Good afternoon. Welcome to day one of our 54th annual TD Cowan TMT conference. My name is Greg Williams. I cover cable, wireless, telco, and fiber here at TD Cowan. I'm delighted to be joined this session by CEO of Cogent, Dave Schaefer. Thanks for joining us.
Hey, Greg. Thanks for hosting me. Always like to thank TD for a great venue. And most importantly, I'd like to thank investors for taking time out of their busy day to hear a little bit about our business.
Sure. A lot to tackle in this conversation. But maybe we'll start off with, you know, the stock had an earnings sell-off, yet another one. But then yesterday, you posted news of the company data center's sales that drove a modest stock rally. Stock's even rallying a little bit today as well. Before we get into that, which I certainly want to immediately, but I'm curious, at this point now, with the data center sales, what are the next catalysts to get the stock working, in your view?
So, you know, Cogent is a 26-year-old company. We have been public for 21 of those 26 years. And stocks go up and down for a variety of reasons. I think there have been some overhangs that have been extraordinary pressures on our stock that probably have decoupled our performance from our stock. The business has been more stable than the stock has been. I think it started with my four sales about a year ago. It continued with the underperformance of the ramp in our wavelength business and probably crescendoed with concerns around the maturity of our 2027 notes. And I think the data center sale is part of a broader plan that is going to help us reduce aggregate leverage and refinance those notes at a reasonable cost of capital. You know, we had historically hovered at about four times net leverage, and then after acquiring the Sprint assets from T-Mobile, that leverage peaked at 6.7 times levered, and this sale will help us get back to that four times levered target, and then re-accelerate our return of capital program.
Got it. Let's talk about that sale. I-squared is the buyer, 10 data centers, 53 megawatts of installed power for purchase price of $225 million. Before we talk about the details of the actual assets and structure, maybe talk about is this a taxable event for Cogent? What do you think
the net proceeds will end up being of the $225? So I think we will have very de minimis tax friction on this transaction. While Cugent has about $1.1 billion of international NOLs, they will not be applicable to this. We do have about $140 million of NOLs from 2025. And because of bonus depreciation and some of the capital that we've spent on the network modernization programs I think we should not be a cash taxpayer at the federal level on this transaction and most states will follow the federal guidelines there are a few states such as California that ignore some of the
federal NOL guidelines got it and when you talk about the valuation I squared's press release noted 53 megawatts of installed power to me I'm just doing simple math, that implies $4.3 million per megawatt. Is that the correct way of thinking
about it? I think it is, Greg, but that's on a gross inbound basis. You know, our power efficiency is about 1.5 in these facilities. So if you increase... Meaning you need power to keep
The light's on.
So the critical IT load is more like 37 megawatts, but the full 53 megawatts is backed up and protected by a generator.
OK, so the actual usable power is closer to 37 megawatts for GPUs or whatever I squared wants to use these for.
That's correct. That's AC-converted, protected power.
Okay. And so I guess it's either 4.3 on a gross or somewhere in the 6 range for usable power. Is that a good relative value marker for the remaining assets, the remaining 14 data centers you're looking to sell?
You know, ultimately, the market will make that determination. Just to remind investors, these were facilities that came along with the acquisition that we paid $1. for the entire network assets. The 482 buildings were a portion of that. We made the decision to retrofit 125 of those buildings and convert them to data centers, and a subset of 24 of the largest were placed up for sale. When we initially put those buildings up for sale, we were asked, what is the ask price? And we looked at publicly traded data center companies and put out a straw man of $10 million a megawatt. Clearly, this result was less than that, but ultimately the market evaluated it and were comfortable with the transaction. we did have a cost basis of $1, have invested $100 million across the 125 facilities that we modernized. About $30 million of that was directly invested in these 10 facilities. And then the remaining facilities will be priced probably by the buyers. And because there's been competitive tension on each of our data centers, I think we'll get a fair value. Got it.
And I think you said you had 87 or 88 megawatts of power to be sold prior to this deal. You just sold 53. That leaves 35 megawatts left to sell?
No, we actually had a total of 109 megawatts across the 24 data centers. So if we step back, Cogent has today 185 data centers, roughly 213 megawatts of inbound power in those facilities at approximately 2.1 million square feet. These ten facilities represent 53 of that power pool and then the remaining facilities have 56 megawatts of inbound power that's available.
The remaining 14 data centers have 56 megawatts of gross inbound power and now how efficient
About similar efficiency.
Okay, so 56 and gross, and then, okay, something less than that.
Roughly 40 of net.
And I think in the prior LOI, we were selling two data centers. I believe one of those data centers sold here, and then one of them is now of the remaining 14 that still could be.
That is correct.
I guess that bodes well, because that means there's at least a data center out there that piqued folks' interest. And another thing I think that bodes well, I guess, Is it fair to say that at this point in time, you have more patience as a seller? Because, you know, maybe this prior sale, you had the refi on your mind. And, you know, now that you've conducted this sale, you know, it sort of helps clear the decks in that sense. And now you can be a little bit more of a patient seller. Is that a fair statement?
You know, I think we've always tried to maximize value. We realize that data centers are not our core business. You know, in the remaining 161 facilities that we'll have, after we complete the sale of all 24, we still only generate about 3% of Kojin's revenues from space and power in those facilities. You know, I don't think we were particularly anxious on this sale. We found I-square to be a very good counterparty. party. They were very efficient in that when the other deal with the first buyer fell apart because the buyer requested us providing financing to them, I-squared was very quick to look at the portfolio, identify the 10 locations that it was interested in out of our pool of 24, and then immediately put together a team of consultants and internal resources to do their diligence. We have been focused on getting this deal complete. That diligence window was fairly short. It was only 90 days. And then, as you can see, from the ending of that diligence, they intend to close within two weeks by June 12th. And then I think we'll have resources available to turn to the others. But I don't think we feel that we're under any pressure to get the next 14 done.
And it's going to be tough to put words in the bio, you'll have to ask I-squared, but what does it sound like through your negotiations that I-squared was looking for on these 10, and what do you think they're using them for?
So I-squared is a $60 billion global infrastructure fund. They do have other communications assets. They're primarily focused on energy, but they've diversified into digital infrastructure. They own a couple of fiber companies. They own a European data center operator, a Latin American data center operator. And they're viewing this as their North American data center footprint, that they intend to invest incremental capital, and they've earmarked a billion dollars initially of equity for this investment. we're getting 225 of that. I believe they are focused on inference, and they feel that the geographic footprint of this portfolio puts them in a very good position to be able to provide
inference to virtually all of the U.S. market. Okay. Even though they're fairly small data centers, there's enough inference to go around? I mean, a typical data center that you're selling
with like four megawatts. So, you know, there's been a lot of attention focused on very large data centers that are used primarily for AI training. And inference requires less power, should be more distributed, and have lower latency to the end user. And I think these facilities are ideally suited for that application and then just switching
gears and the other 14 data centers again are there any updates there any change to the sales process now that these other ten are done and what are the county parties and deal structures look like would you be willing to sell you know the whole thing at a discount or would you do singles and doubles if
you have preferences there you know we've been focused on getting this deal complete. Like I said, it was a fairly tight window and we don't have infinite resources and we chose not to engage a bank to help us. We did this on our own. I think this deal is now two weeks away from closing. Once it closes, we'll be able to divert resources and re-engage with a number of other parties. I think we're open to anything from a single transaction to the remaining portfolio. We've also had a few parties express interest in other data centers that we were not initially contemplating selling. But again, we're open to any or all reasonable transactions for this footprint. That's a good segue to my next
question. You actually have 185 cogent-owned data centers, 3% of revenue, as you noted. Can you remind me what the company is using these data centers for, and would you have room to even monetize those as well?
So many of our data centers predated the Sprint acquisition. We had 54 data centers before the acquisition that totaled 69 megawatts of power and about 634,000 feet of raised floor space. 52 of those 54 facilities were leaseholds, two were owned, fee simple. When we acquired Sprint initially, we thought that we would use only a fraction of the 482 buildings and only a small portion of the power and square footage in those buildings. We then realized that the value that we had in inbound power could be used by others and made the decision to convert 125 of the former Sprint switch facilities to data centers. We're selling 24 of them. we actually are shutting down a few of the legacy cogent leasehold facilities and we would be willing to transact on any of the other facilities if it made
sense. Okay I mean are they considered non-core? I mean if you sold them are they too critical to sell? You know we keep a presence in them to serve
customers, we do generate revenue in RAC and power, and oftentimes our core metro networking equipment sits in one of these facilities. But it takes up a very small fraction of the total space that's available, and we could easily continue to operate in those facilities as just a tenant of a data center operator. Finally, some of the data centers also have a cogent sales office adjacent to the data center space, but within the same building. So, for example, our Pasadena facility, which was the old Earthlink main data center, or or our Seattle facility have sales offices co-resident. So we would probably keep that sales office and maybe sell the data center space.
God, I want to switch gears to another important topic, which was CapEx over the last quarter. It was a surprise to the downside, meaning CapEx was $46 million, well above street estimates, and sort of in the opposite direction of the message $25 million per quarter that we were all hoping for after the sprint refurbishing of the data center. So you noted increased costs, increased DRAM equipment costs, and the hyperscalers are crowding you out of the equipment. So the question I have is, where does CapEx go from here? Do you expect the increases in this equipment cost to be transitory? Typically, routers follow Moore's law, but we now have to step up in terms of CapEx.
is like 35 the new 25. I hope not. So first of all, our CapEx on a year-over-year basis in the first quarter came down by $12 million. That is not as much as we expected it to decline. It was roughly half the rate of decline that we were forecasting. Unfortunately, we've seen a number of factors contribute to increased capital costs. Also, because Cogent does not build outside plant, 100 percent of our capital is either equipment or capitalized labor to install that equipment. There have been three primary drivers of increased equipment costs. One you mentioned, which is the DRAM shortage and the increase in RAM pricing. The second has been general supply chain disruptions and some of the impacts of tariffs still flowing through the equipment supply ecosystem. And then third, there's definitely been a shift in the buyer universe where hyperscalers for their AI efforts have become major buyers of telecom equipment. They have pressured margins for the equipment vendors, and the vendors have responded by raising prices for their more traditional customer base. You know, we buy equipment from three vendors, Cisco, Sienna, and Arista, and we normally operate in a world where existing equipment declines, and most importantly, next generation equipment is typically priced at parity to the older equipment, but provides an order of magnitude improvement in function. functionality. What we have seen, which is very different than anything in our past 25 years, has been sequential price increases. One of our equipment vendors had five price increases in five months. When I was talking to a member of their senior management team saying, is this the end? He just said he didn't know. They were just under such pressure that there may be further price increases. I think this is temporary. I don't want to use the word transitory, because there is the ability to get more functionality out of the equipment at a lower input cost. I think the short-term shocks of incremental buying capacity from the hyperscalers and memory are temporary, and we should see a reversion to the typical rate of price decline. And if you're measuring that for optical transport gear, it's just literally the cost per bitmile carried, falling at about 80% a year, which has been the 35-year long-term average price performance improvement. And then for routing, the primary metric you would use is cost to forward a packet, and that has fallen at about 40% per year for the past 30 years.
So the curve's coming down or should continue to come down in perpetuity, but now the fear is there's like an order of magnitude step up now because input
costs are now just higher. That's correct, and it should start down again
from a higher base, I guess, is the point. I guess it goes back to my question, is 25 million a quarter now a little more aspirational than it
maybe was in the past? I don't think it's realistic in this environment. You know, I've been pressed to say what the new number is, and we're gonna be as capital efficient as possible. We have achieved ROICs substantially above our cost of capital ever since the company's inception. And I think for the next year or so, we will be running above that number. But with five increases in five months, it's really hard for me to predict whether we'll have another seven this year or we're done for 2026 got it um does cogent have
pricing power because obviously in any industry when you see input costs go up you can pass it along to your customer just curious to hear your thoughts on passing along the higher costs to your
customers so capital as a percentage of revenues has typically been around 10 or 12 percent of revenue. So it's not a material input to our cost structures the same way that either OPEX or SG&A are. You know, I do think across the industry, all of our competitors are facing similar types of pricing pressure for equipment. Normally, the cost of internet bits has fallen at a compounded rate of 23% a year. That's a 35-year average rate of decline. It does appear that that is moderating, and our rate of price decline has moderated. Whether that's a permanent reversal of that trend is yet to be seen. I think there is still sufficient competition in our wavelength business where we are a new entrant we typically price at about a 20% discount to the market because there are other qualitative factors such as diversity and reliability and route uniqueness that go into the buying decision I think there probably is a little bit more price rigidity than there is in a more fungible product like transit got it
it's a perfect segue for the last few minutes is the Waze business it continues to struggle to get to that ramp of 500 million run rate by mid 2028 which now looks more and more difficult can you just help us with you know what's happening here you've noted on the last earnings call and previous earning calls that customers are delaying their orders. But there's a bigger fear out there that there's just a simple difficulty to sell the Waves from Cogent. Is it that some folks are worried that the fiber's older or the strand counts aren't large enough? Maybe, or is the product set or does Lumen and Zeo sell as larger RFPs where you only sell one or two products? So we're using it with the sales force. I gave you a lot there, but maybe tackle some of those issues and help us ease our concerns
or what is the concern here? So our Wavelength business is a new business for Cogent. It grew last quarter 90% year over year off of a small base. We have gone from 0% of the market to 3% of the market in a year and a half of actively selling the services. We are continuing to build credibility with customers. We have the largest sales force focused on this customer segment and selling these products. We have more data centers wave enabled than any other provider. We ended the quarter with 1,107 carrier neutral data centers in North America. We can provide wavelengths in any data center to any data center configuration, which means there's over 10 to the 2,800th power number of permutations of wavelength combinations, and we can offer those wavelengths in either 10, 100, or 400 gig speeds. That is a far more robust portfolio than our competitors. You know, we are building credibility in this segment. We are the world's largest carrier of Internet transit traffic, carrying about a quarter of the world's traffic. For wavelengths, we have a ways to go to build that credibility. What is actually helping us, though, is the market is growing and is more robust than we had initially anticipated. it. The incremental use case of AI is driving significant demand. In terms of fiber age and strand count, when you are selling a wavelength, you're selling a finished product. In that finished product, people care about the uptime and the bit error rate of that product. As long as it meets those specs and it is reliable, then they will buy it if it is priced correctly. Where we have a significant advantage is the uniqueness of our routes, 90% of them have no one else on the right of way, where the two other companies you mentioned typically build their fiber along public highway right of way, it is common, in plastic conduit that's less than two feet below grade. The network that we acquire from Sprint is in an armored cable six feet below the middle of a train track. Highly reliable. Typically gets cut about one-seventh as frequently as our competitors. So for the customers that have tried us, And we do have a total of 492 unique customers now using our wavelengths, and those services have been installed in 581 of our 1,107 data centers. They're extremely happy. I would encourage any investor, either themselves or through third parties, to do channel checks and get feedback. I think the reviews on our waves have been equal or better than our competitors in terms of breadth of footprint, speed to install, reliability, and uptime. On every one of those metrics, we've been winning.
And I think you've noted 25% of the wave buyers are the hyperscalers, which I think is promising news because it tells you that the hyperscalers are perhaps smaller than folks thought, given AI demand. When you're selling to these hyperscalers, I know it's tough to see what exactly they're doing and what use cases they're using with this fiber, but is it more for training or is it really going towards inference? And don't you play more into the inference side then? Is that a fair statement?
No, I think we can play in all segments. I think inference will actually most likely occur more over the public Internet. IP transit. Yeah, I'd be transit as opposed to needing wavelengths. You know, hyperscalers use us for typical data center to data center connectivity, content distribution, and AI training. We're also selling to all of the neoclouds or pure AI companies. You know, there's been a lot of buzz around AI training. There are trillions of dollars earmarked to be invested in those training facilities, but many of those facilities have yet to be constructed. Until those facilities are up and operational, they won't need wavelengths. It's one of the last items that they will use. The reason why wavelengths are used instead of Internet access in those facilities is because of the defined latency characteristic of a wavelength versus the Internet. Because GPUs represent about 55% of the capital investment in a training facility, the need to keep those GPUs at 100% utilization is paramount. And that utilization demands a fixed amount of latency rather than a variable latency such as the internet would deliver. And for that reason, I think we're going to continue to see training huge wavelengths. I think the inference component of AI will be much more distributed. And obviously, I think the billion-dollar bet that I-squared is making is, you know, that inference will be much closer to the edge. Training will be more centralized. Got it. And with that, we're about out of time. Thanks, Dave. Hey, thanks, Greg. Thank you all.