Cogent at Goldman Sachs Conference: Strategic Growth and Challenges

On Monday, 08 September 2025, Cogent Communications Holdings Inc. (NASDAQ:CCOI) presented at the Goldman Sachs Communicopia + Technology Conference 2025. The company outlined its strategic focus on internet and VPN services, the acquired Sprint GMG enterprise business, and wavelength services. While Cogent anticipates growth in its core business and wavelength services, challenges remain with the legacy Sprint GMG business.

Key Takeaways

• Cogent’s core business, comprising nearly 90% of revenue, is expected to grow by 5% annually.
• The wavelength business aims to increase its market share from 1% to 25% by 2028.
• The acquired Sprint GMG business has been stabilized but is projected to decline by 1% to 2% annually.
• Cogent plans to reduce capital intensity and grow its dividend sequentially.
• The company is exploring the sale of non-core assets, including data centers and IPv4 addresses.

Financial Results

Cogent’s core internet and VPN services are projected to grow at approximately 5% annually, with an anticipated margin expansion of 100 basis points. The acquired Sprint GMG business, previously declining at 10.6% annually, has reached break-even and is expected to achieve a 20% positive EBITDA margin in the coming years, despite a forecasted decline of 1% to 2% per year.

The wavelength business, a key growth area, currently has a $36 million annual revenue run rate and aims for $500 million by 2028. This expansion leverages Cogent’s structural advantages, such as network reach and aggressive pricing.

Operational Updates

Cogent has enabled 938 sites for wavelength delivery, offering speeds of 10 gig, 100 gig, and 400 gig. The focus is on expanding market share in the inner-city wavelength market and connecting AI training data centers.

The company has converted 125 former Sprint sites into data centers, exceeding its initial target. These conversions involved significant infrastructure investments.

With a sales force of 650 representatives, Cogent targets service providers who purchase wavelength services.

Future Outlook

Cogent expects total revenue growth between 6% and 8%, with a margin expansion of 200 basis points annually. The company plans to continue growing its dividend and may consider share buybacks. While M&A activity is unlikely, Cogent is focused on reducing capital intensity following the completion of network repurposing and data center refurbishment.

Q&A Highlights

Cogent’s wavelength customers include major international carriers, hyperscalers like Meta and Google, and AI companies such as OpenAI. The company is evaluating options for its IPv4 addresses, including selling or leasing.

Data center asset sales are being explored, with trial pricing set at $10 million per megawatt for purchase or $1 million per megawatt per year for leasing.

CEO Dave Schaeffer emphasized Cogent’s aim to capture a 25% market share in the inner-city wavelength market, leveraging unique structural advantages and aggressive pricing strategies.

For a deeper dive into Cogent’s strategic plans and financial outlook, readers are encouraged to refer to the full conference call transcript.

Full transcript - Goldman Sachs Communicopia + Technology Conference 2025:

Mike Ng, Analyst, Goldman Sachs: Good morning, everybody. Welcome to the Goldman Sachs Communicopia and Technology Conference. I have the privilege of introducing Dave Schaeffer, who is the Founder and CEO of Cogent Communications Holdings Inc., which he founded in 1999. My name is Mike Ng, and I cover U.S. telecom here at Goldman Sachs. We have about 35 minutes for today’s presentation, inclusive of any audience Q&A. I’d like to start in saying thank you, Dave, for making it out here to the conference and spending some time with us.

Dave Schaeffer, Founder and CEO, Cogent Communications Holdings Inc.: Thank you, Mike, for hosting me. Thank Goldman Sachs for a great venue. I’d like to thank all the investors for taking time out of their day to hear a little bit about Cogent.

Mike Ng, Analyst, Goldman Sachs: To start things off, I was wondering if you could just talk about the big picture, three to five-year view. What are some of the biggest developments that you’re focused on? Where should we be spending our time as analysts and investors on the Cogent story?

Dave Schaeffer, Founder and CEO, Cogent Communications Holdings Inc.: Yeah. For those of you who have covered telecom for any long period of time, it has been a very tough sector for investors. The Internet has been very value-destructive to the traditional telecom ecosystem. That’s going to continue going forward. Cogent has differentiated itself and been able to grow in that market by focusing on a limited number of products where we have a structural advantage. The bulk of our revenue today, nearly 90% of it, comes from selling Internet-based services. For 25 years, Cogent was exclusively a provider of Internet and VPN services over the Internet. That is going to continue at Cogent to be a significant part of our business. It is going to continue to grow, and we are going to continue to capture market share.

We have significant structural advantages based on our network architecture, the reach of our network into 57 countries, about 1,870 carrier-neutral data centers, and over a billion square feet of multi-tenant office buildings where we sell to end users. A second part of our business is selling that same product set to large enterprise customers. This was part of the acquisition that we received when we purchased the Sprint GMG business from T-Mobile. That is a business that was in decline. It was declining at 10.6% a year for the three previous years to our acquisition. It was almost exclusively an off-net business, and it was a business that had a large number of non-core products. We have worked diligently in taking costs out of that business, streamlining the product portfolio, and getting that business from a negative 80% margin to break even.

Over the next couple of years, we will be able to get that business to about a 20% positive EBITDA margin. It’s a business that is predominantly off-net based on where these large multinationals buy service. It is a business that probably will continue to decline at 1% to 2% a year, but we will be able to harvest some cash. Maybe the third leg of the stool is the most exciting. It has been the repurposing of the Sprint physical assets in order to sell wavelength services. The Sprint network was built between 1985 and 1992 at a capital cost of $20.5 billion. That network was comprised of 19,000 route miles of inner-city fiber, 1,200 route miles of metropolitan fiber, and 482 buildings. It was exclusively designed initially to carry long-distance voice. That network has been dormant since 2015.

We acquired that network from T-Mobile as part of the Sprint GMG acquisition. That acquisition had two components: a $700 million subsidy payment for taking over the declining and money-losing enterprise business, and it sold us the physical asset base for $1. We spent 18 months repurposing that asset solely to deliver wavelengths. We have a startup business and a public company, a wavelength business. We have enabled 938 sites as of the end of the second quarter to get a wavelength that can be delivered within 30 days in any direction to any site in one of three speeds: 10 gig, 100 gig, and 400 gig. The addressable market that we are going after is a $2 billion addressable market. It historically had three use cases, now four. The first use case was international network extensions. The second use case was regional network consolidation and aggregation.

The third use case was content distribution. A fourth use case, which is growing rapidly, is connecting AI training data centers to AI data that needs to be used to build large language models. This is driving an acceleration in wavelength demand. We have five structural advantages that our competitors cannot match. One, we have a network that goes to more endpoints. We’ve got an architecture we can provision faster. Three, our physical routes in 90% of the cases are unique to us. We provide a level of protection and diversity. Fourth, because we acquired this asset for $1, we have the ability to price very aggressively to capture market share. Fifth, because of the way in which the fiber was constructed, it is more reliable than our competitors.

Our ultimate goal over the next three years is to go from 1% market share in the inner-city wavelength market to 25% market share, growing that business from a $36 million run rate at the end of last quarter to a $500 million run rate business. Within Cogent, there’s a core business that’s growing at about 5% a year with about 100 basis points of margin expansion. There’s a legacy acquired business that is declining in low single digits with no margin today and margins improving. Finally, there’s this rapidly growing new business line that is going to contribute very high margin revenue.

Mike Ng, Analyst, Goldman Sachs: That’s a fantastic overview, Dave. Maybe just on the wavelength piece. $500 million annual revenue run rate, $2 billion TAM, the 25% market share. What do you need to do to execute from where we are today at $36 million to get to that $500 million by 2028? Key customer wins, investments in the infrastructure.

Dave Schaeffer, Founder and CEO, Cogent Communications Holdings Inc.: Yeah. The investments have been made. In fact, our capital intensity is declining. We built out a network that only requires two field deployments in order to provision each wavelength. We plug in an optic at each endpoint and then can automatically provision that wave. The architecture that we have deployed is optimized for one and only one product, wavelengths. That is very different than what our competitors do. They typically sell wavelengths off of a multi-service platform where they’re selling remnant or excess capacity, which requires much more customized deployment, much longer deployment windows, and oftentimes results in wavelengths actually being ordered and then never installed. What we need to do to get from $36 million to $500 million in annualized run rate is simple. We need to earn the credibility of the customers. The customers need the wavelengths.

The market is concentrated today among a couple of key players. The customers are dissatisfied with those players for three reasons: poor performance, poor installation, and difficulty in getting pricing that they feel is reasonable. We have the opportunity to win on each of those fronts. We have an existing sales force of 650 sales reps. That sales force is divided into corporate end users and then those that focus on service providers. Virtually all of the wavelength growth is going to come from the service provider portion of our sales force to roughly 300 sales reps that focus on customers that will buy waves. Three quarters of all wavelength buyers are already buying transit services from Cogent. We have a pre-existing relationship. We’ve demonstrated the quality and superiority of our transit services. All we’re asking is for the opportunity to do the same in the wavelength market.

If necessary, we will discount heavily if that’s what it’s going to take to win market share. To date, we have not yet had to be as aggressive as we had originally planned.

Mike Ng, Analyst, Goldman Sachs: When you talk about these customers, just in terms of the type and category and flavor of these customers, I was just wondering if you could just talk about what that looks like. Are these hyperscale customers, what I would call like AI cloud, Neo cloud type of customers? Are these more tier two cloud, Netflix, Salesforce, like those types of companies?

Dave Schaeffer, Founder and CEO, Cogent Communications Holdings Inc.: All of the above plus some. Let’s start with the legacy customer bases. One, they are companies that are international carriers extending their network: Telefonica, Telecom Italia, British Telecom, Deutsche Telekom. These are all companies that have bought wavelengths from us. They are regional network carriers, companies like a MetroNet, an Alo, a FirstLight, a Hotwire, all customers that have bought wavelengths from Cogent. They are traditional hyperscaler operations: Meta, Google, Amazon, Microsoft, all customers of Cogent historically, who are distributing content: Netflix, distributing content; Akamai, Fastly, as third-party CDNs, distributing content. All of those legacy bases have been purchasing from Cogent. Today, we’re only getting a very small portion of their spend. As we prove in the ability to deliver and the reliability of the service, the percentage of their expenditures that will be shown the opportunity to bid on will increase.

We have a ubiquitous footprint that covers that entire addressable market. For the new use case, the new use case comes from hyperscalers who have a second line of business. They’re in the AI business. Meta, Microsoft, Google, Oracle have all been very vocal about that. There are pure play companies: OpenAI, Anthropic, Lambda Labs, NeoCloud, ones that are really building a network solely for AI training. All of these are Cogent’s addressable market. These are customers that are buying from us and will continue to increase their spend with us. Our view of the addressable market was probably more sober when we announced the deal than it is today. We were burned in the dot-com boom 25 years ago. Cogent was founded in 1999. The boom was sparked at Communicopia in 1998 when the CEO of UUNET gave a speech. I was actually there in the audience.

I was in a little kiddie room with a little one-on-one, and he was in a big ballroom. John gave a speech, and he said, "The number of T1s that we are selling for Internet access is doubling every month for the past six months." That was actually a factually true statement. What everybody inferred from that was that the Internet was going to double every month for the rest of time. It then elicited a nearly $2 trillion investment boom that ended in a crash. When we came to the wavelength market, we looked at four independent market studies that all sized the market in North America inner city at about $2 billion. We then went to our existing sales force and said, "I want a bottoms-up market analysis before we agree to bid on Sprint.

I want you each to go to every one of the potential wavelength customers in your funnel and assess how many wavelengths over time and how much they will spend." Those two numbers triangulated to the same number. If I looked at all of the third-party analysis in 2022 when we announced the deal, they were expecting the market to grow 5% to 7%. I actually discounted that and said, "This is a flat market and we will be successful." The reason we took that pessimistic view is that’s exactly what has happened in the transit market. We have grown to be the largest transit provider in the world. We carry a quarter of the world’s traffic. The TAM for that service is only $1.5 billion, and it hasn’t grown. It’s grown in units, but it has not grown in revenue.

We took that same sober approach to the wavelength market. Where we were surprised to the upside is that as AI training really started to take off, the wavelength demand is growing. I do today believe the total addressable market is probably growing at somewhere between 5% and 10% a year. That’s a good fact. It’s easier to get to our goal in a growing market than a static market. The second bonus from that growth came on the data center side. Cogent connects to 1,870 third-party data centers in 57 countries around the world, more than any other provider. That’s where we sell 97% of our transit services in that footprint. We also have 180 of our own data centers. It is a very small business in Cogent. It’s a sub-$30 million business.

With the acute shortage of power driven by AI training, that has created demand for some of these data centers. Our hope is we can sell off some of these facilities. We did ramp up investments, but we were encouraged when customers who would use these facilities came to them and told us what features they would need added to what were former switch sites to convert them to data centers. That work is now done. We are in the process of trying to sell off those facilities. We have, I think, had a couple of fairly significant positive surprises to the upside.

Mike Ng, Analyst, Goldman Sachs: Just on the new use cases, just as a follow-up, could you put a finer point on what some of those things are? You know, the connecting storage to training facilities, that sounds like data center interconnect. Is there a part of this wavelength thesis that’s really relying on enterprise inferencing to really take off in a meaningful way so that the content distributors and some of the traditional hyperscalers see increased demand for wavelengths over this period of time through 2028?

Dave Schaeffer, Founder and CEO, Cogent Communications Holdings Inc.: Yeah. The building blocks of AI have been known since about 2010. It took 15 years for AI really to burst into the mainstream and be something that is now integrated into most applications. The raw material for AI is the data that was collected over the internet. The zettabyte or so of data that is stored around the world is the raw material for large language model development. Because the processing to do that is very power-intensive, usually the processing exists in a different location than where the data resides. If you are building a data center for the purpose of doing AI training, roughly half of the capital invested goes to the GPUs to do that training. The reason why AI is practical today is we have low enough cost compute.

We have enough data to actually compute against and build the tokens necessary to create the large language models and the neural networks that will then be used for the inference phase. We have the toolkits to be able to do that. In fact, large language models are evolving very rapidly, typically on a four or five-month refresh cycle, much faster than traditional software, which is making this much more efficient. The inefficiency of locations is causing the need for wavelengths. While the internet is how all this data was collected and will continue to be the mechanism of collection, it will also be the mechanism of distribution of outputs. The inference phase will occur over the internet from distributed caches around the world. Latency is important for the inference phase. For the training phase, the training has to occur where the power and space is available.

Even though it’s two and a half times cheaper per bit mile to use the public internet to move the bits than to use a wavelength, AI prefers to use the wavelengths because it does not have to buffer the traffic. The internet is not latency determinant. What that means is you don’t know exactly how long it’s going to get to take from point A to point B. A couple % variance in paths can force you to add 10% buffering to your compute power. Since that’s half of your capital expenditure, that is highly inefficient. If you are trying to optimize the number of tokens produced, which is really the primary goal of large language model construction, you’re going to want to do that using wavelengths. An alternative can be dark fiber, which is even more expensive. There are two downsides to wavelengths or dark fiber.

One, they’re more expensive. Two, they’re not protected. They’re vulnerable to any number of real-world inferences, cuts. You know, a jumper gets disconnected in a data center and you’re down. Almost all wavelengths are typically sold in pairs. That’s extremely important because the fact that our routes are diverse from the other two vendors, yet we have the same sitting pairs, gives us a huge structural advantage. The Sprint network was built along railways. Most of the fiber of the other wavelength and dark fiber vendors is along public highway right of way. Railroads and highways go to the same cities, but on typically different physical paths. That spatial separation of a mile or two has tremendous value for resiliency and diversity. The use cases are to get data out of a data repository to a training facility and then return that data once the tokens have been created.

Mike Ng, Analyst, Goldman Sachs: You talked a little bit about the preparation for the data center asset sales that you acquired as part of the Sprint GMG deal. Could you just touch on how much that could potentially be worth? Similarly, I think there are some IPv4 address inventory assets that you’re also considering monetizing. What could that do for your capital structure and success?

Dave Schaeffer, Founder and CEO, Cogent Communications Holdings Inc.: Yeah. We have three assets that we’ve identified as not core to our operating business. The first of them are former Sprint tandem switch sites that are being converted to data centers. There are 24 large sites that have a total of 109 megawatts of existing inbound protected power. We have gone through those sites and converted them from negative 48 DC to AC 120. We’ve refurbished air conditioners, fire suppression systems, distribution frames, generators, all of the infrastructure needed for that critical IT load. We went to the market with a trial balloon of $10 million a megawatt to buy or $1 million a megawatt a year to lease. Both of these numbers represent a discount to where publicly traded comps are. Whether that’s the correct price or not, ultimately, the market will tell us. We invested about $100 million and almost a year in doing these conversions.

The number of facilities that we converted is far larger than what we had initially targeted. On day one, when we announced the transaction in September of 2022, we thought we would convert 45 of the 482 buildings. We actually have grown that to 125. The amount of power that was going to be converted went from the initial 45 to about 150 megawatts, of which the 109 has been viewed as surplus to Cogent’s core business. The second asset that Mike touched on is our inventory of IPv4 addresses. These are the unique hexadecimal numbers that make the majority of the Internet work. The world ran out of those addresses in 2017. We have been leasing addresses. We have raised prices on leases. We have continued to lease out increasing volumes.

We took a portion of that lease revenue and securitized it with an asset-backed securitization that has raised $380 million. We still have about 23 million completely fallow addresses. We are considering selling them, perhaps leasing them, perhaps doing more securitization. Finally, the Sprint physical network is comprised of anywhere from 24 to a total of 144 strands on any given point-to-point path. We do not need all of that inventory to run our IP business, to run our wavelength business, and to account for future growth. We have also considered, and we’ve done just a handful of small dark fiber sales, but we will consider those going forward. Each of these non-core assets represents things that can be sold off to help strengthen our balance sheet. They are not a recurring revenue. The key value creator at Cogent is the recurring revenue and operating margins of that ongoing business.

Cogent will be able to grow its total top-line revenue based on those three categories I described earlier at between 6% and 8%. We will be able to deliver 200 basis points a year of margin expansion. We will do that with declining capital intensity. Those factors will allow Cogent to generate increasing amounts of free cash flow per share that can be returned to shareholders.

Mike Ng, Analyst, Goldman Sachs: Great. I just wanted to see if there were questions from the audience before I moved on. Dave, maybe you can talk a little bit about that declining capital intensity point. You know, you’re expecting to reduce capital intensity, I think, to about a $100 million annual run rate later this year. How do you balance that piece with the needs to invest in the business to compete?

Dave Schaeffer, Founder and CEO, Cogent Communications Holdings Inc.: It’s actually a little more than that because you have to also look at principal payments on capital leases, which is another form of CapEx. It’s just a different accounting treatment. Together, those two numbers are about $140 million. The capital intensity of Cogent prior to acquiring Sprint was the lowest in the industry with the highest growth rate based on the network architecture that we deployed and the product discipline that we applied to the network. When we acquired Sprint, our capital intensity increased for three reasons. We had to repurpose the Sprint network, refurbish the data centers, and interconnect the networks. That work is all behind us. With that, we will be able to have lower capital intensity in our IP business and lower capital intensity in our wavelength business.

One of the key differences in our architecture versus our competitors is it was built solely to deliver wavelengths, not other services. In the deployment of transponders, nearly 40% of all transponders manufactured never become productively deployed. They’re stranded capacity. The architecture that we have deployed is far more effective at reducing that amount of stranded infrastructure. It is a very different topology than the companies that we compete with. That’s what allows us to install faster and what allows us to get better transponder efficiency.

Mike Ng, Analyst, Goldman Sachs: Maybe just in closing, maybe you could talk a little bit about capital allocation. You know, the company’s consistently reiterated intent to sequentially grow the dividend. You’re obviously in the midst of several potential asset sales. Talk a little bit about capital allocation, your willingness to do additional M&A to the extent that the conditions make sense.

Dave Schaeffer, Founder and CEO, Cogent Communications Holdings Inc.: I think on M&A, it is highly unlikely. Between 2005 and 2022, we looked at 825 targets and did a grand total of one. I think we’ve been disciplined. I think that will continue to be our mantra. That doesn’t mean there won’t be things that come available. Right now, there’s nothing that we’re actively looking at. Our leverage has peaked. We knew it was going to peak in Q2 of 2025 due to the investments in the Sprint network and the reduction in subsidy payments from T-Mobile. We are going to consistently deliver from this point forward on a net leverage basis. We’re about 6.6 times levered. That is above where we target, and we do need to deliver. We have bought back 10.9 million shares, or nearly 23% of our cumulative outstanding float. We did some buybacks in the second quarter.

We also have a dividend that we’ve grown sequentially for 52 consecutive quarters, quarter over quarter. We believe we can do all of the above. There is a debate on what is the appropriate mix of dividend, buyback, and delivering, particularly as the decoupling of our share price from our dividend yield, where historically we had a much more market normal dividend yield. Today, we have one that is clearly telling us the market doesn’t believe our dividend.

Mike Ng, Analyst, Goldman Sachs: We’re at time, so we’ll wrap it up there. Dave, it’s been such a privilege to have you on stage here with us. Thank you for joining us, and thank you, everybody, for listening in.

Dave Schaeffer, Founder and CEO, Cogent Communications Holdings Inc.: Thank you all. Thank you, Mike. You did a great job for literally just initiating. Thanks.

Mike Ng, Analyst, Goldman Sachs: Thank you, sir.

This article was generated with the support of AI and reviewed by an editor. For more information see our T&C.