses, and we could get there much more quickly, but at a lower price per address.
What's the discount that you get on the wholesale side versus retail?
It's roughly half of what we would charge if we were providing that address to an end user.
But then, do you have better visibility on a timeframe for getting that revenue? And I guess it's also, what is sort of the, if you were to go out and try and accelerate, you're doing $2 million a year. If you marketed that and tried to ramp that up, could you do more? And what is the, how thick is the market for that? Would that break the market?
So our distribution method is through direct sales. The Amazon, Microsoft, Cox, and Verizon models, which are competitors, are directly to retail end users through their website. So a totally different way to go to market. 84% of our addresses go to other service providers. 14.5% go to corporate end users, and a little over 1% of addresses go to large enterprise customers. We don't have a salesforce that can reach very small retail users. So we either need to continue to go through other service providers or through wholesalers. The decision to do both was something that we're experimenting with and trying this quarter. Our original decision to lease addresses goes back to 2015. Between 2015 and 2022, we only leased addresses to companies that also purchased bandwidth, and it took us seven years to grow to a $12 million run rate.
Once we relaxed that restriction and were willing to lease addresses to anyone, whether they bought bandwidth from us or not, the volume and dollar amount of addresses accelerated meaningfully, and we've gone from a $1 million a month run rate to a $5.5 million run rate in basically three years. I think we can continue that and then augment that with these wholesale opportunities.
Right. Okay. All right. And then on the data center side, you said you're targeting kind of $10 million per megawatt. We haven't seen a whole lot of activity there. Are you open to selling that for lower amounts? I mean, why is $10 million the right number?
So we have never divested or sold off a data center. We looked at the market, and we observed two things. One, an average sale price of about $17 million a megawatt for newly constructed data centers. And then secondly, wholesale leasing activity at about $1.4 million a megawatt a year on a triple net basis. These are old switch sites. They are distributed, which has value, but they are small, which makes them less desirable. We picked $10 million as a reasonable go-to-market price. We have seen some parties interested at full price, some at a discount. Ultimately, the market will determine where these will clear.
Okay. Great. And so give us an update on the wavelength business. You gave some guidance on kind of where the run rate would be for the quarter. How do you feel about that target today, and how should we think about the current trends in the run rates for waves?
I just want to remind investors, Cogent generally does not give quarterly or annual guidance, but rather multi-year metrics. We have said that our total revenues will grow 6%-8% a year with about 200 basis points of margin expansion. Specifically with wavelengths, we said and reiterate that we will be at an annual run rate of about $500 million by mid-year 2028. Effectively a little less than three years from now. When asked about that ramp, we said that we would probably exit fourth quarter at a December times three rate of about $20 million. We could hit that number, but unlikely will not based on the fact that some of the wavelengths that we have sold and installed have not yet started billing for the customer because the customers have not accepted the waves as quickly as we have installed them.
Okay. Great. And then what are customers taking? I mean, some of the checks we've looked at indicate a movement towards 400 gig waves. I think most of yours are hundreds, which is probably where most of the market is. But as far as new customers and things, how are you positioned for that? And what's sort of the mix we should think about, say, starting in 2026 of new waves coming on of hundreds versus 400 or higher?
So the embedded market today, which is about 37,000 wavelengths and about a total of 140,000 wavelengths, the majority of those are today 10 gig waves. That represents 55% of the market by revenue. Roughly 40% of the market is 100 gig, and less than 6% of the market is 400 today. That's the installed base. At Cogent, since we've been selling, we have seen 79% of sales be 100 gig, so nearly double where the market is. And we have seen the 400 gig take rate at around 10% and about 11% for the 10 gig wave. So our sales are skewed towards the higher wavelength side. You had mentioned 400 and higher. Today, there is no commercially deployed wavelengths above 400 gig because of the lack of pluggable optics. The equipment that we have installed is capable of 800, 1.6, and 3.2 terabits per wave.
The term wave is somewhat of a misnomer because the customer is not buying a full wavelength. They're buying a private line inside of a wave. The term of art wave came about when the private line business shifted from fixed microwave to optics, and initially, you would buy a 2.5 gig wave or a 10 gig wave. As wave sizes got bigger, what ended up happening is as waves became fractionalized, much as our 3.2 terabit wave is broken down into either 100 or 400 gig increments. Now, while 3.2 terabits is technologically possible, in the real world, when you look at span lengths, number of jumpers, and number of splices, a 3.2 terabit system will actually deliver about 2 terabits without any intermediate OEO conversion on a transcontinental basis.
If you add that incremental set of transponders to do that OEO conversion, you can't get the full line rate, but then it actually raises your cost per bit mile to deliver.
All right. Great. And then with that, how should we think about ARPU on waves going forward? You've been steady in the 2000-ish kind of range. As the mix changes, should we think about that as creeping up? And where are you on the market for 400 gig waves versus kind of where your competitors are?
I'll start with the competitive point first. We are generally priced at about a 20% discount to the market. A wave is priced based on three inputs: the size of the wave, 10, 100, and 400. The physical length of the path. The longer the path, the more you pay for it, and third, the longer the contract you sign, the lower the price of the wave. Our ARPUs have hovered around 2,000-2,100. You know, I think it varies. If you're on a long-term contract on a short path, 10 gig wave, you're down around $500. If you are on a 400 gig wave that is transcontinental, that number is closer to $7,000-$8,000. I would say the $2,100 average is for a 100 gig wave on a three-year contract with a relatively long path, probably 2,000 miles or so.
Okay. All right. Great. So let's talk a little bit about the legacy business. Let's see that. Let's talk about NetCentric. How has that been trending, and what's the outlook for traffic growth currently?
So, again, just to remind investors, three different drivers of business. 70% of Cogent's revenue comes from the classic Cogent business, the legacy business that Frank described. About 30% of revenues come from the legacy Sprint business that we acquired, which was predominantly off-net and a combination of MPLS and DIA. In the Cogent portion of the business, the revenue split was roughly 75% on-net, 25% off-net. Breaking that down by customer type, the historical Cogent business was roughly 40% of revenues were NetCentric, and about 60% of revenues were corporate end users. Within the NetCentric business, 90% of that business is on-net, and 10% of that business is off-net. The NetCentric business and revenue is growing at about 8% today, year- over- year. That's in line with historical averages. There has actually been a slight acceleration in the past two years in Cogent's legacy business.
So if you looked at Cogent pre-acquisition, we were growing at about 5%. We acquired a business that was declining at 10.9%. We accelerated that revenue decline of the acquired business to - 24.2% over the subsequent nine quarters, but our reported top line rate of decline was only 2.4% annually. The core Cogent business actually accelerated in part because of an improvement in the corporate segment and a stable but above trend line rate in NetCentric.
All right. So let's switch to the corporate business. So talk to us about what the trends are there and how has that recovery gone and what are some of the drivers of that uptick?
So there's both corporate legacy Cogent, which was, again, a split within that segment of 60% on-net and 40% off-net. That business is now growing at between 3% and 4% year- over- year. Not as well as it had grown pre-pandemic at about 11%, but much better than it performed at the trough of the pandemic where revenue growth was - 9%. The acquired Sprint business was divided into corporate and enterprise. Those two segments have declined at 24% year- over- year. It is virtually all off-net. So the off-net corporate business is dominated by the Sprint portion of that business, and that is the primary reason for the revenue decline. We expect the corporate on-net business to continue to grow in those low single digits.
We are through the majority of the forced decline in the acquired Sprint corporate business and should see off-net corporate slightly declining to flat going forward and eventually returning to very modest growth, but the entirety of the acquired Sprint base, which is corporate and enterprise, is still declining at about 2.5% a year.
Okay. Great. And just to kind of finish up, maybe pick two things you think that are the most misunderstood by investors, and what do you think the investors should understand better about those aspects of the business?
So I think the first thing is just the complexity of the accounting associated with the transaction. It's a highly unusual transaction when you take an asset that costs $20 billion to construct and are paid $700 million to take it. And I think the way in which that is accounted for has confused investors, has perhaps had them think our leverage was worse than what it in fact is. And then as a corollary to that part of it, they have had a hard time disaggregating the revenue growth between the acquired customer base and the organically growing customer base. So I think those levels of misunderstanding have hurt Cogent's end user business and its NetCentric business. And then on the wavelength side, that is a brand new business for Cogent. It never existed at Sprint. It didn't exist at Cogent prior to us repurposing the Sprint network.
And I think investors may have misunderstood how long and how difficult it would be to repurpose that network and what that growth rate would be. I think we came in right on time in terms of our schedule to repurpose, and I believe we have achieved meaningful revenue growth, but maybe not as quickly as we'd like, in part because that wavelength market is much broader in terms of endpoints than just core data centers. You yourself wrote a piece on that, talking about how broad that market is.
All right. Great. Dave, thank you very much for the time. Really appreciate it. Thanks, everybody, for being here.
Hey, thank you. Thanks, Frank, for hosting me.