Wonderful. Good afternoon, everybody. Welcome to the Cogent Fireside Chat at the Goldman Sachs Communacopia and Technology Conference. It's my wonderful privilege to introduce Dave Schaeffer, who's the Founder and CEO of Cogent Communications. My name is Mike Ying, and I cover telecom here at Goldman. We have about 35 minutes for today's presentation. First and foremost, thank you so much for coming out here to San Francisco and being with us this afternoon.
Hey, Mike. Thanks for hosting me. I'd like to thank Goldman Sachs for a great venue, and I'd like to thank all the investors for taking time late in their busy day to hear what we have to say.
Wonderful. Well, I'm sure you will not disappoint. Cogent operates one of the largest internet backbones in the world. Maybe talk a little bit about the current state of the business, your strategic priorities, how the company is positioned to benefit from all of these investments in AI infrastructure that we're hearing about this week during the conference.
Yeah, sure. First of all, 84% of Cogent's revenues come from selling internet access or internet-based services. For 25 years, we were purely an internet service provider. Recently, we've expanded our business to include optical transport or wavelength services. Let's start with the backdrop for internet traffic growth. I'm going to bifurcate our market into two segments. I'm going to look at the market by end users. Today, about 2% of Cogent's traffic, but nearly half of its internet-based revenues come from selling to end users. Most of those end users are in large multi-tenant office buildings located in the central business districts of major cities. That business had historically grown at about 11% a year organically. When the pandemic hit, that growth rate declined to - 9%. Today, it's back to being a 3% growing business.
We today have about 35% market share in 1,870 multi-tenant office buildings, where there is about 1.1 billion square feet of office space. We also sell internet services to customers in buildings that are too small or too far from our network to justify us deploying capital. Most of those off-net customers actually came to Cogent through the acquisition of the Sprint Global Markets business from T-Mobile. We acquired a business that was declining and burning cash. T-Mobile paid us $700 million over a 54-month period. We still have just under two years of those payments left coming to us. Our transit business, where we sell bulk internet connectivity in carrier-neutral data centers. The footprint is 94,000 route miles of inner-city fiber connected to 33,000 route miles of metro fiber, serving 1,950 data centers in 58 countries around the world.
Approximately 7,500 access networks from companies as large as China Telecom or Jio buy their upstream from us, as well as all of the major content producers. That business is a volatile business that is based on unit volume growth and price declines. Our transit business today is growing at about 8% in revenue. It is growing at about 15% in unit volume. We are the largest carrier of internet traffic in the world, carrying about a quarter of all internet traffic. Roughly 77% of our traffic stays on net, meaning it goes from one paying customer to another. In 23% of the cases, we only get paid on one side, and the other side is a peer with whom we exchange traffic. This segment of our business has seen an acceleration as a result of AI for two discrete reasons.
One, the traffic that is being collected is the raw material to build large language models and now is increasingly valuable to companies to transmit and collect for that model construction. Secondly, as we are transitioning to an inference phase, we have seen a significant acceleration in traffic growth due to agent-based or agentic AI driving traffic growth across the internet. We expect these trends to continue for multiple years. The internet has gone through a series of discrete waves of growth. It had previously seen a major acceleration in traffic growth due to the acceleration in streaming at the beginning of the pandemic. That has since matured. We saw growth slow and then re-accelerate now based on AI inference. We have a couple of small bolt-on businesses. We lease IP address space out. We are the third largest owner of IP addresses.
When you use the internet, you need an address just like you do to get mail delivered to your house. There are two competing technologies. The more desirable is IPv4. We have 38 million of those addresses. We lease out about 15 million of those addresses and generate about $70 million a year in that address leasing business. We also have our own data center footprint. We operate about 170 data centers, about 1.4 million square feet, and about 150 MW of power. We recently divested of 10 unoccupied data centers for $225 million, selling those facilities to I Squared Capital, and we currently have 14 more facilities up for sale. Finally, the most controversial part of Cogent's business is its wavelength business. This is selling point-to-point optical transport. It is a new business for Cogent. It was a business that did not exist inside of Sprint.
We took the former Sprint long-distance voice network that was built at a capital cost of $20.5 billion between 1982 and 1989. We purchased that asset from T-Mobile for a dollar and then spent two years repurposing that to sell optical transport. Our transport business today is about a $60 million business. Last quarter, it grew 9.2% sequentially, about 62% on a year-over-year basis. A major strategic initiative for Cogent is to grow any of our on-net services. The easiest of those is wavelength. When we add a dollar of on-net revenues, they carry 90% contribution margins. It is why over a 26-year history, we have averaged about 200 basis points a year of margin expansion. Our margin expansion took a material step back with the acquisition of the Sprint customer base, and as we have purged those less desirable revenues, we have resumed our margin expansion.
That's great. There's a lot there that I would love to go over the next 25 minutes or so in the session. Maybe starting out with Waves. $60 million business, I think 3% market share relative to your long-term target of 25%. Talk a little bit about how Cogent's wavelengths may be different or competitively different from those of your competitors. Thematically, what's driving increased demand for point-to-point optical transport? Is it speed? Is it privacy? Is it data center interconnect, just as we have more distributed AI training? What do you see that's happening that's driving demand for Waves?
It's all of the above, Mike. First of all, the cheapest, easiest way to move bits is over the public internet. There is this thesis that the world converges to a single network. That, in fact, has happened. It's called the internet. It has cannibalized every other telecom service, and it has disadvantaged many legacy service providers. A wavelength, which is in fact a private line service, it is a point-to-point fixed connection, has three characteristics that cannot be replicated on the internet. The first and most important is defined latency. You know exactly how long it takes to get from point A to point B. The internet is a non-deterministic network, and by its very architecture, is dynamically routing traffic over different paths to provide maximum resiliency. The wavelength market or private line market has been around since the 1940s.
It was initially delivered over fixed microwave by AT&T. MCI entered that market. In the late 1990s, the market pivoted to using optical transport. The term wavelength came into use. In fact, the first private lines that were sold on optical were dedicated wavelengths. What is sold today is a subset of a wavelength. We do not actually sell a full wavelength to any customer. We sell them a private line at either 10 gig, 100 gig, or 400 gig interfaces delivered over a wavelength, but it is a subset of the spectral capacity of that wavelength. Initially, the market was dominated by government and large corporate users. The market really exploded with the advent of switch-based long-distance resellers. That market then collapsed. It was then replaced by an ISP market, which also collapsed.
For the past 20 years, the market has had three primary use cases: international carrier network extension, regional carrier network aggregation, and content delivery. In the past three years, there is a fourth and significant new use case, AI training. AI training requires wavelengths because approximately 60% of the capital cost of a training facility is in the GPUs. You want those GPUs running at 24/7 and not requiring buffering. The internet would slow down that GPU utilization, so companies generally pay about 2.5 x as much per bit mile for optical transport as they do for internet services. It is also an unprotected product, which means that any physical point on the path can result in a failure if that point fails. That almost always requires customers to buy wavelength services in pairs. One as primary, one as backup.
That market had become dominated by two major players, Lumen and Zayo Group. We had the opportunity in the Sprint acquisition to get paid to fix a broken enterprise business and to then be able to buy that $20 billion asset for a dollar. An asset that doesn't produce revenues is not an asset. It's a liability. The carry cost on that fiber was about $140 million a year. We continue to reduce that cost by selling off certain facilities, by optimizing the network, and the way in which we generate profit is by repurposing that fiber. By taking that 19,000 route miles of inner-city fiber, connecting it to an additional 12,000 route miles that we've secured across North America, and then connecting it to about 22,000 route miles of North American metro fiber. We then interconnect that fiber to 1,137 data centers.
We built a network that is truly unique, optimized for wavelengths, quick to deliver, can deliver any of three speeds, and deliver an any-to-any solution. So over 10 to the 3,000th power number of permutations. As we compete in this market, we have five discrete competitive advantages. We have more endpoints, faster delivery, unique routes, higher reliability, and lower price. We have been selling wavelengths for about a year and a half. I wish we had sold them even faster and sooner, but we have gone from zero to 3% market share. The uniqueness of our asset gives us a competitive advantage, and we've got the tailwind of this fourth use case in AI training that I think will allow us to reach our revenue target of $500 million in wavelength sales.
On that wavelength revenue target, could you talk a little bit about the pace at which we'll get there? In the past, you've talked a little bit about data center construction delays as something that's affected customer acceptance. So where are we in that customer acceptance life cycle? Are there any other delays that may be impacting wavelength acceptance?
As of the end of Q2, we had installed and began recognizing revenue on about 2,500 unique wavelengths. Those wavelengths terminated in 608 unique data centers. They were sold to 548 unique customers. That is about one tenth the number of waves that we ultimately need to sell to hit our revenue target. A wave is priced in a three-dimensional grid. It looks at the length of the path, the size of the wave, and the duration of the contract. We continue to see a migration to longer wavelengths and a migration to higher throughput. What has been frustrating to us is that some demand is not being fulfilled because customers, while they want the service, cannot accept the service due to limitations in their architecture. Those limitations could include the fact that a data center is not built, a data center has insufficient power.
Perhaps there is insufficient server capacity, insufficient memory, insufficient GPUs, insufficient pluggable optics, perhaps insufficient router equipment for those facilities. We have focused on the $7 trillion of announced AI training capital that is being spent. About $1 trillion of that $7 trillion has been spent. $6 trillion is yet to be spent. Also, it is not a linear expenditure in the sense that there are certain items in the supply chain that may be constrained, and most of the capital is spent, but not all of it, until all of those items are delivered. It has frustrated us. I think we will continue to see a gain in market share and a willingness of customers to broaden their view of Cogent from being just an internet provider to now being also a provider of optical transport.
Great. I want to touch on the data center sales. As you mentioned, 10 data centers sold. What was the price per megawatt on those data centers, and then do you expect the remaining 14 facilities that are for sale to transact at a similar price?
We sold a total of 54 MW across 10 facilities, approximately 500,000 sq ft at $4.3 million per megawatt. We have 14 additional data centers up for sale. We have LOIs on four of those additional. The pricing is similar. Each data center is slightly different based on its geography, its square footage, its existing power, and its potential for power upgrades. All of these facilities were originally built as telephone switches. When we first acquired Sprint, we did not intend to invest in the switch footprint, and rather focus almost exclusively on the wavelength business. About a year after we announced the acquisition, we had a number of inbound inquiries, and it became clear to us that the 230 MW of power that was provisioned for those phone switches had economic value. We evaluated all 482 facilities. We chose 125 to convert.
We earmarked $100 million of capital that we did not have. It increased our leverage. Between June of 2024 and June of 2025, we did the repositioning of these former telephone central offices into data centers. We then placed them in the market for sale. We quickly signed some Letters of Intent. We closed our first sale on June 29th of 2026, resulting in 10 facilities and $225 million in gross proceeds. Because of the tax attributes that we had and the fairly de minimis transaction cost, we will recognize about $224 million as a tax gain, but are able to shelter that and should be able to retain virtually all of that gain. We then used a large portion of that, $175 million of the $224 million, to repurchase debt at a discount in the open market.
We repurchased $138.9 million of our 2032 secured debt at an average of about 90 cents on the dollar. We also repurchased about $50 million of our unsecured debt at a slight discount of about 99 cents on the dollar.
Great. I wanted to ask about Cogent's margin outlook. It is something that Cogent has been able to address, driving margin expansion because of the wind down of the Sprint integration costs. Now that the bulk of the work is behind you, what cost efficiency initiatives are you prioritizing next? How are you thinking about margins going forward?
Yeah. Let us maybe go back and look at Cogent's margin trajectory pre-Sprint. We went public in June of 2005 with 0% EBITDA margins. It was our first quarter of turning EBITDA breakeven. Over the next 18-year period, we averaged 220 basis points a year of margin expansion. That margin expansion came from the fact that roughly three-quarters of our sales were on net, one-quarter of our sales were off net. Those on-net sales carried 90+ percent contribution margins and allowed our margins to grow from zero to 40.5%. In May of 2023, we completed the acquisition of the Sprint operating business from T-Mobile. That business had - 60% EBITDA margins, was declining at 10.9% a year, and represented 42% of the revenues of the combined company. In the first quarter post-transaction, our EBITDA margins fell from 40.5% to below 2%. Our annual EBITDA was running at only $5 million.
Now, to offset that, we got $700 million, $350 million of it in the first 12 months, the other $350 spread over the next 42 months in the form of a subsidy payment from T-Mobile. With that added in, our margins actually expanded all the way up to 47% initially and then receded back to about 15%. The underlying margins without the subsidy payments have gone from a 2% margin to about a 20% margin today. We anticipate being able to grow our margins annually at at least 200 basis points a year until we resume margins in the mid-40s. We have a long runway ahead of us. That margin expansion comes from three primary drivers, cost reductions, product rotation, and new product sales, primarily of on-net services. Over the past three years since closing the Sprint transaction, our aggregate reported revenue went from being positive 10.2% to - 5%.
We've had 12 out of 12 quarters of negative revenue growth, yet in 11 of those 12 quarters, we've actually delivered EBITDA growth, not just margin growth, but absolute EBITDA growth. That should continue. We still have about $8 million of cost savings as of the end of the quarter, the end of Q2 to achieve. We also will achieve about a $7 million annualized savings from the divestiture of the data centers, and we are eliminating $36 million in expenses associated with integration. This is above and beyond the contribution margins from on-net sales. In the most recent quarter, Q2 of 2026, our on-net sales were 82% of sales. Cogent's revenues actually have grown every quarter since the deal is closed by approximately 29% or about 8% per year.
That's been masked by the fact that the acquired Sprint revenues have declined by 69% during that same period, resulting in the total top line reported revenue trajectory of - 5%. But on a going forward basis, we should see more of our EBITDA growth come from top line growth because the acquired Sprint business has become only about 15% of the combined company's revenue as opposed to being 42% at closing.
Great. I wanted to get one in on capital allocation. Where are you in terms of refinancing? How do you think about further debt reduction, network reinvestment, and then restoring capital returns to shareholders?
To remind investors, Cogent has returned approximately $2 billion to equity. Virtually all of that has been tax deferred. Our dividends, which have been treated as a return of capital, were about $1.7 billion and about $300 million in the form of share repurchases. We have dramatically reduced that capital return program in order to deliver. The acquisition of Sprint increased our net leverage from 4.2 x- 6.7 x. Today, we are at 6.23 x net leverage as of the end of the quarter. We have a stated policy of getting back to 4x net leverage before we will meaningfully return capital to equity. That return of capital will probably be a combination of dividends, hopefully treated as return of capital, and also in the form of continued buybacks. In order to provide the company adequate liquidity, we have been repurchasing some debt at a discount.
We also know that we have $700 million of our unsecured debt that comes due in June of 2027. In order to address that maturity, we are working on converting the unsecured debt to secured debt. We have ample capacity within our indentures to do that, and then using that secured capacity to do a new $700 million secured offering. We have been working with another bank on this. There is not yet a formal transaction launched. We have been in discussions with both existing bondholders and potential new bondholders, and are looking to optimize our cost of capital.
Great. In the last minute that we have, I would just love for you to talk a little bit about key things that you think investors should be focused on as it relates to the Cogent story.
Listen. It has been a tough year for Cogent's equity. That equity has sold off meaningfully. The fundamentals of the business are actually stronger today than they have ever been. Those fundamentals are somewhat masked by the complexity of a transaction, where you get $20 billion of asset value for a dollar, you get an operating business doing $485 million in revenue with - $300 million of EBITDA, and you are paid $700 million to take it. I think it is that accounting complexity, the difficulty investors have in parsing out the underlying Cogent story from the totality of the combined company that has penalized us and raised our cost of capital. We firmly believe our cost of capital will come down, and we will be able to continue our track record of returning capital to equity.
Great. Well, Dave, thank you so much for participating in our conference. It's been a privilege to have you on stage here.
Hey, thanks, Mike. Thank you all very much.