Ready to get started? Great. Awesome. Good afternoon, everyone. My name is Sam Mayer. I am an associate on the telecom research group here at TD Cowen. I am very pleased to be here today with Dave Schaeffer, founder and CEO of Cogent Communications. Dave, thanks so much for joining us today.
Hey, Sam. Thanks for hosting me. As always, I want to thank TD Cowen for a great venue. It gives me an excuse to get to Boulder, and I would like to thank the investors who hung around late in the day to hear what we have to say.
Yeah, for sure. On the 2Q print, had an EBITDA miss, but partly on one-time severance cost. Wave of softness, but was largely baked into estimates. Noted constructive commentary on potential data center sales, a pending refinancing, CapEx coming down, yet the stock still sold off again. We are curious to get color from you on what you think the main factors driving the stock sell-off are.
Yeah, I think there is significant investor concern around the fact that our $750 million of unsecured debt went current in the second quarter, and we are taking steps to refinance that debt as secured debt. We will most likely shrink the amount that we are looking to raise.
Yeah.
Probably doing a shorter tenure to give us a little bit more call flexibility. I think the number one concern was around that debt maturity. Secondly, our top line continued to decline. For 18 consecutive years, when Cogent was organically growing as a standalone company, it had average annual growth rate of 10.2% per year. We acquired Sprint, which three years prior to our acquisition, had a revenue decline rate of 10.9% a year. Sprint represented 42% of the combined company's revenue, Cogent 58%. With that large of a revenue base declining, our revenue growth for the past 12 quarters has been negative. A lot of that negative growth is by design because the Sprint revenues were generating negative EBITDA margins of approximately 60%, $300 million of negative EBITDA on a $485 million run rate. We worked diligently on purging unprofitable products.
That meant killing non-core services, moving customers where possible on net, and migrating access services to fiber rather than fixed wireless or coax or twisted pair. Finally, we purged services outside of the 58 countries in which we are licensed. As a result, the revenue base of Sprint accelerated its rate of decline. Today, that revenue base represents only 15% of the combined company's revenue, and that drag on top line has, I think, discouraged growth investors. During that same period, our EBITDA grew sequentially and year-over-year 11 of the 12 quarters, and our EBITDA margins expanded by over 2,000 basis points on a combined basis. Most of that improvement came through cost cutting. A small amount of that improvement came through the growth in on-net versus off-net.
We report a higher number of EBITDA because we include the subsidy payments that we receive from T-Mobile. Last year, the underlying EBITDA was $192 million, the reported number, $292 million. Investors are concerned because those payments end in February of 2028, so a little less than two years from now. We probably have about another $50 million of run rate of cost savings to take out. That is about $8 million in the underlying cost basis, $7 million of cost reduction due to the sale of our data center footprint to I Squared, or at least a portion of it, 10 data centers and 54 MW, and then finally, from our reduction in headcount related to integration expenses. Just to remind investors, when we announced the Sprint transaction, we outlined that we were going to be spending about $60 million a year on integration efforts.
Almost all of that spending were internal resources, and that spending would be complete by the end of 2026. We had reduced that number down to a $36 million run rate going into the second quarter. We ended up reducing headcount by 6%, and that additional headcount reduction had us exiting the quarter at about a million-dollar-a-month run rate. That number will go to zero by year-end. We go into next year with nearly $50 million of EBITDA tailwind.
Great. Awesome. I want to touch on CapEx. CapEx came in at $38.5 million in 2Q, coming down sequentially, and you've guided to further expectations for declines in 3Q and 4Q. With the expectation for CapEx to moderate going into 3Q, why is this happening? Have your vendors taken down price? Are you simply ordering fewer routers? Do you think we could think about $30 million-$35 million as the quarterly range to look at CapEx going forward?
We actually hope our long-term CapEx is below that number. Prior to acquiring Sprint, our CapEx was running at around $100 million a year. Sprint's was about $30 million. With the synergies that we achieved, the network consolidation, we thought we would get down to a run rate of about $100 million a year, and we would also be spending about $40 million a year on principal payments on capital leases. The Cogent IP network, which spans 94,000 route miles of inner-city fiber, 33,000 route miles of metro fiber in 308 markets, 58 countries, is the largest IP network in the world, built entirely on IRUs. We expect that principal payment number to remain constant. On the CapEx side, our CapEx did come down sequentially by $18 million. We had a surge in capital spending for three reasons. One, we ended up converting 125 telephone locations into data centers.
That ended up being a wise investment because we sold 10 of those facilities for $225 million in cash. We had a GAAP gain of $130.4 million. We actually had a tax gain of $224.1 million, and we're fortunate that we had sufficient NOLs to shelter that gain. We still operate over 170 data centers, and we have 24 of those remaining data centers, or excuse me, 14 remaining of the 24 that we had earmarked for sale. The second reason CapEx surged is we had to reconfigure the Sprint network. The Sprint network was a TDM voice network. We repurposed that asset into a wavelength network. We extended that network into 1,137 carrier-neutral data centers. This required a significant capital investment. Then third, we have been hit actually by equipment price increases, which is truly unprecedented in technology.
Whether it be in routing or transport or servers, customers expect prices to decline following Moore's Law, which represents about a 55%-per-year price performance improvement. What we have seen in the past 18 months is an unprecedented spade of price increases. We buy equipment from three vendors. Cisco is our largest, Ciena our second largest, and Arista our third. We have had six price increases in calendar year 2026 from Cisco. We've had three from Ciena and two from Arista. This is unusual. We've also seen equipment availability stretch. Normally, equipment would be ordered and shipped within 90 days. Now we're being quoted delivery times for much of this gear being 18 to 24 months. This has resulted in an uplift in our capital spending. We did experience $18 million of savings on a sequential basis.
We expect that number to come down, but I think the $25 million a quarter, which we still think is a long-term rate, probably needs to wait until some of these supply chain issues get resolved.
Do you anticipate any more price raises coming from equipment vendors, or where it sits now, do you think it might have plateaued?
I wish I could answer that question. In talking to management teams at all of the vendors, they are today bemoaning supply chain constraints for components, particularly memory, and a lack of contract manufacturing capacity. We have seen a shift in behavior. Cisco had historically been an enterprise-focused business with only about 20% of its aggregate revenues coming from service providers. Ciena was very different. Ciena was a 100% service provider company. What we have seen over the last year and a half is hyperscalers coming in and putting massive POs in place. Today, at Ciena, nearly 60% of revenues come from four hyperscalers, where that was virtually zero 18 months ago. Those single POs, in many cases, are over $1 billion. Unprecedented. At Cisco, the majority of their service provider products are today being sold to hyperscalers.
Those companies have spent $1 trillion, and they have announced nearly $6 trillion on a capital roadmap over the next four years. The good news is capitalism solves supply constraints. All of the vendors are ramping up their input capabilities, their manufacturing capabilities. While I don't want to call it transitory, I don't think this shortage will be permanent, and we will see a reversion back to Moore's Law economics in telecom equipment.
Yeah, for sure. Staying on CapEx for one more point. Your on-net building count came down again this quarter. Is that driving any of the CapEx declines? Can you help us understand why the building count just doesn't stay flat and leave those buildings on net?
Our total on-net building count increased, where virtually all of those increases were on the carrier-neutral data center footprint, not the multi-tenant office building footprint. Cogent's corporate business grew for 18 years at overall 11%. We were growing that footprint pre-pandemic at about 3.5% a year. Today, we have about 1 billion, 40 million square feet of multi-tenant office space and another 13 million square feet of single-tenant end-user buildings. When the pandemic hit, we saw vacancies in our footprint expand from a historical run rate of around 6% to nearly 21%. While we have seen a moderate improvement to about 19.8% in our footprint, that is still far from the 6% we had pre-pandemic. Secondly, we've seen the number of employee entrances per day in that footprint decline to zero in the pandemic, and it has reverted back to about 65% of pre-pandemic levels.
As a result, we have continued to grow our corporate on-net revenues, but at a more moderate 3% growth rate. We have seen a number of buildings be converted from office to residential. There was one of the buildings in New York that got quite a bit of press attention a few weeks ago, as that conversion was not done well and the building became unstable, and I believe that building was in two blocks of your headquarters office—
That's right.
...which probably made access to the office a bit challenging.
Yeah.
We have been extremely disciplined about what types of buildings we connect to. Cogent's end-user business is predicated on two unfair competitive advantages: using fiber from third parties to get as close as possible to the building, minimizing our actual construction, and two, cream skimming the market, picking only the largest buildings. The average building is about 550,000 ft, 41 stories in height. We, for example, serve your building near Vanderbilt and 42nd and have dozens of tenants, from law firms to financial services. Yet, if you looked across the street, you might see a small single-tenant building. Even though we have fiber passing by it does not have a large enough addressable market to give us an adequate return on capital. The fiber overbuilders for residential have a very different model. Cogent's model has always been corporate dense locations or carrier-neutral data centers.
Because there are a few new speculative multi-tenant office buildings being built, we're not expanding the footprint. In fact, because of these residential conversions, we're actually exiting certain buildings.
Yeah, great. Want to pivot to wavelengths. Wavelength revenue was $14.8 million in 2Q, just a touch below estimates. You've noted in the past two primary reasons for the Wave slowing. Clients aren't ready to take orders, and that Cogent is still new to the game. But during the quarter, Lumen had noted their strategic wave sales were up 35% year-over-year on a much larger base. Can you give us any sort of color to help reconcile the strength elsewhere with the delays that Cogent's seeing?
Yeah. So on wavelength sales, Cogent is a new player in that market. It's a mature market that has been around for more than 75 years. They used to be called private lines. They just got a fancier name called wavelengths today. The market used to be dominated by AT&T, and then MCI entered. The services were delivered over fixed microwave and were transitioned to fiber in the early 2000s. We had the opportunity to acquire Sprint's operating business, which I touched on, but also acquire the Sprint voice network and repurpose it as a wavelength network. We had a great deal of credibility as an IP provider. We're, in fact, the largest provider globally. We had an existing sales force. We had relationships with many of the customers, but they had no frame of reference for us selling wavelengths.
Initially, we had to prove our value proposition, starting to deliver those wavelengths, getting customers comfortable with that. I think we are over that hurdle in the fact that in the last quarter, we reported that we had sold wavelengths into 608 of our 1,137 wave-enabled locations, and we have 548 unique wavelength customers. Those customers today buy about 2,500 waves. We added 188 waves sequentially in the quarter. Revenues grew 9.2% sequentially, 68% on a year-over-year basis. But maybe the most important stat that we reported was that 77 of the existing wavelengths upgraded in a quarter, either from 10 - 100 or 100 - 400. That progression is demonstrating the customers' confidence in our wavelength business.
Our major wavelength competitors anecdotally speak about the strength in the market, but what they don't do is give you the granularity to see the growth in that particular product suite and measure that growth by units and ARPU. We have been extremely transparent. Yes, our wavelength business we had hoped would grow faster. We have gone in six quarters from 0% market share to 3% of the aggregate North American intercity market. We have an ultimate target of getting to 25% market share. We feel we will get there, probably a little slower than we had originally hoped, but we will get there.
It's a good segue from the 3% market share to the 25% long-term target. Even if we view the longer-term target as pushed out a few years, it seems like it still implies some sort of J-curve growth. Curious on what your sense of timing for that is and what you would need to see for an inflection to happen.
I think our rate of growth will accelerate for three distinct reasons. One, customers now view us as a credible supplier equal to other suppliers. Two, customers will no longer have these artificial constraints caused by either supply chain or power constraints to use wavelengths. Then three, I think the entire market will grow as we see many of these new AI training facilities actually come online. Much of the capital that has been deployed to date is not yet fully operational.
Right. And you've mentioned before also, customers have been delaying orders. We've been speaking about that for over a year now. To that point, do you still have orders in the backlog for a year ago, or have those customers ever walked away from those orders or have gone to a competitor? Some of those orders for over a year ago; what's happened to them since?
While we have had a modest number of cancellations, the majority of the orders in our backlog are installing. It is how we've gone from zero to 2,500. We also do not yet have enough empirical data to model out the monthly conversion rate. We used our IP business and its backlog and conversion rate as a proxy, but what we have seen is that the wave market is different. We are seeing strong demand. The demand comes from five discrete types of customers: regional networks for interconnectivity, content distributors to push our content closer to the edge, international carriers to extend their networks across the terrestrial U.S., and then two relatively new applications. One, hyperscale-based AI training, and second, neoclouds. We have sold to each of the major hyperscalers multiple wavelength orders, and we have also sold to all of the neoclouds.
Your colleague actually reported on one of our sales to one of those hyperscalers, and I spent a day on the phone calming that management team down, telling them that the leak did not come from us, but rather through independent channel checks.
On the data center sales, you continue to market the 14 remaining facilities. You had mentioned on the call, received LOIs for four sites, but would not expect a sale until 2027, expecting better tax treatment. Could you help us understand the tax NOL piece and the mechanics of it? Could you possibly sell those four facilities in 2026 and then bring the rest in 2027 for the better NOL treatment? Any color there would be super helpful.
We were very fortunate in that we had a $224 million gain before tax, a $130.4 million gain for book from the 10 data centers that we sold to I Squared. Those 10 data centers comprise 54 MW of power. The remaining footprint that is for sale is 14 data centers and 55 MW. At least one of those data centers has significant incremental power above and beyond what is at site today, and we think that warrants a premium. The current term sheets are from credible parties. They have not been countersigned by Cogent, as we believe we can negotiate those prices to something more appropriate. For the remaining 10 in which we do not have LOIs, we are continuing to conduct tours and talk to a number of interested parties.
Our preference would be to have a transaction that may be announced in the latter part of 2026, actually close in early 2027. At that point, the annual limitation on our NOL usability elapses. We effectively reset and would have the ability to shelter that gain. Since the primary objective is to use those proceeds to delever, the less we spend in taxes, the more we will have to either reduce our total debt load or buy debt at a discount.
Yeah, for sure. With the last few minutes, I wanted to touch on questions to see the longer-term vision of Cogent. As we look out the stock, what kind of events will the company have to lap before Cogent can reach its longer-term 6%-8% top-line growth and 200 basis points of margin expansion per annum? Once the wavelengths business has seen a meaningful ramp, what would you expect the next leg of growth ahead for Cogent would be?
Yeah. Cogent grew at 10.2% with 220 basis points pre-Sprint. When we acquired Sprint, we have had now 12 quarters of negative growth. The Sprint-acquired revenues are now only 15% of the base. In the last quarter, our revenue decline was $3.6 million, but $4.5 million of that came from off-net services, predominantly Sprint services. Because that base has become a small enough part of the total, we should be able to let the Cogent growth rate show through. Cogent's growth rate over the past three years cumulatively was 29%. Now, that did include wavelengths, which are growing 65% off of a small base; IPv4, which is also a smaller base growing at 18%, and our corporate business, which is growing at about 3%, and our on-net NetCentric IP business growing at about 7% or 8%.
I think we have many years of growth in each of these products ahead of us. I think the burn-off of the Sprint revenue is almost complete. We actually thought it was complete a couple of quarters ago when we had two successive months of positive revenue growth, only to have a large enterprise customer churn a meaningful number of locations. But with that, we will revert back to not a 10% growth rate because of that remnant Sprint business, but a 6%-8% grower. In EBITDA margin, we have delivered roughly 2,000 basis points of margin improvement in 12 quarters. That is an abnormally fast rate of margin expansion. That margin expansion will moderate to that roughly 200 basis points a year. And if we look at Cogent pre-Sprint, our EBITDA margins were 40.5% in the last quarter before the acquisition. They fell all the way to 1%.
They are now back to 20% without counting the transfer payment from T-Mobile. With that, they are 30%. We envision those margins as those transfer payments roll off, continuing to grow and plateau in the mid-40s at roughly that 200 basis points a year. So our EBITDA growth will be substantially better than our revenue growth for the foreseeable future. And couple that with the moderation in CapEx, I think we will rapidly delever and be in a position to again re-accelerate the return of capital to equity.
Great. With that, looks like we're just about out of time. Dave, thanks so much for joining us again.
Hey, Sam. Thanks for hosting me.
Of course.
Thank you all for hanging around.