All right, let's welcome to the afternoon sessions of day one of our TD Cowen 52nd TMT conference. My name is Greg Williams. I cover cable, wireless, and telco here at TD Cowen. I'm joined in this session by Dave Schaeffer, the founder and CEO of Cogent. So let's get started. Dave, thanks for joining us.
Hey, Greg, thank you for hosting me. I'd like to thank the investors for taking time out of the day, and thank TD Cowen for a great venue.
Sure, great. So let's talk about the Wave business. You targeted 800 Wave-ready data centers by the end of the year, but as of last earnings, you're just at 104. So you still feel confident in hitting 800 by year-end, but what gives you this confidence? And, you know, the path from 800 from here, will it be linear or ramp more in Q3, Q4?
So for clarification, Greg, we're actually Wave-enabled in 419 data centers as of end of Q1, and we had actually provisioned and sold Wave in 104 of those. In making a facility Wave-enabled, the easy part is putting the transponder shelf in to accept the Wave. But if that is all we did, the provisioning window would be comparable to that of our competitors. So what Cogent is doing is optimizing the Sprint network, which was originally designed to carry long-distance voice, to sell wavelengths. In order to make that happen, there are four discrete steps that have to happen. The first is we have to physically connect the Sprint network to our metro footprint, and we've done that in over 100 locations across North America, and that process is complete at this point.
That was actually probably the riskiest of the processes because there were a few instances where we actually had to do greenfield builds and permit. The second effort is to deploy a transponder shelf in each of the targeted 800 data centers. As I said, we've got roughly 420 of those done at the end of the quarter, and feel comfortable that'll be complete by year-end. That transponder shelf is a four RU unit. It typically fits in an existing rack. There may be a handful of sites where we don't have four RUs out of a 42 RU rack and therefore need a second rack, but that should be straightforward. The third effort is also pretty straightforward, and that's deployment of a multi-dimensional reconfigurable add-drop multiplexer, or a ROADM.
That is at the intersection of the long-haul network and the metro network. The number of dimensions is dependent on how many physical paths long-distance fiber is coming in. In the simplest case, it's one direction, in the most complex, it's six or seven. So you need an eight-dimensional ROADM if you have seven inbound paths and one connection to the metro. Think of that ROADM as a traffic cop to direct the wavelengths to the correct data center.
Sure.
Then the most difficult part of the process is only about 40% complete at this time. Cogent's metropolitan network in North America is comprised of 14,000 route miles of metropolitan fiber configured in 800 physical rings. Attached to those rings are 1,860 multi-tenant office buildings and 800 data centers. Those rings commingle both data centers and multi-tenant office buildings. For an IP network, that is the optimal design. Because we do not intend to sell wavelengths into multi-tenant office buildings, we need to segregate the multi-tenant office buildings from the data centers. That will result in 800 rings at the end, 14,000 route miles, but roughly 250 rings containing just data centers, 550 rings containing multi-tenant office buildings.
That reconfiguration work requires us to do multiple site visits to each endpoint and do that reconfiguration. Once these four foundational steps are in place, we will be in a position to provision a wavelength in less than two weeks, and in any-t o- any scenario, going from any data center to any data center.
Do you think that you'll get to the 800 on the linear ramp, or do you think this is gonna be... It sounds like it's a lot of work, I mean, whether it's the Sprint network or the transponders or the multiplexers or even these rings. You know, maybe I'm trying to—I'm just trying to figure out, you know, which of these steps is sort of the bottleneck, and when, you know, how should the cadence of this build look?
So all the steps are happening in concurrence. There are nearly 10,000 discrete work projects, and nearly 1,000 of our 1,900 employees are involved in these reconfigurations. It will all be complete by year-end. We are able to provision wavelengths today, but each wavelength takes custom engineering and will take 90 to 120 days. Each of these custom designs slows our reconfiguration work down. So while we've installed about just under 800 wavelengths, and we have about 2,400 in the backlog, we are trying to focus on doing this foundational work, and therefore, be in a position to clear the backlog and accelerate sales early next year. So I think the way to think about the cadence is slow between now and the end of the year, some incremental proof points, but then a material acceleration in our wave deployment.
Right. And you sort of answered my second question now is, when you build it, then you can finally sell it. And so we've talked about revenue, which is top of mind, 'cause you've got about $3 million per quarter of wave revenue right now. You rescinded the $20 million run rate from you mentioning a few months ago. And, you know, I guess when could we reach a $20 million run rate or any other target you'd be comfortable giving as we think about the revenue cadence now going forward?
So we initially thought that the bulk of our wavelength orders would be in the largest data centers where we sell transit. We quickly enabled those facilities first, and we were half right. Most of our wavelength demand does have one of the ends in those large data centers, but what we have seen is virtually all of the demand into those data centers is from a smaller data center, as opposed to large to large. With that, we need to do all of this foundational work to be in a position to hit those revenue targets. I think we should be able to meet our long-term, five-year target from closing of $500 million in revenue. We were hoping it would be more linear, meaning one year from closing the transaction, that would have been May, we would have been on an annualized rate of $100 million.
Instead, we're probably closer to $15 million this May. We should catch up part of the way by next May, and by the third year, i.e., May of 2026, be back to an annualized rate of about $100 million each year, about $300 million, getting us to that $500 million number, by May of 2028.
Got it. That's super helpful. And then you mentioned the backlog. You had 2,400 wave connections backlog. How much annual revenue do you think that represents? Like, can I assume, like what, $1,600 a circuit, 2,400 wave connections, that's $40 million of revenue, roughly a little more than that. Is that the right way of thinking about it?
I think the average in the backlog is slightly higher than the average you laid out.
Okay.
In pricing a wavelength, there are three dimensions to that pricing equation. The first is the size of the wave, and waves come in three different increments: 10 gig, 100 gig, and 400 gig. The most prevalent size is 100 gig, by far and away. 400 is the rarest of the sizes today. The second dimension is distance. So different than transit, which is a distance-insensitive product, waves are sold based on the mileage a wave traverses. The longer the wave, the more expensive it is. And then the third dimension of pricing is how long of a contract the customer commits for. Most of these are two-year contracts. I think our ARPU in the funnel is slightly above the average that's installed today.
Got it. The $500 million target in a couple of years, it's obviously a big number. It shows, you know, a fair share of market stealing, for lack of a better word. And you know, I get that you can provision faster, and you have unique routes, and you know, you can even provide a pricing discount of, just call it 30%. So you know, you've got some compelling reasons why you'll win that share. But you know, the flip side of that, you know, aren't wave customers typically entrenched with Lumen or Zayo in a bigger solution, so it's hard to splice out a wave product? Does it make it a little tougher to you know, steal share? I'm just trying to think about that dynamic as well.
So I think your question is partially correct, in that there is some incumbency advantage, there is some switching cost. However, waves are typically purchased as an independent product, not part of a larger solution. There is little or no cross-selling that goes on. I think the advantages that we have of diversity, ubiquity of endpoint, speed of delivering, will help us capture market share, both by displacing existing vendors and capturing incremental waves. You know, Cogent is the largest provider of transit services in the world, and we got there because we provision quickly, and we're less expensive. I think those same two characteristics are necessary on waves. Wave prices deflate just like transit. Now, because of the route specificity, the rate of price decline is not as acute, but it does still represent deflation.
So we will capture share, in large part, as customers' unit volumes go up. Also, waves are typically purchased as part of a greater solution, not in terms of bundled services, but rather diversity. So in internet, you can buy just one connection. If there's a failure, the internet reroutes your traffic. In the case of a wavelength, it is deterministic. If there is a cut anywhere along that path, that wave is down. So for that reason, most customers buy waves in pairs.
Right, and diversify their, their network-
Absolutely
If you will. You mentioned multiple times about how you can provision faster than, you know, perhaps your competitors. But I do see Zayo and Lumen adopting network as a service solutions in a big way, talking about that. So with that, you know, I would surmise that perhaps they can catch up on this quick provisioning and take your advantage away. So maybe mention the unique advantage you have on why you can provision faster. I know you mentioned the four steps to get to that low provisioning. Where is your advantage there versus where Lumen and Zayo have not, you know, don't have that?
I've been doing this a long time, and there are a lot of buzzwords that go around. I quite honestly don't know what network as a service means, and I've yet to have someone adequately explain that to me. People are buying a specific service. They go to a location, and they want internet connectivity; that is, transit or DIA. They are buying a wavelength between two points. In order to provision that wavelength, you can pre-provision every single permutation possible. If capital was free, you may in fact do that, but that's illogical. So for 800 data centers, there are 799 factorial divided by two number of permutations. That is more permutations than there are stars in the known universe. So I don't think that kind of pre-deployment of capital makes any sense.
So then, your term, network as a service, means I've got a specific requirement, and I need to fulfill it. What Cogent has done is built an architecture that at each endpoint, a technician needs to physically be dispatched and deploy a pluggable optic, a transponder, a low-cost, can-be-white-labeled device into that shelf. Once that is in place, a person in the network operations center can build a physical path through the intermediate devices. Having a standard configuration of all those intermediate devices is optimal. There's a second level of optimization that needs to go on. So you're trying to optimize the spectral efficiency of your long-haul network. In that scenario, what you would want is every wavelength to drop at a central hub in each market. Think of it as a regional airline, where all the flights dump into the hub and then get redistributed.
On the other side, you're looking to minimize the amount of extraneous transponders. In that scenario, you would never drop and home run every wavelength into every data center. Both of those are not realistic. So what you're trying to do is come up with a topology that anticipates the most likely set of scenarios and provision them quickly. That's the closest definition I could come to network as a service, but that's far from what you think of as software as a service.
Right. Great, but so you're mentioning the topology as an advantage. Maybe I can ask it another way. I mean, what you're doing is those four steps to get to, you know, wave-ready data centers is: you've got to upgrade your Sprint network. Well, I can't, Lumen and Zayo upgrade their network. Then transponders were two. Well, why can't they put transponders in? Three was multiplexers, and four was, of course, the metro-
Sure.
How much capital are you spending, and why, why would they not, you think?
It's not a question of capital, and it's not a question of can they do it. Anybody can do it. Anybody could do what Cogent has done in transit. You need to be determined, you need to have a specific plan, and you need to be willing to walk away from other products and revenue streams.
Right.
Cogent starts with a clean sheet of paper. We have a network that was built to carry long-distance voice. It's been sitting dormant for a decade. We are taking that physical layer and configuring it optimally for wavelengths. Our competitors sell many other products over a heterogeneous network, and therefore, they do not optimize their wavelength services. Wavelengths are not the primary revenue stream at either of the companies you mentioned-
Yep
... or any of the companies we compete with. So like with any multidimensional product, the chain is no stronger than the weakest link. The weakest link is the most difficult product you have to deliver.... One of the things that differentiates Cogent in the marketplace is we are not a traditional telecom company. We sell a handful of products. We don't try to be all things to all people. In fact, we're spending money de-provisioning a bunch of services that we're losing money on, that we acquired from Sprint. You know, our competitors all talk about selling solutions and moving up the value chain. Again, I don't know what those terms mean.
Right. And, I mean, it sounds like the typical telephone narrative, they have bigger castles to defend, and so would you put that business case in front of them to possibly cannibalize larger businesses, slow down the process than you need, and by then it's too late?
I'll even say it more bluntly. You know, we are proud of the fact that we are in a dumb pipe business. Cogent's tagline is, "Smart people buy dumb pipes." Most of our competitors do not want to admit they're in a dumb pipe business, but want to tell you they're selling network as a service. They're selling value-added solutions. Every time you try to add an additional service, you add significantly to the complexity of the sales process, the provisioning process, the billing process, the network management process. Each of those dimensions of complexity increase your overall operating costs, slow down your ability to deliver service. Cogent started by selling internet. We added VPN services, colocation, IP address leasing, and now wavelengths. That's all we sell. You want something else, go somewhere else.
Sure. I want to change gears and talk about some of the asset sales that could be compelling, one being data centers. You noted Cogent has or will have 230 megawatts of data center capacity, but a surplus, I think, of 100 megawatts, which you could either lease at $1 million a megawatt per year or outright sell, I think, at 10 million megawatts-
Correct.
- per year, or $10 million per megawatt. So what needs to be done to make these facilities leasable or sellable? I mean, is that ready now, or is there another-
There's more work to be done, but we are in the market with these facilities. These were former Sprint switch sites. When we acquired Sprint, we acquired 482 fee-simple-owned buildings. Those buildings totaled 1.9 million sq ft and had 230 MW of power. Prior to the acquisition, Cogent had 55 data centers, 69 MW of power, 53 of the 55 were leaseholds, 2 were owned fee-simple. We identified 47 of the 482 buildings as suitable for data center conversion. Within those facilities, every one of those will have a 1,000 sq ft and 0.5 MW PoP room where we house our equipment. They will all have a retail data center that looks just like Cogent's existing data centers, 8,000-10,000 sq ft and about a megawatt of power.
And then in 21 of the 47 facilities, we've identified 1 million sq ft and 100 MW of excess provision power today, that we are looking to monetize through sale or wholesale leasing. In order to make these facilities leasable, we had to connect the facilities to the metro networks. That is done 'cause we had to do it for our wavelength business. Two, we had to depopulate these facilities of telephone switch gear. There were 22,500 cabinets of dead telephone switches. We've taken 16,500 out. There's still about 6,000 left that we're removing every week. We also need to check the power systems, the generators, the fire suppression systems, the perimeter security, to make sure these facilities are marketable. All that work is going on.
Then we need to take a negative 48 DC plant and invert it to AC, and we are doing that. So of these 21 facilities are in various stages of being ready. They're all far enough along. We began the marketing process about 4 weeks ago, and we continue to hold tours literally on a daily basis. We went to 116 counterparties, and we'll see what the market brings back in terms of either sale or lease opportunities.
Right. And if you sell, hypothetically, all 100 megawatts at $10 million per megawatt, that's $1 billion in value itself.
Which will be a bargain in today's data center.
That's where I was about to go, ask is, is if you're leasing at $1 million per megawatt, I'm assuming 50% margins, it's a 20x, so 20 times EBITDA multiple then. Is that the right way of thinking about it?
Well, these are triple net leases.
Okay.
The EBITDA margin will be much higher. The leases will have CPI adjusters.
Mm.
We are indifferent to whether we lease or sell the facility. These are not sale leasebacks, and it really depends on the capital structure of the potential counterparty. You know, what we have is conditioned space that is ready for someone to put servers in this year with no permitting and no additional changes from the power grid.
Sure. If you're leasing at $10 million per megawatt. That leasing price implies about $83 per kilowatt. The leasing market is around, I think, $120 per kilowatt these days. So, you know, why would these be priced so far below the market rate? Maybe it's because of the characterizations of these data centers. So maybe you can tell me, are they smaller? They're not internet facilities. Maybe they're not ideally located. I'm just curious about the leasing price that you're implying with $1 million.
It's actually all of the above, plus these are triple net, so they do pay the raw inbound power costs associated with it. But yes, we are pricing these at a substantial discount to where the market is today, realizing that this is surplus capacity that we're looking to monetize.
If you did go and lease them, I mean, there's a fear that maybe you're not gonna get credit from investors. We're gonna do, like, a sum of parts valuation, and we should put a very healthy multiple on data center lease, you know, EBITDA. So would it make sense maybe to sell the data centers? And, you know, is that run in your calculus of investors?
Absolutely.
Yeah.
Now, on the sale, we would be subject to tax. While we have a significant amount of NOLs, the majority of those NOLs are not in the US. All of the data centers are in the US, so we would pay both state and federal capital gains. We have no basis in these assets.
Right. I guess my final question then is, you know, if you have no basis and you've got to pay some taxes, what would you do with the proceeds on selling these data centers and any of the assets, whether it's dark fiber or IPv4 addresses as well?
So we have many assets that we can look to monetize. We have cash on the balance sheet. We would look to add incremental cash. We have some unfavorable IRUs that we acquired that we would look to take out at a discount. We could look to reduce our debt, we could look to enhance our dividend, or begin a buyback, or all of the above. I think, that's kind of putting the cart before the horse. I think, you know, our EBITDA is going to decline this year. You know, prior to the Sprint acquisition, we did $233 million of EBITDA in 2022. That stepped up to $352 million in 2024. Our Q1 EBITDA was $115 million. If you annualize that, it's $460 million.
We're not gonna do $460 million, because our subsidy payments from T-Mobile stepped down in June from $29.2-$8.3 million. We will maintain EBITDA relatively flat for full year, even though we're getting 7 months where we have effectively $20 million less a month coming in. This being offset by both Wavelength sales and by cost savings through our integration synergies. Next year, for full year, we'll have the lower subsidy for the full year, probably in 2025, EBITDA, roughly flat with 2024. As a result, we will see our leverage first go down and then tick back up. So, you know, perhaps raising capital through sales makes sense. Perhaps doing it just by borrowing against the cash flow. Our most recent securitization of our IPv4 was just another proof point that we have multiple accesses to capital.
Sure. And with that, we're about out of time-