Okay. Good morning, everybody. Welcome to the Goldman Sachs Communicopia and Technology Conference. I'm Jim Schneider, the telecom analyst here at Goldman Sachs. It's my pleasure to introduce Dave Schaeffer, CEO of Cogent Communications. Thank you for being with us today, Dave.
Hey, thank you, Jim, for hosting me. I'd like to thank all the investors for their time and Goldman for the opportunity.
Excellent. I wanna start, I think, with a topic that I believe is near and dear to your heart, that is wavelengths and that business. You know, one of your competitors has had some news out recently related to AI and the use of dark fiber for transport between data center properties, so that caused a lot of buzz, I think, in the sector the last couple of weeks. You know, we've also heard you talk about your ability to provide dark fiber to some of these same customers, and when you evaluated some of these opportunities, you found them to be, I think you said subpar returns.
Maybe talk about just zoom out for us, for a second, and help frame for us the size of the wavelengths market you see, both domestically and internationally.
Yeah. So the wavelength market, which is optical transport with a fixed latency and no dependence on, packet size, is about $7 billion globally. About $3.5 billion of that is in North America. About $1.5 billion of the $3.5 billion is metropolitan. That means within one general market, and about $2 billion is long haul or inner city. The North American market is about 130-140 thousand aggregate wavelengths today.
Mm-hmm. And how concentrated do you think is the customer base for that, for that market, as you, as you see it? And can you maybe say something about how many customers, in your view, are sort of really in a position to actually be able to handle and manage dark fiber capabilities?
I'll directly answer that question, but let's kind of start. The cheapest and easiest and most flexible way to move bits from point A to point B is the public internet, period. It is cheaper than wavelengths, it is cheaper than dark fiber, it's cheaper than constructing your own fiber. If you move to a wavelength product, on average, you will pay about two and a half times as much per bit mile as you would over the public internet. If you move to a dark fiber product, you will pay even more. The reason I can't quantify what more means is it depends on what percentage of the wavelengths on that dark fiber you actually like. The absolute most expensive way to move bits would be to build your own fiber between two points.
So in terms of the market, there's a couple of hundred customers that value these attributes in scale that buy wavelengths. They tend to fall into two primary categories. Companies who are looking to extend their network. That may be to link islands of traffic together, a cable company, an international carrier that's looking to tie markets together, and then the other are content generators who are looking to move content around. And that could be for one of two very different purposes. One could be to distribute content to more edge locations. The other could be to move content in order to facilitate AI training. So the limitation today is space and, more importantly, power for AI training, and in many cases, the data is not co-resident with where the space and power is available.
So companies are purchasing wavelengths or dark fiber in order to connect the content to where the training is occurring. And in building new fiber, many of these sites are in locations where there is no fiber at all. So if you are a hyperscaler building a new data center campus from scratch, you actually have a pretty specific model you would like to follow. Now, in the real world, you may not get everything you want, but in the ideal world, you would get about 2,000 acres, you would have about a gigawatt of power, you would have some water available on site for cooling, you would have relatively inexpensive power, and then finally, in the ideal case, you would get an IRB bond to help you facilitate that construction.
That would be kind of the wish list you would start with, and then you'll end up taking what the market gives you. Now, those new campuses have no fiber connectivity, so what those hyperscalers are looking for is for companies like Cogent and others to build out to those facilities. And they typically fund about half of the construction cost and expect you to fund about half of it. And they typically have sufficient market power that they can drive the returns down to low single digits, and that's why we have found these unattractive.
Mm-hmm. And maybe, you know, how many customers are able to handle the dark fiber kind of deals, you think?
Yeah, I mean, I think for these very large deals, it's probably less than 10.
Mm-hmm.
... for smaller specific routes, where maybe someone just has one particular requirement, a sub-sea landing station, maybe back to a major metropolitan market, there may be a market of, you know, 50-75 companies that are actively looking for dark fiber. And then there's always been a government and research dark fiber market. And for a number of years, some of the current suppliers have only focused on that market. You know, we have been on the other side of that market. We are probably the most prolific buyer of fiber in the world. So in the core Cogent business model, we didn't build any fiber, we didn't own any fiber. We bought from others, and prior to the acquisition of Sprint, we had bought from 328 different suppliers.
Since we've acquired Sprint, even though we're an owner, we've actually expanded that universe, and last quarter, we reported 356 different suppliers. And the reason is, we were in about 110 markets where we needed to connect the Sprint network to our metropolitan network, and we would always do that with as much purchased fiber as we could get to get as close to the route that we wanted as we could, and then we would build just what we had to build.
From a competitor standpoint, who do you see as your main competitors in this part of the market, besides, let's say Lumen and Zayo?
You know, I think they are the main competitors on a nationwide basis. There are clearly regional players, Windstream and Uniti, more in the Southeast. Crown Castle, maybe a little more in the Northeast and in Florida because of their FPL acquisition. FiberLight in kind of New York, or excuse me, FirstLight. FiberLight is more in the Southwest Texas region, FirstLight more in New York State and, Maine and New Hampshire. So it's more regionalized.
Mm-hmm. In terms of your competitive advantage, I think you've said you want speed of provisioning to be one of your advantages. You're going from, what? 90 days today to a matter of weeks in some amount of time. And give a sense of, like, once you get there, how much time advantage in terms of provisioning does that give you relative to your competition? And then maybe speak to any other points of competitive differentiation, like unique routes or otherwise, that you think will give you a leg up in the market.
Yeah. So as we thought about the Sprint network that we acquired, we thought that the highest and best use was to take that network and use it for wavelength services. To do that, we had a number of foundational steps we had to complete. We had to physically connect the network into our metro networks. We had to reconfigure our metro networks to support wavelengths in addition to IP services. We needed to put transponder shelves at the endpoints in those data centers to accept the wavelengths, and then we needed to deploy ROADMs at the intersection of the metro and long-haul network. Today, the companies that we are going to be competing with in the wavelength market are doing so on a network that was not optimized for wavelengths.
It sold many different products, and that meant that each time a wavelength had to be provisioned, it was a bespoke process that would typically require six field visits by technicians and custom engineering to architect each wave on a one-off basis. We took a very different approach. We learned from our 25-year history in transit, where we have become the largest transit provider globally. We identified 800 data centers. We needed to have optical continuity and a standard architecture between them. And then, when we got a wave order from any to any of those facilities, all we would have to do is deploy two field dispatchers to plug the optics in at the endpoint and then could automatically configure that wavelength. That'll bring the provisioning time down and improve the accuracy and reliability of that provisioning.
And about unique routes for you?
Unique routes are an important component, so a wavelength is an unprotected product. As such, any cut on anybody's fiber needs to have a backup path. 90% of the Sprint routes are unique. They were built differently than most other networks. First, they were built with SMF-28 fiber, which is a better fiber for coherent transmission, which is today's industry standard. Two, those routes were buried in an armored cable instead of a plastic conduit. They're typically six feet down versus two, which made them much less susceptible to cuts. The uniqueness of the routes, the quality of the fiber, the ubiquity of the endpoints, the ability to provision quickly, and again, we have an asset that we have a cost basis of $1 in. That gives us a capital advantage over companies that built their networks with either equity or debt that they're still servicing.
In terms of revenue and the revenue ramp you can expect, you'd reported just under, I think, $3.6 million last quarter. You've talked about getting to $500 million annualized over time for wavelengths. Give us a sense about, you know, when this really starts to go nonlinear. You say you expect a big step up in 2025, but when can we kind of, like, sort of approach that, you know, that $100 million dollar plus revenue run rate per quarter or so?
Yeah. So our foundational work is critical to be able to scale this model. That work will be completed in about three and a half months, by the end of the year. We're on track to complete that work. That will allow us to provision the backlog that we've built, as well as a new backlog. We currently have sufficient staff and processes to probably provision about five hundred wavelengths a month. That seems to be about right for the current market and our backlog. We currently provision about 3,000 IP ports a month. We then have churn that we are replacing, but on a gross basis, we're provisioning about 3,000 . We're using similar tools, similar people, similar processes.
I think we have argued and modeled that by May of 2028, five years from the acquisition, we would be at 25% market share or a $500 million run rate. We are behind that now, obviously, at 3.6 a quarter, roughly $15 million a year today. That number will ramp up materially in 2025 and in 2026. Probably by mid-year of 2026, we'll be back on pace to show that kind of annual $100 million a year growth rate in wave business, and that's kind of where we think we'll be in years three and four to get to that $500 million number.
Mm-hmm. Very good. And I guess just maybe talk about the backlog. You've sort of touched on it before. You've got sort of a backlog of, I think, it was 2,700 or so. I mean, do you think you can get through all that backlog? Do you think you can get through the backlog before it potentially expires or goes away, and just your confidence level in achieving that?
So some of it will expire because of the prolonged installation periods. We have let some customers out of contracts only to get additional orders from them. We continue to build new orders. We also have not really focused the sales force on building a much larger backlog at this time because of these elongated provisioning cycles. I think as the foundational work comes to closure, we will effectively turn the sales engine on at maximum capabilities. Obviously, there's a market discovery process, there's building credibility with customers, and I think the most important thing for us to build a bigger funnel is to provision quickly. In other words, people want to just keep giving you orders if you can't actually install them. Once you demonstrate you can install within a couple of weeks, your customer base will have confidence and give you incremental orders.
That's the exact same process we followed in the IP business that got us to where we are.
You talked about different speed grades for wavelengths, the 10, 100 gig, 400 gig. Obviously, there was a long time getting to a hundred, but obviously that's most of where the sales are today. But a lot of people have asked me, you know, investors have asked me, "Well, given what, you know, this potential, you know, gen AI data center demand looks like, might we see a faster upgrade to a 400 gig cycle?" What's your view on that?
So AI will be an additional driver of wavelength demand, but price per bit will continue to fall because optronics will continue to improve. We've been through multiple applications that have driven demand on the internet. I think the same is true of the wavelength market. Applications come, and then new applications replace them. I think the migration to 400 today is driven mainly by companies trying to minimize cross-connect expense in data centers. So, waves today come in three flavors, practically, 10, 100, and 400. There was a time when there was one gig wave. There was actually a time when there were half gig waves and then 1 gig, 2.5. 10 is now practically the smallest wave that is sold in the market.
Most of the market is 100 gig waves, and we are seeing a migration. It's a few percent of the market to 400. We will also see a migration beyond that to 800 and 1.6 terabits. So the wavelength equipment that is available today, in ideal circumstances, can support 3.2 terabits per wave, but it never works that well in the real world. You've got reflections, you've got jumpers... you've had fiber that may have been cut or different types of fiber that have been spliced together over the years. So the reality is, you'll get something less than the full rated speed. So a 3.2 terabit system, if you deploy it, will probably deliver in the real world about 2 terabits per wave.
The customer won't accept a single two terabit handoff, so you will use a muxponder to break that down into increments that they can accept. Virtually all of the equipment today is hundred gig enabled. Very little of it today is 400 , and there is no commercial 800 gig router ports available. There are lab versions, but I think we're still a couple of years away from people taking 800 gig handoffs.
Mm-hmm. Maybe if we could switch to the core corporate business for a moment, it's kind of your bread-and-butter business. You know, I think you were, you know, post-pandemic, fairly cautious previously about the amount of recovery, kind of given, you know, in-office trends, post-COVID, and so on. But since then, it's stabilized and it's growing okay at this point again. Maybe talk about some of the key drivers and reasons for that recovery, and how kind of close do you think we are to kind of getting back to that core, sort of 2%-3% sequential or 10% plus annualized revenue growth rate in the core?
Yeah. So it is growing, but I don't think it's growing satisfactorily at about 4% annually. So our corporate business, where we're selling internet connectivity in skyscrapers, about 1,860 buildings, about 1 billion sq ft of office space in the central business districts of major North American cities. That business, for 14 and a half years, grew at an 11% growth rate, pretty darn consistently. We would get a little bit of variation, but it was about a 2.5% sequential growing business. That growth rate decelerated in the pandemic to negative 9% growth for about a year, and then as people returned to offices, we saw that growth rate gradually improve, and today it's at about 4% year- over- year, or about 1% sequentially.
You know, the occupancy rate in our footprint went from a vacancy rate of 6% up to 18%. It's back down to about 16%, so it's coming down, but still far worse than it was pre-pandemic. Secondly, a number of companies, I think, had been reticent to make a big architectural decision and change their networks. I think that resistance is slowly going away, and that business is improving. I think we'll continue to see gradual improvement in that corporate business, but I'm out of the market of trying to predict when we get back to pre-pandemic levels because I've been wrong so many times.
Then that absolves me of all responsibility. Maybe, can we talk about the core competitive dynamics in the business? Because I think historically, there's been a lot of pricing pressure in fiber, on the corporate side. You've talked about that, I think, fairly openly. I mean, are you starting to see any of those dynamics change structurally, any moderation in pricing or potentially improvement in those dynamics?
So let me separate our end user business from our NetCentric business, 'cause they really do have very different pricing dynamics. So in our corporate end user business, we've actually seen effectively static ARPUs, but a higher bit volume, so the price per bit comes down. So a typical corporate customer buys a connection that's on twenty-four seven, and it's at full rate bi-directional. Over time, those connection speeds have continued to increase but have kept ARPUs relatively flat. We first saw a transition from a hundred meg to gigabit, and we're now seeing a transition for corporate end users from one gig to 10 gig interfaces, as their equipment in their local area networks has a 10 gig WAN interface, and they can now connect at that speed. In our NetCentric business, the dynamic is very different. We're selling bulk internet connectivity in a data center.
We're differentiating ourselves based on price, and the average price per megabit has fallen at about 23% a year for the past 25 years and will probably continue to fall at that rate. We have been able to expand margins, lower capital intensity, and undercut our competitors and gain market share. Our NetCentric product is a metered product. It is rarely sold on a per connection basis, but rather, you buy a connection with some level of commitment on that connection, and then you pay a metered rate above that at a slight premium. The vast majority of that NetCentric business is usage-based or metered. Those customers have become accustomed to prices declining, and they probably will continue to decline. So on the corporate side, we typically compete with the incumbent phone company, a competitive carrier, and sometimes a cable company. That would be the corporate competitive landscape.
On the NetCentric side, for internet transit, it would be companies like Arelion and Lumen, much lesser extent, some regional players like Tata or NTT, Telecom Italia, Sparkle, Telefónica, Orange Business Services, but it's a very different competitive dynamic, and there, the decision is really a one-dimensional decision based on price. And, you know, we have continued to maintain a significant pricing lead over our competitors, and for that reason, many of our competitors who historically been in this market, have dropped out.
Mm-hmm. I want to come back to the asset sales, but make sure we have time to get to the financial questions. So maybe we'll go there next. I think your last earnings call, you talked about your sort of long-term annual growth targets being 5%-7%, 100 basis points of EBITDA margin expansion annualized, on a multi-year basis. You know, what are the kind of key assumptions underlying those targets, and do you see, you know, potentially—when you sort of get there and, you know, how do you think about arriving at that target?
So let's start with where we've been. So Cogent went public in two thousand and fifteen, and between two thousand and fifteen and two thousand and twenty, we grew our total revenues organically, with not a single acquisition, at 10.2% per year, compounded. With that growth in revenues, we delivered 220 basis points a year of margin expansion. The reason for that excellent performance was roughly 75% of our growth was coming from on-net services, which carry a virtually 100% gross margin, 95% EBITDA incremental margin, and only 25% was coming off-net. Now, let's layer in two additional types of business to get to the growth rate you're talking about. We're going to layer in an acquired enterprise business that we took over from Sprint, that we're actually intentionally shrinking.
It will stabilize and be about a 20% margin business with little or no growth, maybe a 1% growing business. Our core Cogent business will probably slowly continue to recover back up to close to that 10% growth rate that we delivered for the 15 years pre-pandemic, but for the reasons I described earlier on the corporate side, we're not there yet, and then third, we're gonna have a very rapidly growing wavelength business off of a very small base, but that base will get big quickly because of our competitive advantages.
So as we looked out over a five-year period from the announcement of the acquisition of Sprint, we said, combining these three businesses together, we would grow at somewhere between 5% and 7%, and because of the off-net concentration in that enterprise base, we would probably only deliver about 100 basis points a year of margin expansion.
Fair enough. You know, last time you were at Communicopia two years ago, you just announced the Sprint acquisition. Last quarter, you talked about, you know, you were 62% of the way through the original $220 million cost saves target you've outlined. Maybe give us a sense about, you know, how are you feeling about the progress toward those targets? Any update or are you thinking about those numbers being higher over time? And then how quickly can you get to the next 40% versus the first 60%?
So we outlined a total of $220 million in cost savings, and we said those would take 36 months to complete from closing. In the first 15 months post-closing, we had accomplished 62% of that target in actual cost savings, and you saw that in our gross margin expansion and our reduction in SG&A. The ability to take further costs out are governed by the contracts that we inherited. T-Mobile paid us $700 million, not because we were nice people or they liked us, but because we took a series of obligations on that we have to roll off over time. We are doing that. We still think it will take the full three years to squeeze all of the cost savings out.
The good news, however, is we have identified other areas that we are taking out additional cost savings, and I think over the next couple of quarters, we'll have enough confidence in those incremental areas to lay out a revised target for investors. We have not done that as of yet, but I think that two twenty number will go up.
It's good. Last quarter, your net debt EBITDA leverage ratio was about 3.2, I believe. That's in your stated target of two and a half to three and a half times. You know, once the T-Mobile payments, you know, roll off, how do you think about kind of addressing those headwinds to your leverage? Do you think that you're actually gonna get enough cost saves to kind of, like, do much better, so the EBITDA is actually underlying much stronger than that to get to that same level? And, you know, to what extent are you comfortable kind of operating towards the high or even above the high end of the target leverage range?
So prior to the Sprint acquisition, our leverage had actually been above that range at 4.7 times net leverage, and that was in large part because we did not slow down the growth rate and the dividend quickly enough with the pandemic. Again, I'll fully admit my mistakes. I thought the pandemic was more transitional and, you know, would last a year, not three years, and we kept the growth rate at 2.5 cents a share sequentially, as opposed to the 1 cent we've reduced it to. As a result, our leverage drifted up. We materially delevered in the first phase of the acquisition of the Sprint network due to the higher payment stream from T-Mobile and the accelerated cost savings.
We will probably see our leverage drift up from here till mid to late next year at probably peaking around 4.2-4.3 times, which is above our long-term range. And at that point, we will see the leverage each quarter sequentially decline on a net basis. This is also assuming that we will continue to grow our dividend at the same pace we have recently, and you know, we have grown it now for 12 consecutive years. We also will opportunistically buy back stock. We did some last quarter. You know, it'll be just based on market conditions, and that could impact our aggregate leverage. We'll make those decisions as we go, and this is probably a tie-in to your question you alluded to, which is asset sales.
We have three asset bases that are not today in our financial projections, because we just don't have enough information to know how much we will get from them. We have roughly a million sq ft and 100 MW of surplus data center space that we're looking to either lease or sell in the market, and we are in the process of repositioning those assets from former switch sites. We have dark fiber across our 19,000 route miles of intercity fiber and 1,200 miles of metro-owned fiber that we acquired from Sprint that we're gonna look to divest of. We have a large inventory of IP addresses. We recently validated those addresses by securitizing a portion of the rental income, raising $206 million in an ABS offering.
But I think we also may look, if it's an opportunistic time, to try to sell some of those addresses. So I think there are three levers we have beyond the operating business to adjust aggregate leverage. We have today, $462 million at quarter end on our balance sheet, plenty of cash, and access to capital, so we feel pretty good about our liquidity position.
I think you've timed that masterfully. So we're, I think we're out of time, but thank you very much, Dave, for being with us today.
Hey, thank you, Jim.
Thank you.