All right. Good afternoon, everyone, and welcome to TD Cowen's 10th Annual Communications Infrastructure Summit. My name is Anton Rinnert, and I'm an Associate on the Cable and Telecom team at TD Cowen. I'm joined here today with Dave Schaeffer, founder and CEO of Cogent Communications. We'll have 40 minutes for the discussion today, and with that, let's jump into it. Dave, welcome, and thank you for taking the time to join us.
Well, Anton, thank you for hosting me. Always thank TD for a great venue, and I want to thank the investors who stuck around at the end of a rainy day-
Thank you.
to hear a little bit about Cogent.
Yeah. All right. Well, so starting with GenAI, it's been a hot topic recently. So with... We saw the fiber contract wins at $5 billion levels, at one of your competitors. Let's just start with maybe your general thoughts on these deals. And, you know, to be fair, economical for one company can be uneconomical for others, depending on the capital and where your network resides. Is that fair?
Yeah, it is. So let's maybe back up and look at the demand driven by AI, which is real. There is incremental use cases to marry remote computing power for large language training models to large datasets, and those require significant amounts of bandwidth. We also benefit from the fact that we are primarily an Internet Service Provider, and the Internet has really been responsible for the collection of the data that is the raw material for those training models to run off of, and the low cost of data collection is what has really enabled AI to take off. Now, because of power constraints, oftentimes, the data centers that will do the heavy computing associated with those training models are not adjacent to where the data sits, and for that reason, there's a need for transport.
You know, that transport can occur either over the public internet or through dedicated connections. Those dedicated connections can actually come in three flavors. You can have wavelengths, which are the least expensive and more flexible. You can purchase dark fiber from someone who already has inventory, or you can go one step further and build your own dark fiber. Some hyperscalers have gone to that final build stage, but it's been mostly for subsea applications and not terrestrial. You know, we looked at a number of the hyperscale generative AI bids and chose not to participate, and our reasons were threefold. One, it requires significant capital to build out to brand-new facilities where there would be only one customer, and that had too much market risk and too low a rate of return.
Two, while we have substantial dark fiber inventory that we acquired from the acquisition of the Sprint GMG business from T-Mobile, we have not been very active in selling it to date because we've focused all of our resources on enabling wave services over that network. And then finally, we did not have spare conduit that we could use to pull new fibers. So I think in that deal that was announced, there are really three separate deals that comprise that $5 billion number.
Okay, and as we kind of think about the industry opportunity as a whole, I mean, how many $5 billion deals do you, do you think there are? How many new data centers are, are being created?
So two very different questions. There is, today, almost a land rush for data centers. The constraint is available power. Typically, in the developed world, the markets have been accustomed to about a 1% per annum increase in power consumption. With the application of AI, we've seen a step function, so data centers today consume between 2%-3% of the world's electric power, and the current demand would point to something more like 5%-6%. To build the generative capacity, not generative AI, but Power Generation-
Mm-hmm
... has long lead times and significant environmental impact. To build the transmission lines, to go from the generating plants to where the data centers also have long lead times. So what is happening today is a desire by these computing locations to use power that is in surplus in existing markets, which means that there needs to be new fiber built to those facilities. We typically add about 100 data centers a year to the Cogent footprint. We have 1,688 carrier-neutral data centers in 54 countries attached directly to our network.... This year, it seems like the pace of new construction has accelerated, and we'll probably do North of 120 or 130 data centers. And that's on top of these private data centers that are being built.
And then in terms of the bandwidth demand to link those centers together, I think that is real. It is an incremental application, and it will drive more bandwidth demand for both wavelengths and dark fiber.
Okay. I guess as we kind of think about dark fiber, I mean, you mentioned some of the challenging economics, given it's construction intensive. Is it safe to say the opportunity there then is far smaller than the opportunity for you in waves?
I think in our priorities, one, we're gonna continue to grow our Internet-based business.
Right
... which today represents 90% of our revenue. That is either in the form of IP Transit services, DIA or various types of VPN technologies, VPLS, MPLS. Secondly, we will add additional transport services, initially focusing on wavelengths. So we have been in the process of enabling 800 carrier-neutral data centers across North America, where we can deliver a 10 gig, 100 gig, or 400 gig wavelength. Today in the industry, those wavelength deployments generally take 3 months-4 months and require a significant amount of custom engineering. What we are doing is standardizing an architecture that will allow us to deliver a wave from any data center to any data center in 2 weeks. I think that will differentiate us, along with the unique routes and the aggressive pricing that we bring to market.
Once we have completed that wave enablement, there will be surplus excess dark fiber in the former Sprint network, and we fully intend to sell off some of that fiber. Now, there may be a need to construct extensions, and I think that's a business we're probably gonna leave to others, just based on the very low returns, and we have better uses of capital, including giving it back to shareholders.
Got it. And you kind of started going into the waves, and that brings us to our next topic. You're targeting 800 wave-ready data centers by the end of the year. And I believe as of last earnings call, you were at 574, but-
554 .
554, sorry. These still have longer provisioning times, correct? So, you're selling to 156,
Correct.
Are these with quick provisioning times then?
No, these are much slower.
These are also slower. Okay.
Yeah. So today, if we need to provision a wave, it would look much like the same process our competitors use. That means custom engineering on the route and will generally mean about 6 field dispatches and a fairly lengthy install window. Once we have completed this rearchitecture, because we had the flexibility of starting with a clean sheet of paper, we are effectively taking a network that was built for long distance voice, never contemplated wavelengths, didn't even contemplate the internet when it was built. It has been sitting dormant for a decade. We are stripping out the old phone equipment. We're extending it to connect to our metro networks, and then we're architecting it to be optimal for wavelengths. By the end of the year, we can eat into the backlog we have, plus we can sell new orders with a 2-week install window.
Got it, and that kind of brings us to the next question is the revenue run rate. So at the moment, you're at, I believe, less than $15 million annualized waves revenue run rate, but you're targeting $300 million by 2026, correct? So this means you'll have to accelerate pretty quickly once you reach the 800 by the end of the year. Are we gonna see kind of a step function upwards in early 2025, or what does this kind of look like?
So you will see some progress between now and the end of the year, based on certain customers needing us to do these installations the old-fashioned, custom way. We have a funnel of about 2,700 waves. The ARPU on those waves is around $1,670. So if we convert all of that funnel, and some of it will not convert, but if we just converted that funnel, coupled with our current run rate, it would get us to about a $70 million runway. As we are able to install services, we will demonstrate to customers that we can in fact achieve these better provisioning windows along unique routes, and we expect our sales cadence to significantly accelerate.
So I believe our long-term target is to get to $500 million against a $2 billion addressable market by mid-year 2028, May of 2028, which was 5 years from our transaction closing. We are very comfortable with that target. We think that the demonstrated funnel that we've built when we couldn't provision in anything different than our competitors, really demonstrates the depth of demand for these services.
Yeah. And so, I mean, what gives you kind of the confidence in the $500 million number? I mean, you mentioned a $2 billion opportunity. Are you just kind of taking a percentage of the market that you think you can achieve, or is there kind of more detailed math that goes into it? How are you kind of thinking about the $500 million?
We thought about the market both on a top-down and bottoms-up approach.
Right.
Let's start with the top-down, because it's the easiest for a generalist to understand. There is a $7 billion global optical transport market. About $3.5 billion of that is in North America, and about $2 billion of that is intercity, and about $1.5 billion is intra- city or metro wave. We are primarily focused on that intercity component. We may include a small amount of intra- city as part of a total solution, but we're primarily an intercity carrier. In the transit business, we have become the largest player in the world, carrying a quarter of the world's traffic, and it took us 17 years to do that, but it was a much more competitive market. We had no brand recognition, we had no pre-existing customer relationships, and we had to build a sales force.
Today, we have 280 salespeople that focus on the wholesale market, and they will be able to go to their existing customer base and add on this wave product. So the idea was, with the unique advantages we bring to the table, we should be able to get to the same kind of market share and do it in about a third the time. So that's the top-down view. Let's start with the bottom up. We took each of those reps, we canvassed them against their existing deck of accounts, said: How many of these accounts are potential wave customers? How many waves do they currently buy from others? What percentage of their business could we win, and how many waves could we in fact sell each month? On that bottoms-up approach, we actually came up with a little higher number than $500 million.
We looked at our gain in market share, both on a top-down and bottoms-up view.
Got it. And you kind of touched on your key differentiation to, I guess, succeed in the market with your kind of unique routes, time to provision, lower price. Do you anticipate the competition to play defense? And what do you expect to kind of see on that end?
Well, I think we've already seen a little bit of defensive action in that dark fiber sale that you mentioned at the beginning of our discussion. You know, for 20 years, that particular vendor had been reticent to sell dark fiber. In fact, had regretted the sale they had made to us 25 years ago. That's clearly changed. I think it's a recognition that the wavelength market is going to become more competitive. It is a big market. There's room for multiple players, but I think the ubiquity of our footprint, the pricing that we can bring, give us some real advantages. Remember, 3/4 of the wave purchasers are already buying internet access from Cogent today. The other 25% of the market, that's not an internet customer of Cogent, does in fact have a knowledge of who we are and has interacted with our sales force.
So it's very different than going out and saying, "Cogent who? Who's my salesperson? What are you calling me about?" And now it's, "We know that you have wavelengths, and we have to win that business on a route-by-route, wave-by-wave basis." So I don't think this is a market share gain by big global deals, but rather proving in the value on each and every link that we sell. And it's simple; if we can deliver it faster, at a better location, with greater reliability, and do it at a lower price point, we will win share.
Yeah. And you mentioned your backlog earlier. Is there kind of a concern that some of the backlog might shrink and maybe some of the potential connections go to your customer competitors instead?
Oh, absolutely. So we have been clear with these customers that we are indeterminate in how quickly we could provision these waves. We only have the 554 facilities out of 800, and even in those facilities, it could be 3 months or 4 months to get a wave installed, and it is not our goal to do these on a one-off basis, so we're going to encourage you to wait till year-end. We have allowed certain customers to get out of contracts. I suspect that will continue, but it's been fairly de minimis. On the flip side, we're continuing to see new customers sign contracts with us, and I think the real inflection will come as we demonstrate credibility and scale, that customers can get confidence in our ability to deliver and the reliability of the service.
Okay. And is it possible that the waves business could maybe cannibalize some of your IP transit business or potentially the other way around? Or maybe on the other hand, could there be synergies, such as, like, selling customers both?
So if customers were perfectly rational and did not care about control or these other attributes of latency and packet size, and were just looking at the lowest cost, easiest way to move bits, there would be no wavelength market. But we know from 25 years of history that customers do value these attributes. We have chosen, excuse me, not to participate in that market because of the way in which our IP network was built. When we had the opportunity to acquire the Sprint network with all of that surplus fiber, the idea was to take a pair of fibers and build a wave-optimized network, just like we did in the IP-optimized architecture of our network. I think they are complementary. There is a place for both services. I'm sure there will be situations where waves cannibalize transit, and there'll be other situations where transit cannibalizes waves.
In general, Cogent has developed a reputation of being the most interconnected network in the world, and a transit product is the combination of transport, i.e., wavelength, routing, and interconnection to other networks. We have more data centers around the world on net than anyone, by several hundred facilities. We have more networks connected to us, and we sell at the absolute lowest price in the market. I think that is going to continue. That business continues to grow in the kind of 20% volume range, prices decline, and we continue to grow revenues. I think wavelengths will complement that.
All right, well, maybe let's move on to a different topic, maybe CapEx. It came in a little bit higher in the second quarter at $49 million, but I mean, understandably, you are still working on the integration and you accelerated some of the data center conversion efforts. But how should we think about CapEx going forward and the overall outlook there? What do you consider maybe normal CapEx levels, and when do we get there?
Yes. So let's start with the two businesses prior to the acquisition. Cogent was spending about $35 million a year in maintenance CapEx. That includes regularly augmenting capacity and deploying new routers and transport equipment. That is going to continue. We also spent about $30 million a year extending the network into new locations, whether it be multi-tenant office buildings or data centers. That will continue as well. When we acquired Sprint, there was about $30 million a year being spent on maintenance CapEx for that business, primarily maintaining the physical plant of the network, whether it be the fiber itself or the buildings. Even though they were effectively dormant, with only 7% of the revenues touching the Sprint network, those assets still needed to be maintained. Now we're using them, they will still need maintenance, and we'll spend about $30 million.
So in total, about $100 million of long-term run rate CapEx. On top of that, we will spend about $40 million a year in principal payments on capital leases. From an investor's point of view, that's cash out the door, just like CapEx.
Mm-hmm.
It's just treated as a finance lease on the balance sheet rather than a straight capital purchase. Then, in addition to that, we are spending about $50 million one-time to conjoin the two networks, to bring them together. We spent about $30 million of that in 2023. We'll spend about $20 million of that in 2024, and that will be over. The final piece of the puzzle is the data center enhancement. When we acquired Sprint, there were 482 fee simple owned technical buildings. They comprised 1.9 million sq ft and 230 MW of power.
We went through that inventory and concluded 45 of those facilities were suitable for data center conversion, meaning we could clear out the old phone switches, we could connect them to the metro markets, and we could put inverters in to take the negative 48 DC power and convert it to AC. We were embarking on that strategy to supplement the Cogent pre-existing data center footprint. So prior to the acquisition, Cogent had 55 data centers, 634,000 sq ft of raised floor and 69 MW. Of those facilities, 2 were fee simple owned and 53 were in leaseholds. Our plan was to end up with 100 kind of similar-looking data centers, all about 10,000 sq ft-12,000 sq ft, with roughly 1 MW per facility.
We realized that we had an additional 1 million sq ft, 108 MW of inbound power, and 88 MW of conditioned power that we were not going to use as a business. With the accelerated demand for power because of generative AI, as your opening question asked, we have decided to invest money to make those facilities marketable. We are rapidly doing that. We are taking those facilities and offering them to the market in three models: You can lease the facility for $1 million a megawatt a year on a triple net basis. Leases can be anywhere from 3 years to 20 years in duration with CPI or escalators. You can lease the facility at the same rate with an option to buy, or you could just outright buy the facility for $10 million a megawatt.
We went to market the last week in April to market these facilities. We went to 115 counterparties. That list has grown to 133. We are still in active discussion with 58 of them, and we have multiple LOIs. We have not consummated any deals. None of these facilities are totally ready to go, but we think there is a significant amount of value that could be harvested from this pre-existing surplus power footprint.
Okay, and so how far along are these discussions? Are they still preliminary, or they're pretty far along now, you would say?
It varies by counterparty-
Of course.
-and by location. We're in various stages with all of them. There are no deals ready to announce publicly.
Yep.
None of the facilities are today ready to be occupiable. We are still in the final stages of that repositioning. About 10 of the 21 facilities with surplus power will be occupiable by year-end.
Got it. And then, you know, I wanted to kind of touch on the, the cash, on your balance sheet. I mean, you, you did a $206 million ABS raise, a few months ago, and you did a another $300 million debt raise. And then you paid down the, the unfavorable lease. But you still have a good amount of cash left on the balance sheet. Was this the plan from the start, or, what are your plans kind of for the idle cash on your books?
So we have $462 million at quarter end. We initially looked to tap the high yield market last fall, when we began contemplating buying out of the unfavorable IRU that we acquired from Lumen, from Sprint, and our ability to buy that out at a 12% discount seemed attractive. However, we could not raise enough money in the ABS market, so the high-yield market was closed. We went to the ABS market, and we did something that no one had ever done before, which is we went out and securitized rental income from IPv4 address. Totally new asset class for a new issuer. We did raise $206 million. We were uncomfortable using all of that to retire that IRU, so what we did was a second high-yield deal when the market opened.
We bought out the lease for $114 million. It resulted in a roughly $15 million savings to Cogent, and it also was foundational to allow us to exit out of our Lumen IRU that we need to exit as part of our targeted $220 million in cost savings. You know, in terms of the excess cash, you know, we understand our net leverage is going to go up over the next year because of the decline in subsidy payments from T-Mobile. Prior to announcing the deal, Cogent's leverage had ticked up from 3.5% to 4.7 x levered on a net basis at the end of 2022. With the transaction, we have de-levered down to 3.14 x on a net basis.
We will most likely see our leverage tick back up to about 4.2 x as a result of those smaller subsidy payments from T-Mobile, and then we will gradually de-lever again. We are committed to returning capital to shareholders. We have 48 sequential consecutive quarters of growing our dividend. We intend to continue to be able to do that, and in addition to that, last quarter, we took advantage of some market volatility, and we bought back 153,000 shares. In total, we have bought back 10.6 million shares at a cost of about $230 million.
Got it. And as we think about kind of the data center sale proceeds, if you were to sell all 100 MW of your excess data center capacity, that could be $1 billion in proceeds. Does your thinking kind of change in terms of how you're going to use those proceeds? Could you maybe be doing more buybacks? What are your thoughts there?
We actually have three assets that are not baked into our operating business plan-
Right
... The excess data center space, the excess IP space that we control, and the dark fiber that we touched on earlier in the discussion. Our goal is to monetize all of these when appropriate. You know, we will be a cash taxpayer on that gain because we have no basis in these assets. You know, that excess cash could be used to retire debt, it could be used to accelerate returns to equity, which would be buybacks or a faster rate of dividend growth. We would evaluate all of those. Currently, our debt does not trade at a significant discount, so therefore it would not make sense at the low coupon rates that we have to buy out the debt. So I think we would wait until our next maturity tranche in 2026. We may just hold the cash on the balance sheet. We may do a buyback.
It's really highly dependent on market conditions. These are good problems to have.
Yeah, that's true. And so I guess as we're talking about your other assets, your IPv4 addresses, you've... You're leasing 12.8 million of them. You've securitized most of the leased ones, but you have a total of 38 million. So you mentioned maybe potentially selling some of these as the market improves. Do you anticipate the market value to kind of get better in the short term, long term, given the scarcity or ongoing demand?
So IPv4 is a finite resource. There were 4.3 billion addresses created at the beginning of the DARPA net, that became the ARPANET, that became the Internet. There are no more that could be created. The Internet moved to a second protocol, IPv6, which has 2 to the 128th power addresses, but has never been widely adopted, and only accounts for about 7% of traffic. So as a result, the average price per address in the past 13 years has went from $4 an address up to $60. It has pulled back some to about $50 today. The two largest buyers had been Amazon and Microsoft. They are both now leasing out their inventory at $3.60 per address per month. We began leasing at roughly $0.30 a month.
We've recently raised our prices to $0.51 for new leases. We're going to evaluate raising the installed base to a comparable price and then may go back and raise prices again. We also have accelerated our leasing activity. So our view of IPv4 monetization is three-pronged. It is, one, continue to lease at a higher rate.
Mm-hmm.
Two, potentially an additional securitization of more leased revenue. And then third, as the market reaches kind of an optimal point, sell that excess inventory. And we're open to all three.
Okay. You mentioned kind of experimenting with raising the prices on the install base. These trials haven't started yet, correct?
For the new sales, they did. It began April 1st.
Okay.
We ran that trial between April 1st and August 1st. We saw no material diminution in the rate of sales. With that information, we have notified customers that starting September 1st, 25% of the installed base will be increased to $0.51 from the $0.29 average. We will then measure for several months the impact on churn. The churn on these addresses is very low. It is about 0.8% per month or per year, versus 1.2% per month for bandwidth sales, so about 15 x stickier. If we see the address space remain much stickier than bandwidth, we will raise the entire base. That'll probably happen by year-end, and then we'll evaluate, should we further increase the new sales, given the fact that Amazon and Microsoft have created a pricing umbrella that's so far above our price point?
You know, I am kind of curious why the discussions around these addresses, they've kind of only begun fairly recently. But you've had them for decades, and I think before the past year, IPv4 wasn't really mentioned at Cogent very often, or at all. What kind of-
No one asked. You know, we had been marketing these addresses on leases since 2015. Actually, we got one question in 10-year period-
Okay
... in 2012, when someone asked us, and at the time we said, "We are sitting on a large inventory, and we'll use it at some point in the future." And then for the next 12 years, no one mentioned it. One of our research analysts covered Amazon, found out they were leasing, went to the registry, saw Cogent's inventory, and called us up. We said: Of course we are. And then, at the same time, we had a number of banks, including your bank, reach out to us to do an asset-backed securitization of our fiber network. We concluded that that was not practical, based on the fact that that network spanned 12,000 unique jurisdictions. It would just be too hard to perfect the security interest and transfer those assets into a special purpose vehicle. So we went to a number of banks.
Six banks had contacted us for ABS, and we said, "We got this thing called addresses." Four of the six banks had no idea what we were talking about and said-
Mm
... "We don't get it. We're not interested." Two did say: Yeah, this is a very securitizable stream of revenue. They did the work. We went to market with that and had a very well-received inaugural offering. Obviously, once we securitize them, we did have to have a little more discussion about them because we have a new base of investors whose only exposure to Cogent is the notes that are backed by these addresses.
Got it. All right, and with that, we're out of time. Thank you so much, Dave.
Hey, thanks, Anton.
Thank you.
Thank you all very much.