Always pleased to have Dave Schaeffer from Cogent here with us. He is kind enough to come every year, and great to hear the latest from you, Dave. Thank you again.
Michael, thank you for hosting me, as always. Thank BofA for a great venue, and thank investors for taking time out of their day to hear what we want to talk about.
Outstanding. I had a number of questions that I want to go through. I will leave a few minutes at the end if there are any questions from the audience. When I mention that, there should be microphones going around, so just raise your hand and we will take time for that. Dave, I think you would admit that growth has been disappointing and below your own projections for the last year or so due to a number of factors. Can you just remind us maybe what went right and what went wrong, or maybe not wrong, but didn't meet expectations in the last 12 or 18 months, and then why you believe that is going to change in the next 12 or 18 months?
Let me start by laying kind of the backdrop. Cogent had done no acquisitions between June of 2005 and May of 2023. In that 18-year period, we had a compounded average growth rate of 10.2% a year. Our corporate segment actually was outperforming that, growing at 11.2% a year over that 18-year period. Because the majority of our sales, 76% of our revenues, were on net services, we were experiencing consistent margin expansion. Our EBITDA margins expanded at approximately 220 basis points a year.
Because the on net's about 90% incremental margin to growth, right? Correct. So that's like...
No, the actual total margins-
Yes.
-had expanded. So we went public in June of 2005 with zero EBITDA margins.
When we acquired Sprint, our EBITDA margins were 40.5%. We were acquiring a business that was in decline and burning cash. The business we acquired for the three previous years had declined at 10.9% a year, and it represented 42% of the revenues of the combined company. We expected our total revenue growth to go negative. We also knew that the company we acquired was operating a number of unprofitable products and network services and had a negative 60% EBITDA margin. We outlined a plan to take costs out, and more importantly, we had a program to repurpose the physical assets of that business that had been sitting dormant since 2015, and turn that old TDM voice network that was built at a capital cost of over $20 billion in 1980 dollars into a modern wavelength network to serve international carriers, regional carriers, and content distributors.
We did not anticipate the tailwind of AI, so that was a blue bird. I will not profess I was clairvoyant. In September of 2022, when we signed the deal, we knew that AI was going to burst upon the scenes and create a $7 trillion capital investment cycle. What went right? The ability to take costs out were actually better than we expected. We actually raised our cost reduction targets, and we achieved those cost reductions quicker than we expected. What went wrong was we lost more of that revenue than we expected. Now, in a way, that's a blessing in disguise because that revenue was negative margin. Today, the Sprint customer base has gotten margin zero.
It doesn't help, but it doesn't hurt. It has shrunk from 42% of our total revenues to only 15% of our revenues. We today have a business that is at least break even in those Sprint customers. The second part of the transaction was the establishment of a new business, a wavelength business. We actually reenabled the Sprint network on budget and on time. We did not exceed our goals, but we did not fall short. We also made the election after the transaction was announced to invest an additional $100 million into the physical buildings that we acquired and convert those former switch sites to data centers. We had a program to divest of 24 of the 125 that we did convert. We chose those based on their location, size, and proximity to major markets. We then completed that.
There was a great deal of skepticism that the company already was spending money on building a wave network and then committing an additional $100 million into what was viewed as a speculative endeavor. We, I think, proved that investment to be correct. We sold 10 of those 125 facilities for $225 million, $224.1 million net, because there was very little transaction costs associated with that divestiture. We did that in-house, and we used those proceeds to reduce debt on our balance sheet. I think on the data center monetization, it was not originally anticipated.
We did spend more than we expected, but we got more than we spent back, and we still have 14 data centers that we plan to divest of using those proceeds to further delever, and we will be left with the remaining footprint that while we view it as not monetizable through sale, we should be able to lease out that capacity.
Dave, just quickly, on the 14 remaining data centers, do they look like the 10 that you already sold, meaning location, occupancy? Should we expect similar valuation per megawatt for the remaining 14?
We had never sold a data center prior to this. We had no real view on how well we will do, other than we felt that we would do much better than the $100 million invested. We ended up getting $4.3 million a megawatt. The 10 facilities we sold were about 53 megawatts, getting to the 225. The remaining facilities were 54 megawatts, so about equivalent.
At least from our perspective, look relatively similar. I think the expectations should be about the same. Each of the facilities has some different characteristics, whether it be physical site or availability of incremental power. Almost all of these facilities are empty with no real revenue associated with them. We have the added benefit that when we divested of the first 10, we reduced a negative drag on our EBITDA by $7 million a year. We were effectively paying the real estate taxes, personal property taxes, and maintenance of facilities with no revenue.
These are old wire centers, right?
These were...
They were not purpose-built data centers. They're old wire centers you converted with $100 million to make them into data centers. So they're different maybe than a traditional data center.
That's correct.
Correct. Right.
What these were, if you look to the typical facility, it was 15 miles from a downtown, on six or seven acres with a 40,000-foot building with 4-5 megawatts of inbound power, full of old Class 4 and Class 5 telephone switches.
Yep.
There were 23,800 cabinets of telephone switch gear that we removed from the footprint and disposed of. We had to convert these facilities from negative 48-volt DC power plant, which is how a switch site operates, to AC 120, which is how a data center operates. We had to refurbish the air conditioning, the fire suppression systems, the security systems, and do some cosmetic refurbishment to then make these basic data centers. These are still facilities that have only about 100 watts per sq ft and are not today able to support high-density liquid-cooled cabinets. Now, they are modifiable to that, but they are smaller, much more distributed than the types of data centers that are in the headlines today that people are protesting, where people are trying to stop the construction of a 1 gigawatt, million sq ft facility on 1,000 acres of land.
These are much more innocuous, and they, I think, will have a great deal of applicability for inference. In fact, I Squared, the purchaser of the first 10, did not disclose to us during the purchase process what their ultimate plan was. But subsequent to buying them and now standing up a management team to start to commercialize these facilities, they've disclosed to us that their ultimate plan is to make these inference facilities and kind of have token repositories distributed around the country where they can then be providing inference as a service to neo clouds and hyperscalers. They ultimately intend to invest $1 billion, at least that's what their public statement is, 225 of that came to us. Now to what else went right and wrong. I want to finish the answer to your-
Yeah, please.
-first question. Cogent's core business actually accelerated. 84% of our revenues come from selling internet-based services. That business had been negatively impacted by the pandemic. It had seen its growth rate fall from that 10.2% down to 5% pre-acquisition. In the three years since the acquisition, the core Cogent business has actually seen its growth rate re-accelerate to about 9%. That business, over a three-year period, grew 29%. The acquired Sprint business rate of decline accelerated and declined in totality at 69%. There is both good and bad post-transaction in our two customer bases. Probably the biggest area of disappointment has been in our wavelength business, our brand new business.
We knew that it was going to take close to a year and a half to two years to do the conversion. We actually started on that conversion prior to the acquisition, after the deal was announced, and we completed the enablement of an initial footprint of 800 data centers by December 31, 2024. So about 18 months after the deal was closed. Right on plan, right on budget. Where we were disappointed, as we hoped to be able to sell more wavelength services on an interim basis as we were bringing these facilities online. We sold about 1,000 waves in that interim period, not the 3,000 that we had hoped to. Then, once the entire footprint was enabled, we have continued to grow that business. Last quarter, it grew 62% year over year.
It grew 9.2% on a sequential basis and has performed well, but off of a small base. We today have 3% market share in the North American inner city wavelength market. Ultimate goal is to get to 25% share. We were hoping to do that by mid-year 2028. I don't believe sitting here in the fall of 2026 that is realistic. I think the ultimate goal is realistic, but the pace to get there is going to take longer. I think the two reasons why our ramp and revenue growth was slower than we anticipated were one, our need to build credibility in this new product. I think we had earned and have a great deal of credibility in our core product suite, but I think many customers did not associate Cogent with wavelengths and were apprehensive about our ability to deliver the quality that we represented.
I think we've solved that problem. We have now sold wavelengths to 548 unique customers. We've delivered service in 608 of the ultimate 1,137 locations that we brought online. I think the second challenge has been the supply chain challenges in the entire technology and telecom ecosystem that we did not anticipate. Now, on the positive side, we get the tailwind of a new incremental use case for wavelengths, that is AI training, connecting a data source to a training facility and doing it with defined latency. While that is more expensive and harder to administer than using the internet, it does allow for more optimal utilization of the GPU capacity. Since 60% of the capital cost of a training facility is in the GPUs, maximizing their efficiency is critical. And with that, we...
Dave, how does it maximize the efficiency of the GPUs?
Every large language model works on the same basic architecture. They each have variances, they each do it slightly different, and they are evolving. But what they do is take a large set of data that was collected over the internet.
Literally exabytes of data. They tokenize that data, breaking words into recognizable tokens that GPUs can process. They then pose a question to that set of data that they already know the answer to. The GPUs crank through that data multiple times and develop a pattern upon which the predicted answer and the actual answer are matched. They do that recursively, repeatedly, and over time, they build a neural network that can then be applied to a new set of data. Most of the data storage exists in preexisting data centers and major data collection points, Northern Virginia, Ashburn, San Jose, Greater Chicago market, Greater Dallas market, Miami market, Amsterdam and the Netherlands, the U.K., mostly in England, Frankfurt. As a result, there's a couple of dozen locations globally where most data sits. Most of the training is occurring in locations that are remote from those locations.
They are locations where there is power and space available. We're seeing new data centers being constructed in some very unobvious locations, Southern Louisiana, Central North Dakota, Western Wyoming, Western Texas. They're going to those locations because power, land, water are available. So in order to train, you have to take the data out of the data center where it's stored, move it to where the GPU is, and then return it back after the training is complete. That process is best done over a wavelength, because a wavelength is a deterministic point-to-point service. On a per bit mile basis, it's actually about 2.5x more expensive than using the public internet. But if you use the public internet to transmit that information, you would have to buffer that information, or you would have to be able to degrade the productivity of your GPUs.
All of the AI training is today done using wavelengths as an incremental demand set to what we expected from regional networks, from international carriers, and content deliverers. This trend is going to continue for multiple years, creating a huge amount of demand. What that training demand has done, though, is tax the entire supply chain, whether it be power generation, whether it be power transmission, whether it be land, whether it be cooling systems, switchgear, fire protection. At every level, this surge in demand has created shortages. There is a tremendous memory shortage, where prices have gone up by a factor of five. Telecom equipment, which would typically be orderable and deliverable in 90 days, has now got a three-year delivery window. You have got a new class of buyers.
At a traditional telecom equipment supplier, 90% of their revenues were coming from service providers like Cogent or AT&T or Deutsche Telekom. Now, 60% of their demand comes from one of four major hyperscalers. This is a shock to the system that has created a number of supply chain constraints. That impacts us two ways. One, it has raised prices for equipment that Cogent and other service providers buy. It has created telecom equipment inflation, which is unheard of. For 50 years, price performance had consistently improved. This is the first time in the digital age where price per unit of productivity is going up. That will get reverted, and we will return to the Moore's Law type of price declines, but that is probably going to take a couple of years to work through all of the systematic constraints.
Can I ask you, Dave, before I forget, you mentioned 3% share today, hoping to get to around 25% over time, and obviously AI has been the unexpected, right? Positive, incremental. How much is AI training spending on wavelength today in general? Or not in general, but how much are they spending today?
Today, AI training, which is coming both from hyperscalers and neo clouds, represents about a quarter of the total market, but it is the fastest growing segment.
Okay.
It is not always easily measurable because the major hyperscalers also operate traditional cloud distribution businesses. They typically buy wavelengths through a centralized purchasing mechanism, and they don't always disclose to Cogent or any other supplier what the ultimate use of that wave is. So when a Microsoft or a Google or a Meta or an Amazon buys a wavelength, they don't put a little check mark and say, "This wave...
Yeah, you don't know exactly how they're using it. You can't differentiate. Is your market share of that 25%, is that in line with your 3% share or is it higher?
I think it's a little higher with that group of-
Okay.
-customers for two reasons. One, for the existing hyperscalers, we had longstanding 15-plus year relationships and have been able to leverage those. Secondly, for the neo clouds, our ability to have such a broad footprint and rapid deployment schedule-
Yeah.
-has differentiated.
Your provisioning is much faster than others, I think, and for wavelengths, correct. Yeah.
Yeah. That really helps us with someone like an Anthropic or an OpenAI or a CoreWeave or a Nebius or a Gcore or Lambda Labs, this next wave of companies that each have slightly different business models but are really only in existence for AI training at this point.
Yeah. Let's just say that, just round numbers, say your 10% share of training wavelength traffic or spending, who has the other 90%? Do they have a 90% because they're building connectivity, providing wavelengths to these out of the way AI training facilities that you never had or will connect to because they're not part of that number of data centers that you're providing wavelengths? Is that why? Then who has that 90% that you don't have, if I have that number roughly right?
That is not correct. We today go to 59 single-tenant facilities. Cogent is willing to build to a facility, but needs to have sufficient commitments from that single tenant. There are four ways in which we can get to a single-tenant facility. The first is we just build there.
Yeah.
We're happy to do that if either the customer signs a long enough and large enough multi-year contract, or they sign a shorter contract and give us enough of a non-recurring payment to cover our capital expenditure. Because if we build to a single-tenant facility and for whatever reason they stop using us, whether their business model fails or the facility becomes dormant or they switch providers-
Yeah.
-we then have a completely stranded asset.
Yep.
The second way we can do that is, in many cases, these companies have already built their own fiber as part of developing these facilities back to a major aggregation point or a point where they can splice into our network. Our network is probably one of the most diverse in terms of physical routes, and has more splice points available to it. We are, in many cases, connecting to those network locations by the customer's own fiber. The third model is we buy fiber. Cogent's core business had always been predicated on buying IRUs, indefeasible rights of use. We have purchased 94,000 route miles of inner city fiber, 34,000 route miles of metropolitan fiber from 383 different suppliers around the world. We then take that dark fiber and put our own optronics on it.
Oftentimes, when we go to a remote location, we may piece together two, three different fiber purchases to get very close to that location, and then do literally maybe the last 100 yards, the last half a mile into the facility. We also have that ability to use our network of providers to allow us to get close to these facilities. Then the final option is for the data center owner to buy lit services from someone else back to a carrier neutral to then ride our network. That is more complex, but it's also more desirable. The internet, by its very design, is resilient. It constantly reroutes traffic around problems. That's the fundamental design of the internet. A wavelength is an unprotected point-to-point service. Which means any physical interruption along that path and the service fails to work.
Almost no customer buys wavelengths as a single service. They almost always buy a primary and a backup wavelength. In the ideal case, they would like to have the same two city pairs, identical latency, but complete vendor and complete path diversity. That ideal rarely exists in the real world. What you do is get as much diversity as you can. An example would be if I'm going from western Nebraska back into Denver, I may be able to get diversity 80% of the way, but there may be an Indian reservation I connect through in which there's only one physical fiber path. I'll have a pinch point where part of that path will have a common point of failure.
Then when you connect in Denver, you would like to ride a network that's completely diverse from 1501 15th Street, that's a major aggregation point in Denver, to say, 707 Wilshire, a major aggregation point in L.A. There you can get a completely independent wave, and you might buy one wave from one vendor and a second wave from us. So Cogent actually has done quite well with customers that buy waves from others and using us as diversity, and conversely, people that buy from us as primary, we use other vendors for that diversity. The final point that makes this even a little more complex is that customers sometimes substitute dark fiber for wavelengths as a fungible product.
It is yet again more expensive, it is harder to implement, it has a longer lead time, but it may be the only way to serve some of these remote locations because no one would build to that location speculatively. And in those cases, they may light that dark fiber and then look to buy a wavelength to protect as much of that physical path as they can. So the wavelength market continues to grow and the historic use cases are pretty static. It's really AI that's driving the growth.
Okay. I have one or two more, but I did want to open up to the audience, if there are any questions. I know we have a microphone that can go around. Just raise your hand if you have any questions. And while we're waiting for that, I just want to go back to the question, because I really appreciate you explaining that to me. So who has the most market share then in wavelengths, if my 10% estimate was even roughly right, who makes up that 90% and why are they being selected? And then second part to the question is, how have AT&T and Verizon now emphasizing this market as well as a growth area, how has that impacted your business and your wins?
In the inner city market, Lumen is the dominant player. Zayo is the second largest player. Those two companies became dominant players by rolling up many other network players-
Yeah.
-and gaining market share. It has been a fairly concentrated market, and it's a market that historically AT&T and Verizon did not participate in. They have recently announced that they would like to get back into that market, but have not yet made meaningful investments or inroads into that market.
Verizon's had an announcement or two, though, right, with some partnerships with hyperscalers.
They are, Verizon at least, has announced a new build-
Correct.
-for a hyperscaler-
Yeah.
-along with the sale of some of their inventory.
Correct. Yep.
Verizon is actually one of our 383 counterparties we buy select fiber from.
Yeah.
They're willing to sell the fiber, whether it's to us as a service provider or a hyperscaler, but they are also willing to do brand new builds. To remind investors, about a decade ago, Verizon announced a program called One Fiber.
They built out metropolitan fiber networks in 69 markets out of footprint, which was really unprecedented for a phone company. They spent about $3.5 billion building that network, primarily to support their wireless backhaul requirements.
They have subsequently tried to monetize that by selling wavelengths. Verizon is a wavelength seller in the metro nationally. AT&T is a dominant seller in its footprint, but does not sell wavelengths in the metro out of footprint. Cogent is not focused on that metro market because we have no real competitive advantage. Our primary competitive advantage is in the inner city market. If we look at the total North American market at about $3.5 billion, roughly 150,000 discrete wavelengths. The dominant players in the metro are Verizon and AT&T second, followed by Zayo third, Lumen a distant fourth, and Cogent way behind.
And then on the inner city piece, AT&T and Verizon, actually Verizon's predecessor, MCI, initially developed an inner city transport network using microwave. They converted to fiber in the late 1990s, but they were burned twice. First by switch-based resellers who all went bankrupt in the LD market, and then by regional ISPs. At that point, both AT&T and MCI exited the wavelength business and really left the market to the new entrants. It was Broadwing, it was Williams, it was Qwest, it was Level 3. And over time, Lumen has consolidated many of them. Zayo's consolidation started in a different tack. They bought a bunch of local metro markets and eventually linked them together. The two of them have been the dominant suppliers of inner city wavelengths.
And where Cogent has an advantage is we come to market with unique routes that those- carriers don't have and therefore provide diversity that they cannot provide.
Okay. That's very helpful. We have about a minute and a half left. So I'm going to leave it for one more question, Dave. I want to stay on wavelengths here for a second. You kind of answered this part earlier, but part of what you attributed the slower growth to earlier in the year was you said Cogent's provisioning service so fast and customers just can't take it or accept it. Are you still seeing that issue where customer's not accepting delivery once you provisioned it? Or is that time shrinking?
We are still seeing that issue, and I think it will continue for some time.
That's the equipment issue? Because there was some question about if customers were even just maybe over-provisioning or overbuying wavelengths, not knowing exactly who was going to deliver and who wasn't, and then because you weren't forcing them to take delivery, there was no penalty, right? So there was no risk in overbuying if that is in fact what they were doing.
There is actually a penalty.
Okay.
These contracts are take or pay contracts, but as a new entrant into the market, we are careful about when we force bill customers-
-and pressure them into taking something that they would like to delay. Customers are not always transparent with us on the reasons for the delay. In many cases, the customers are at the mercy of their suppliers. They signed a lease to be in a data center that was supposed to open in June, and it is not going to open until November. Or they go to a data center and the data center operator says, "I am capped on power now and will not have incremental power available for sale for four more months." Or they may find that they are behind in receiving their GPUs, they are behind in receiving servers, and it could be a combination of multiple constraints. All of these things have caused customers to do two things, to delay and also be more agile than normal.
We have had dozens of cases where customers bought a service to a data center in a market. We provisioned it, and they came back to us and said, "Hold off. We need you to move to another data center in the same market because the first data center was out of capacity for us.
Okay.
Now, I do think that as the supply chain issues get resolved, and as the industry matures, these problems will dissipate. We are in a situation where $7 trillion of capital has been announced and only $1 trillion of it has been spent.
That was excellent, Dave. Thank you for our time. Thank you so much once again, always appreciate it.
Thanks, Michael.
Dave, thank you so much.
Thank you all very much.
Thank you again.