Thank you for coming to Day One of the KBCM Technology Leadership Forum. This is a 25-minute fireside chat with us. We have Dave Schaeffer, CEO of Cogent Communications . Dave, thanks for being with us.
Hey, Brandon. Thanks for hosting me. I'd like to thank KeyBank for a great venue. Most importantly, thank investors for taking time out of their day.
Let's start on waves, Dave. I think investors have high hopes for this business that you're starting up. It's taken a little bit longer to ramp. What have been the factors, from your perspective, that caused this business to ramp more slowly than expectations?
I'll actually break that into three components. We always knew that it was going to take us about a year- and- a- half to take the former Sprint voice network, interconnect it to Cogent's existing metro network, and reconfigure it. We actually slightly beat that schedule in terms of number of sites and the speed at which we did that interconnection. The second piece has been the ability of us to build a funnel showing that there's adequate demand out there. I think there is a great deal of belief among investors that the wavelength market is real, that it is growing, and that AI has been an incremental use case that has driven more wavelength demand. It is a market that is highly concentrated with two players controlling the bulk of the market and then a handful of smaller regional players, the remainder of the market.
What we needed to do was three things: one, demonstrate that people would be interested in Cogent, and do that before we really had a network. We built a funnel of almost 10,000, what I would call expressions of interest. About 1,000 of those turned into orders. About 1,000 of those pushed forward and have been installed. About 8,000 of those early expressions that sat around for a year, year- and- a- half fell out. As we began to give customers specific implementation schedules, we started to build a much larger and more certain funnel. We ended the last quarter with a strong funnel.
We had on earnings day about 4,700 orders that included orders that have been installed but not accepted by customers, orders that are in provisioning, orders that are signed but not yet provisioned, and finally orders that the specific route and locations and price have been agreed to, but a contract has not been signed.
Real quick, on the 8,000 that fell out, how do you get confident that those didn't go to one of your competitors?
I think they did. I think they stayed with our competitors because of our inability to install them. In that period from deal closing in May of 2023 through the end of 2024, we told customers we were actively working on putting wavelength services into a defined list of 800 data centers. We even gave the customers the list. We also said we would take an order, but we had no ability to tell the customer when we would install it. We’ll take the order. You should continue to look elsewhere. If we can wave-enable your two endpoints, we’ll contact you, and we will try to manually install that order by engineering that path. We did that for about 1,000 waves. The process that we have going forward is a much more automated process.
What we did is pre-enabled now 938 data centers where we could provision a wave from any data center to any data center in 30 days, and we can do that in one of three interface speeds: 10G, 100G, or 400G. That is radically different than how the market works today. The private line market, which is really what wavelengths are dedicated in circuits, had been around for nearly 80 years. For the first roughly 55 of those years, you had to buy them from AT&T and then a little bit MCI. The market changed in the late 1990s, and third parties like ourselves and Level 3 and BroadWing and Williams built their own networks. We chose not to build a wavelength network. We built only an IP network. Others built networks and sold waves. Waves were always sold as surplus capacity.
Kind of think of it as the airline model where you want to sell full price tickets, but you’ll go to Priceline.com and give them your surplus. That market developed into a fairly robust $2 billion market. We realized that we had some architectural advantages that would allow us to capture share. We could go to more points. We could build a network without surplus capacity, but rather one that’s designed for quick provisioning, one along unique right-of-ways, one that would be more reliable, and one that was less expensive. I think our approach to this market is really very different than the two major competitors.
When we think about sort of the 4,700 units in your backlog, are those now sort of signed contracts in backlog? Are those still indications of interest on orders?
There are four categories. There are waves that have been signed contracts, been installed, but customers not accepted.
Yeah, you mentioned that those are.
Those should be between $200 and $300.
$200 and $300 that should have been installed last quarter, right?
Should have been counted in install.
OK.
Because the customer wasn't ready, we didn't start billing them, and we only recognize revenue when the customer accepts. We had the same issue in Q1 where our unit installs were far greater than our revenue install. The second bucket are orders that are signed that we're still doing work on. We're still provisioning.
How big would you say that bucket is?
A few hundred.
A few hundred.
Third, there is a bucket of signed orders that have not yet been provisioned.
Sure.
In large part because the customer has told us when we contacted them, we're not ready yet. We still want it, but we're not ready. We have a signed order, and we're ready to go. The final bucket is those orders that the customers agreed on the exact route, the exact interface, the exact speed, and the price.
But.
All of the key terms.
Without a sign.
It has not signed.
What is signed contracted at this point, would you say, in the seven?
Probably close to half of that.
Close to half. How do you sort of bridge us from sort of getting to your target, exiting this year of, call it $20 million plus in quarterly waves revenue? How does that install metric look? The way that I thought about it is you have a couple hundred, let's say $250, that should have been installed in Q2, plus you have the backlog. How should we think about sort of bridging from Q2 to Q3 and Q4?
At the end of the quarter, we had about 1,500 waves installed and billing. About 1,000 of those were done in that manual process between deal signing and January of 2025. We've installed and billed almost 500 the first half of the year. We need to install and bill about another 500. Each, 1,500 in total to get to 3,000 installed billing.
Exiting this year?
Exiting the year, which gets us to that $20 million run-rate from the $9.1 million run-rate we demonstrated in the last quarter.
OK. As we think about taking that backlog number, one of the metrics you sort of gave us is that you expect sort of 5% of the backlog to install each quarter. Should we use the contracted backlog as a better indication of, like, on that 5%, what should be installed?
It's a combination of both. We typically have not given KPIs like backlog or book to bill. I think in the long run, those are inappropriate. I think investors should hold us accountable for GAAP revenue that's reported and audited. I think at the early stages of wave deployment, it is absolutely necessary to give investors these incremental indications of demand: a funnel, a conversion rate, a pipeline that's being provisioned, a pipeline that's provisioned and not installed. We understand until we prove credibility, we need to provide these kinds of metrics for the next couple of quarters. At that point, we should step back and revert back to the much more long-term, multi-year metrics that we have historically given.
Sure.
To remind investors, we do not give quarterly or annual guidance. What we have said is we will grow our aggregate revenues at 6%- 8%, and we will deliver about 200 basis points a year of margin expansion. That is meant to be a multi-year guide, not a specific product at a specific quarter. Now, to refresh investors' memories, from 2005- 2020, with no acquisitions, Cogent Communications Holdings Inc. actually grew its revenues organically at 10.2% with 220 basis points a year of margin expansion. Within that, there was some variability. It was not a spreadsheet with exactly the same numbers. When the pandemic hit, our growth rate slowed to about 5%, and our rate of margin expansion decelerated to 100 basis points. When we acquired the Sprint Global Markets business, our revenue rate of growth turned negative. Our margins actually compressed. They stepped down.
If you looked at the underlying EBITDA without the help of the subsidy payments from T-Mobile, our quarterly EBITDA went from $60 million to $5 million. To offset that, we got a subsidy payment from T-Mobile of roughly $29 million a month or $87 million a quarter for one year. We grew that underlying EBITDA about $20 million during that period. We went from $5 million to $25 million, and that subsidy payment stepped down to approximately a $25 million a quarter number. We have continued to grow underlying EBITDA, and we have a smaller subsidy. Last quarter, we reported $73.5 million of EBITDA, inclusive of the $25 million payment.
Yep. One more question on waves. I think a couple of quarters ago, you said you could get to sort of a 10,000 backlog number. You sold 1,500, I think, in Q2. You're at 4,700 total. How are sales progressing that give you confidence to get to that 10,000 number exiting this year?
We have actually been pleasantly surprised at the customer receptiveness to a new entrant in the market. We have a large existing sales force and we have credibility with most of these customers as we are their transfer provider. We are confident we will get to a backlog of about 10,000 by year-end. That is actually a stronger backlog than we expected, and we have done it with less discounting than we expected to use.
OK. Switch gears a little bit. Talk about sort of the data center monetization opportunity. We've talked to a lot of companies today. A lot of companies seem to have sort of 5 MW of power in central offices or maybe smaller data center facilities. You sounded sort of incrementally more cautious on getting a sale done in your Q2 call. How would you sort of characterize where you are in that process and whether or not we should expect some sort of transaction to occur this year?
Those are two very different questions. In terms of where we're at in the data center conversion, we are substantially complete. We have tried to caution investors that we may not ever monetize these. I don't think that is likely, but that is a possibility, too. We have never done this before. We have now interacted with over 160 counterparties. There are over 60 counterparties that are still doing work. We have received six letters of intent for this footprint. These facilities were a subset of former central offices. We picked the larger ones with the denser power. We had to go in and clear out the old telephone equipment, interconnect them to the market, convert them from DC to AC. All of that work is done. Now we're in the process of finding the right counterparty that is willing to put risk capital on.
What does that look like from your perspective? Because you see these big, chunky deals from a data center standpoint, 250 MW, right? You don't see a lot of just, hey, we need 5 MW or 8 MW, with the type of power that you have.
It's interesting. If you look at Digital Realty, which operates 300 centers and is focused on hyperscale builds, they actually said in the last quarter, the 0- 1 MW was the strongest demand set they had. An advantage of this footprint is it's very distributed, and it has some expansion capability. What it is not, it is not competing with a 900-acre campus in rural Georgia that's going to build a 3 GW campus of buildings.
What Digital has, though, is assets sort of positioned in the right markets. These assets are a little bit spread out, to your point, on distributed. I mean, is there what do you think the customer set looks like? Who is the six companies that you're sort of negotiating with? Help us understand what you're thinking.
I would argue that we're in all of the key markets, and we are interconnected to the key aggregation sites. For example, if you went to Chicago, we're 3 mi south of Midway. We are directly connected to 350 Cermak with dark fiber. If you went to Dallas, we're literally off of 35 mi in Fort Worth, connected to 1950 Stemmons. If you went to Anaheim, we are directly connected to One Wilshire. If you went to Merchantville in Jersey, we're directly connected to 401 Locust. We are connected to all of the major aggregation points. I think there are three use cases for these facilities. One is as just small colo facilities where existing data center operators need to fill in their footprint.
They may be in some markets that we're not in, and then conversely, we're in markets they're not in, and they want to add this to the footprint. The second group tend to be more portfolio-wide deals in which private equity are looking for a vehicle to build a data center business on. The challenge is none of these have revenue today. They're empty facilities, and that's problematic for a PE shop because they're not in the business of buying assets. They're in the business of buying businesses. While they will pay a very high multiple, they need to see some operating revenue, even if it's cash flow negative, with the idea they could scale it.
The third and maybe most interesting use case are the guys that are trying to get ahead of the curve on AI and view these types of edgier facilities as available for inference as opposed to training, which needs to have much lower latency, needs to be much more distributed, and tend to be much less power intensive.
Got it. Let's switch gears. I want to talk a little bit more about the core business, right? You guided us to a 6% -8% long-term growth rate. I think one could probably assume waves contribute of that growth rate a good amount. Maybe can you unpack where the core business is from a corporate and a net-centric standpoint?
An enterprise.
From an enterprise standpoint, try to separate it from the decline from the T-Mobile business?
Yeah. When we acquired the T-Mobile enterprise business, Sprint GMG , it was declining at about 10.5% a year. That's right off of their public filings for the three previous years. We actually accelerated the rate of decline in that business to over 20% a year by intentionally terminating certain products and grooming certain locations. 80% of that business was enterprise. Our enterprise business today represents about 13% of Cogent's aggregate revenues, and it is a declining business. We've been able to moderate that rate of decline to low single-digits and should get it to being a completely stable business. About 17% of the revenue that we acquired got classified in the corporate. They were the same type of customers. They were just smaller. They didn't fit the $5 billion revenue or multinational definition of an enterprise.
Those continued to decline while our underlying corporate business was growing at low single-digits.
Is corporate low single-digits sort of continuing into Q2, and that'd be your expectation for this year?
The on-net portion, yes. The off-net portion is still being groomed from some of those acquired customers.
OK. We're sort of getting past that, right?
That's correct. That is probably pretty close to revenue growth, neutral to slightly positive at this point, and should return to being a low single-digit grower.
The off-net corporate piece.
Yes.
Got it.
Finally, the net-centric business, which does include waves, does include IPv4 leasing addresses. That business is today growing in the high single-digits.
What about excluding waves and sort of IPv4?
Low single-digits.
OK, that has been sort of a deceleration then.
It has been a deceleration. Part of it has been traffic deceleration. Part of it has been location of purchases, and then finally, concentration of customers. In our transit business, prices are lower in Western Europe and North America than they are the rest of the world. The bigger the customer, the lower the price per meg. There's oftentimes some variability quarter over quarter. Are you signing more small deals in Asia-Pac or Africa or LatAm? That pulls ARPUs up versus doing a really big hyperscaler deal in the U.S.
When you think about net-centric, waves, IPv4, growing low single-digits, I guess volumes in terms of traffic is what, upper single-digits? Pricing decline has historically been 20%. How do you think about growing that business going forward?
Over the past 20 years of Cogent being public, that net-centric business without waves, without IP addresses, has grown at an average of 9%. It has been exceedingly volatile.
Yeah.
The growth rates have been as high as + 26% year -over- year and as bad as - 24% year- over- year, much more volatile than the corporate business. I think that 9% long-term growth rate in the net-centric transit business is the right way to think about that business.
OK. In the near term, barring any sort of big inflection in traffic, we would expect it to be sort of low single-digits?
Yeah, we're below the long-term trend.
OK.
Coming off of three years where we were substantially above trend.
Last question, and we have like one minute left. You had previously talked about getting to about $350 million in EBITDA for this year. You're tracking below that. How do you expect EBITDA to sort of trend sequentially as we exit this year?
It will improve sequentially each quarter through both a combination of the remaining cost-cutting initiatives as well as the ramp in high margin sales. We will probably do less than that number. Just to remind you, when we acquired Sprint, we were doing $260 million of EBITDA. The first year out of the box, calendar year, we did $352 million. That included the subsidy payments that more than compensated for the negative EBITDA from T-Mobile acquired business. The second year, no one believed we would replicate that. We had a $104 million headwind, and we, in fact, overcame that headwind, and we did $348 million. The following year, we have a similar headwind just based on the deceleration of the subsidies. We're overcoming most of it, but probably will not overcome all of it.
Got it. With that, we're out of time. Dave, thank you so much for being here.
Thank you.
Thanks, everybody, for your attention.
Thank you for coming to
I'd like to thank KeyBank for a great venue. Most importantly, thank investors for taking time out of their day.
Let's start on waves, Dave. I think investors have high hopes for this business that you're starting up. It's taken a little bit longer to ramp. What have been the factors from your perspective that caused this business to ramp more slowly than expectations?
I'll actually break that into three components. We always knew that it was going to take us about a year- and- a- half to take the former Sprint voice network, interconnect it to Cogent's existing metro network, and reconfigure it. We actually slightly beat that schedule in terms of the number of sites and the speed at which we did that interconnection. The second piece has been the ability of us to build a funnel showing that there's adequate demand out there. I think there is a great deal of belief among investors that the wavelength market is real, that it is growing, and that AI has been an incremental use case that has driven more wavelength demand. It is a market that is highly concentrated with two players controlling the bulk of the market, and then a handful of smaller regional players, the remainder of the market.
What we needed to do was three things. One, demonstrate that people would be interested in Cogent and do that before we really had a network. We built a funnel of almost 10,000, what I would call expressions of interest. About 1,000 of those turned into orders. About 1,000 of those pushed forward and now have been installed. About 8,000 of those early expressions that sat around for a year, year- and- a- half fell out. As we began to give customers specific implementation schedules, we started to build a much larger and more certain funnel. We ended the last quarter with a strong funnel.
We had on earnings day about 4,700 orders that included orders that have been installed but not accepted by customers, orders that are in provisioning, orders that are signed but not yet provisioned, and finally, orders that the specific route and locations and price have been agreed to, but a contract has not been signed.
Real quick, on the 8,000 that fell out, how do you get confident that those didn't go to one of your competitors?
I think they did. I think they stayed with our competitors because of our inability to install them. In that period from deal closing in May of 2023 through the end of 2024, we told customers we were actively working on putting wavelength services into a defined list of 800 data centers. We even gave the customers the list. We also said we would take an order, but we had no ability to tell the customer when we would install it. We would take the order. You should continue to look elsewhere. If we can wave-enable your two endpoints, we'll contact you, and we will try to manually install that order by engineering that path. We did that for about 1,000 waves. The process that we have going forward is a much more automated process.
What we did is pre-enabled now 938 data centers where we could provision a wave from any data center to any data center in 30 days, and we can do that in one of three interface speeds: 10G, 100G, or 400G. That is radically different than how the market works today. The private line market, which is really what wavelengths are dedicated in, had been around for nearly 80 years. For the first roughly 55 of those years, you had to buy them from AT&T and then a little bit MCI. The market changed in the late 1990s, and third parties like ourselves and Level 3 and BroadWing and Williams built their own networks. We chose not to build a wavelength network. We built only an IP network. Others built networks and sold waves. Waves were always sold as surplus capacity.
Kind of think of it as the airline model where you want to sell full price tickets, but you'll go to Priceline.com and give them your surplus. That market developed into a fairly robust $2 billion market. We realized that we had some architectural advantages that would allow us to capture share. We could go to more points. We could build a network without surplus capacity, but rather one that's designed for quick provisioning, one along unique right-of-ways, one that would be more reliable, and one that was less expensive. I think our approach to this market is really very different than the two major competitors.
When we think about sort of the 4,700 units in your backlog, are those now sort of signed contracts in backlog? Are those still indications of interest on orders?
There are four categories. There are waves that have been signed contracts, been installed, but customers not accepted.
Yeah, you mentioned that those are.
Those stood between $200 and $300.
$200 and $300 that should have been installed last quarter, right?
Should have been counted in install.
OK.
Because the customer wasn't ready, we didn't start billing them. We only recognize revenue when the customer accepts. We had the same issue in Q1 where our unit installs were far greater than our revenue install. The second bucket are orders that are signed that we're still doing work on. We're still provisioning.
How big would you say that bucket is?
A few hundred.
A few hundred.
Third, there is a bucket of signed orders that have not yet been provisioned.
Sure.
In large part because the customer has told us when we contacted them, we're not ready yet. We still want it, but we're not ready. We have a signed order, and we're ready to go. The final bucket is those orders that the customers agreed on the exact route, the exact interface, the exact speed, and the price.
But.
All of the key terms.
Not a signed.
It has not signed.
What is signed contracted at this point, would you say, in the seven?
Probably close to half of that.
Close to half. How do you sort of bridge us from sort of getting to your target, exiting this year?