Cogent’s Network Transformation

Introduction

When Cogent closed its acquisition of Sprint Wireline, it expected to generate $100 million of annualized run-rate Wavelength revenue by May 1, 2024. The company missed that target by ~85%, materially raising investors’ skepticism in the company’s long-term ambitions of generating $500 million in run-rate revenues by mid-2028. $500 million of revenue at 95% incremental margins is the single biggest swing factor in Cogent’s future equity value. Every $50 million of incremental Wavelength revenue equates to nearly $1.00/share of pre-tax free cash flow. This matters a lot.

Understanding why Cogent missed its guidance has significant downstream implications on its future equity value. Can it not serve this market well? Is customer demand below expectations? Or was something else going on? The quote below helps set the stage for the rest of this Insight.

“There are nearly 10,000 discrete work projects and nearly 1,000 of our 1,900 employees are involved in these reconfigurations. It will all be complete by year-end. We are able to provision wavelengths today, but each wavelength takes custom engineering and will take 90 to 120 days.

Each of these custom designs slows our reconfiguration work gap. So while we’ve installed about just under 800 wavelengths and we have about 2,400 in the backlog, we are trying to focus on doing this foundational work, and therefore, be in a position to clear the backlog and accelerate sales early next year. So I think the way to think about the cadence is slow between now and the end of the year, some incremental proof points, but then a material acceleration in our wave deployment.”

— Dave Schaeffer, TD Cowen Conference, May 2024

There is a tension between selling and provisioning Wavelength services and completing the reconfiguration work. In our conversations with the company, it became clear that Cogent de-emphasized selling Wavelengths earlier this year, preferring to focus its efforts on the foundational reconfiguration work instead of disappointing customers with delays. In doing so, it sacrificed current revenue and backlog growth.

We’ve written about Cogent in the past, specifically about Cogent’s business opportunity in Wavelengths and its “hidden” assets (part 1, part 2).  Most of what we’ve written has been an attempt to quantify upside potential (from Wavelengths) and downside protection (from Cogent’s non-core assets).  This post will attempt to consolidate our thoughts and our work on the network transformation and reconfiguration that is underway, as mentioned above in Dave’s quote.  It will be divided into two sections: (1) Cogent’s long-term network and go-to-market strategy and (2) how that applies to the wavelength network.

This summarizes some of the work we’ve done over the past 12-18 months, but especially in the last few months as it became clear that Cogent would miss its $100 million run-rate revenue target.  We’ve worked hard to understand the dynamics that contributed to such an egregious guidance miss, and if we should have longer-term concerns about Cogent’s growth potential in the Wavelength market (long story short, we do not).  

With that, let’s dive in.

Cogent’s Network and Go-to-Market Strategy

Cogent’s tagline is “Smart People Buy Dumb Pipes.”  It was the first (and possibly only) telecom company acknowledge it sells commodity services (connectivity) and to design its company around that acknowledgement.  Listen to other companies and you’ll hear phrases like “value-added services” and “solution sales” and “differentiated network.”  They try their hardest not to be a dumb pipe.

Without confusing everyone with telecom jargon, it’s useful to think about telecom in three layers:

1. The infrastructure layer

2. The network services layer

3. The application layer

The infrastructure layer includes all the physical network architecture, which could be copper, fiber, coax, etc., as well as all necessary equipment to move data around the network.  The network services layer sells connectivity services to customers.  These could be voice services, data services, video services, etc.  The application layer is where much of the equity value has accrued – Google, Meta, Uber, TikTok, etc.  These are the businesses built upon commodity network services. 

Historically, Cogent has operated a thin infrastructure layer to support its commodity network services layer, which it sells to customers on a retail or wholesale basis.  Its network services businesses sell a small assortment of services: (a) internet connectivity to customers in skyscrapers, (b) wholesale internet connectivity to customers in data centers (i.e. IP Transit), and (c) VPN services to corporate customers in skyscrapers.  It also has an off net corporate business and has an IPv4 leasing business and a wholesale colocation business, both of which leverage unused physical and digital space in its portfolio.  Those won’t be addressed in this post, but check out our other Cogent Insights for more details on them. 

Cogent took advantage of the massive fiber overbuild from the telecom boom in the late 1990s and early 2000s.  One of the powerful insights Dave Schaeffer had when he started Cogent in 1999 was that each strand of fiber would have “virtually unlimited” capacity over its lifetime due to the advances in optical technology.  This insight allowed Cogent to build and operate a massive global network on a single pair of long-haul fiber (plus denser metro fiber).  Its network now carries close to 30% of global internet traffic.  This insight also meant the companies who constructed the fiber plants faced massive long-term price degradation, which is why essentially all of them have been restructured over the last two decades.  Cogent was able to take advantage of the telecom meltdown in the early 2000s to acquire a handful of companies that could enable its network strategy, and to buy wholesale fiber leases at great prices.

Cogent’s network strategy is straightforward, but is worth discussing in some detail:

1. Acquire the infrastructure layer efficiently.  In Cogent’s case, until recently it assembled its infrastructure primarily through long-term IRU fiber leases from 328 suppliers.  Now, it’s purchased Sprint Wireline for $1 and is receiving $700 million of cash payments from T-Mobile, which certainly qualifies as cheap and efficient!  Cogent has the lowest-cost global network in the world. 

2. Shed all non-growth, non-commodity services from any acquired businesses.  Focus on scalable growth businesses with high incremental margins that can be run efficiently with a small network operations team. 

3. Optimize the topology of the network to become the lowest cost operator of network services in each end market.

Importantly, its go-to-market strategy is closely tied to its network strategy:

1. Offer commodity services at the best price/value in the market.

2. Sell services only in the highest volume locations in the market.

3. Use price and superior service levels to win share.

Cogent could not operate its desired go-to-market strategy without having first implemented its hyper-efficient and differentiated network strategy. 

At the risk of making this post too long, perhaps one of the best ways to illustrate Cogent’s network and go-to-market strategies is to walk through a case study of its corporate connectivity business in skyscrapers - what it calls Corporate On Net. 

Case Study: Corporate Office Buildings

For background, Cogent offers fiber-based internet and VPN services in nearly 1,900 skyscrapers.  These are <0.1% of total corporate office buildings, but 11% of total office space, totaling over 1 billion square feet.  Its buildings are connected by metro fiber rings in each market, typically into the equipment room of each office building.  However, Cogent goes one step further than many other operators in these buildings – it also pre-wires each floor through the vertical utility shaft (aka the riser), enabling efficient installation of services to end-customer suites.  In contrast, other providers may have to connect new customers through the riser on a bespoke basis if they hadn’t previously connected customers on that floor before.  This allows Cogent to guarantee a 17-day installation window (vs. 90 days for others) and average 13-15 days of installation time. 

Cogent combined its efficient long haul backbone architecture with a specialized, high-density metro network build to serve the best price per mbps of service in skyscrapers, offering 2.5-65x the speed for the same price as its competitors.  However, it didn’t build its whole footprint with greenfield capex.  In the early 2000s, the company implemented an aggressive M&A strategy, acquiring distressed telecom assets to turbo boost its organic strategy. 

Let’s walk through some important acquisitions that enabled Cogent’s success in the skyscraper market. 

In 2001, Cogent had connectivity in 166 office buildings with 65 million square feet of commercial space.  It also had options to build into 967 buildings with an additional 296 million square feet of space.  Cogent then commenced on a significant acquisition and restructuring campaign. 

• On February 4th, 2002, Cogent acquired Allied Riser Communications.  Formerly a public company, Cogent acquired ARC via a reverse merger, through which Cogent became the successor public company.  As of its last 10-K filed (in 2000), Allied Riser had built connectivity into 845 buildings with over 300 million square feet of office space.  It also had agreements to build into 1,950 building with over 450 million square feet of space. 

• On April 24, 2002, Cogent acquired OnSite Access, which brought another 16 million square feet and building access agreements to another 3,400 buildings totaling almost 1 billion square feet. 

• On September 12, 2002, Cogent acquired key assets from Fibercity Networks.  Cogent acquired the customer base and Fibercity’s building access agreements. 

Through these acquisitions, Cogent augmented its organically built business which brought more customers, more buildings, and more agreements to connect buildings in the future. It then got to cherry-pick the assets, customers, and employees it wanted to keep going forward.  

Walking through history, below is a summary table of Cogent’s network statistics, which shows about 12x growth in office buildings connected and a 15x increase in addressable office space from 2001 to 2023, supported by an >8x increase in metro rings and a 9x increase in metro fiber miles. 

Source: Company filings, Recurve Capital estimates

Cogent’s on-net office buildings are, on average, 41 stories tall and 542,000 square feet.  According to the company, there are about 93,000 tenants in its addressable market.  Cogent has about 26,000 on-net corporate customer connections, representing about 28% penetration in its buildings, though with many customers purchasing VPN and internet connectivity services, its unique customer penetration is considerably lower.  With superior price/value for its connection speeds and best-in-class service levels, historically Cogent has steadily increased its market share within this footprint.  It wins about 40% of its quoted proposals, suggesting attractive long-term share growth potential vs. its <20% unique customer penetration today. 

Let’s summarize what we’ve learned in the Corporate Office Building case study. 

1. Cogent built out significant infrastructure via acquisitions and organic network growth. It shed all undesirable parts of its acquired companies to focus on growth markets and to optimize for its desired network strategy.

2. The company pre-installs critical infrastructure so that it can provision services quickly.

3. Cogent prices its services at parity but delivers 2.5-65x more bandwidth than competitors.

4. With its win rate greater than its current market share in this footprint, Cogent has steadily gained share and stands to gain more over time.

Now, let’s transition to the new network Cogent’s is building to serve the Wavelength market. 

Cogent’s Wavelength Network Transformation

As detailed in the chart above, Cogent’s legacy network topology consisted of about 3,200 on net buildings (~1,900 office buildings and ~1,300 carrier-neutral data centers), all connected via ~70,000 miles route miles of long haul, inter-city fiber, and about 25,000 miles of metro (intra-city) fiber (route miles are unique miles, fiber miles indicate the number of strands x route miles).

Cogent had built its metro rings to run through both CNDCs and office buildings.  This was ideal since all its customers were buying access to the same service, i.e. the internet. 

Wavelengths are a different service which can benefit from topology that is optimized differently from IP networks.  Let’s walk through why:

1. IP traffic is carried via “best efforts” through networks according to the most efficient path given routing protocols and customer customization.  Packets arrive when they do, via whatever path makes sense at that time.  Sometimes traffic from Los Angeles to New York may traverse through Texas and Atlanta before making its way north.  Other times, it may go through Salt Lake City, Chicago, and then over to New York.  Buyers of internet connectivity don’t know (or care) about the path – they just want connectivity to everything out there, and they are fine with “best efforts” service levels.  Buyers of bulk internet connectivity get ubiquitous connectivity and pay the lowest price per mb. 

2. In contrast, Wavelengths are specific point-to-point pipes used for bulk data transfers that never touch the internet.  If a Wavelength customer wants to open a pipe between Los Angeles and New York, the path must be specified span by span.  Additionally, Wavelengths are negotiated on other KPIs, such as speed/capacity, latency, uptime, and of course price. 

Wavelengths are, by definition, a bespoke service because of the specifications they demand.  There could be (and are) popular routes that are desired by many customers, but they may not be the exact same endpoints.  Even in our LA to NY example, the details can be quite different.

Customer A might want connectivity from 1 Wilshire in LA to 60 Hudson St. in NY via Phoenix -> Dallas -> Atlanta -> Ashburn -> New York.  

Customer B might want connectivity from CoreSite LA3 to CoreSite NY1 via San Francisco -> Portland -> Seattle -> Chicago -> New York. 

Both are moving traffic between Los Angeles and New York, but the paths are different and the endpoints within those cities are different.  Not all providers can support those paths or those endpoints, though the most popular routes and sites tend to be well covered.  Most customers want multiple providers for each endpoint – a key difference vs. traditional connectivity markets, which have providers fighting for exclusive service contracts.

Cogent will be selling Wavelength services from its 800 CNDCs in the US and there are 799! / 2 unique routes within a network - more routes than there are stars in the universe. Of course, it is operationally and financially unviable to pre-install that type of connectivity (and this explains why the internet is so amazing – customers get all this connectivity and more at an incredibly low price that deflates in perpetuity). 

So, how will Cogent built a scalable network to address a bespoke service?  By innovating and bringing a new topology to the market.  As discussed above, Cogent built its network by connecting cities via long haul, inter-city fiber, and it connected its on net locations via metro fiber rings in those markets.  Everything was optimized for delivering internet services. 

Its legacy network looks something like this (forgive my art):

Legacy IP Metro Fiber Ring. Hub facility connects both MTOBs and CNDCs on the same rings. Source: Recurve Capital

There are hub locations with rings that connect both MTOBs and CNDCs along the same ring.

The Wavelength network will be operated 100% out of CNDCs and will run parallel to the IP network in the same facilities.  The fiber used to build this business will be segregated completely from the MTOB network.  This effort requires Cogent to build more rings, which is in process and is scheduled to be completed by year-end 2024. The modified metro rings will resemble this:

Wavelength Metro Fiber - MTOBs are separated from CNDCs. Source: Recurve Capital

Dave Schaeffer said at a recent conference that the network reconfiguration would result in 550 rings with MTOBs and 250 rings with CNDCs, suggesting over 3 CNDCs connected per ring. 

Zooming into a metro area like Atlanta, we can look at my terrible attempt at drawing rings:

Data center locations are from company data, rings are Recurve Capital’s illustrative examples.

The yellow icon is Sprint’s major switch location in Atlanta, where we know Cogent has connectivity with Sprint’s fiber and which is now being marketed as one of Cogent’s wholesale colocation facilities.  If this location acts as the Atlanta hub (not sure if this is true – again, this is illustrative), the company might build 5 or 6 rings from that hub to connect all the regional CNDCs together. 

Cogent is doing this work in every market across the country until all 250 CNDC metro rings are completed. As of the date of this post, the company claims it is about 40% complete (i.e. 100 of the 250 rings are configured.  

Cogent’s Wavelength network will connect all the metro-area hubs with wavelength connectivity (via line systems), and all the metro rings will be ready for rapid provisioning by merely plugging in some transponders at the terminating endpoints. 

Let’s put it all together and walk through an illustrative example of the workflow for provisioning a Wavelength.

Route: Digital Realty PHX1 (endpoint A) -> Albuquerque (hub) -> Dallas (hub) -> Memphis (hub) ->  Atlanta (hub) -> Equinix AT1 (endpoint Z)

Below we’ll try to show the conceptual steps on a few maps. 

Endpoint A:

Digital Realty PHX1 must be connected on a metro ring to the Phoenix hub (red line).

Illustrative metro ring and wavelength path (red) in Phoenix

Long Haul:

Inter-city long haul route: all the route’s hubs will be connected via a line system (Ciena RLS).  Full optical continuity will be pre-built between them.  Traffic moves hub-to-hub, likely without any need to traverse metro rings.

Illustrative path between PHX and ATL via the specified route

Endpoint Z:

The Atlanta hub is connected to Equinix AT1 via a metro ring (red line).

Illustrative metro ring and wavelength path (red) in Atlanta

The wavelength will be fully enabled for customers from Digital Realty PHX1 to Equinix AT1 when the four following steps are completed:

1. A technician in Phoenix plugs in a transponder in Digital Realty PHX1.

2. A technician in Atlanta plugs in a transponder in Equinix AT1.

3. Technicians at Cogent’s network operations center detect both optics, then configure everything already in place.

4. A technician must connect the customer’s equipment to Cogent’s.

Having the hubs pre-connected with line systems is analogous to having fiber connectivity pre-installed up the utility shafts in office buildings, as we discussed above.  Once this infrastructure is in place, provisioning services becomes less bespoke and standardized, using pluggable transponders and a technician workforce already accustomed to turning up services for customers across Cogent’s CNDC network. 

Due to the custom nature of Wavelength contracts’ negotiated technical attributes, it is unreasonable to pre-provision every potential route in the network.  However, Cogent has developed a network strategy that will enable the company to install and turn up new wavelength contracts in a two-week time frame.  To achieve those service levels and unlock the full capabilities of Cogent’s network strategy, its metro rings must be completed. Imagine if some are connected to others, but there isn’t contiguous connectivity around the whole ring. In that scenario (where Cogent is today), provisioning a Wavelength is much more similar to the bespoke workflow that common among other Wavelength providers. This is why Cogent is highly motivated to complete the ring reconfiguration so that it can then materially accelerate its pace of installations and, ultimately, Wavelength revenue growth.

It’s important to compare Cogent’s capabilities to others in the market.  Lumen and Zayo are the two largest providers of wavelength services in the market.  We have had dozens of conversations with competitors, customers, suppliers, and others in the fiber and Wavelength markets since the Sprint acquisition was announced.  The feedback has been very consistent: installation times are frustratingly unpredictable and a cause of significant customer dissatisfaction.  Average lead times seem to fall between 3 and 9 months, with some exceptions at both ends (e.g. we’ve heard 6 months quoted turn into 18 months).  Most believe installation times less than 30 days would be game-changing, although they also maintain a healthy dose of skepticism until they see Cogent perform at the levels it is targeting. 

While it is theoretically possible for the main providers in the market to mimic Cogent’s network strategy, it is difficult to overstate how fundamentally differentiated Cogent’s network strategy and returns-based corporate culture are when compared to its competitive set.

Let’s summarize Cogent’s network and go-to-market strategy in Wavelengths:

1. It is combining the Sprint fiber backbone assets with Cogent’s metro footprint to address the market. As in the Corporate Office market, it is using a combination of organic capex and acquisitions to build its hyper-efficient infrastructure layer.

2. Cogent is building a purpose-built Wavelength network to pre-wire connectivity in such a way that it can provision any location to any other location within two weeks.

3. Cogent will differentiate on route uniqueness, installation times, and price (as needed). Hyperscale customers care about price, but often it is not the most important parameter for them.

Closing

We have spent significant time and resources trying to understand and appreciate the magnitude of Cogent’s network transformation to help solidify our conviction levels in Cogent’s ability to succeed in the Wavelength market.  This is especially critical since the company missed its 12-month Wavelength revenue run-rate target by about 85% - an outcome that damages credibility.  The company’s explanation is that they believed there would be $100 million of demand for Wavelengths with endpoints that are among its connected market hubs (like the Phoenix hub and the Atlanta hub above), but instead, the orders Cogent received have endpoints deeper in the metro market, which require the rings to be completed to access with Cogent’s standardized workflow described above. This appears to be more a failure of investor communication than a disappointment on customer demand or the company’s ability to serve the market.

While it was disappointing for the company to miss the initial targets from a mark-to-market perspective for the stock, the net conclusion from our research suggests no material changes to our medium-term expectations for Cogent’s Wavelengths, although the path likely will be less linear than the company had expected.  If anything, our study on this topic has increased our confidence in the company’s probability of success.

We feel strongly that Cogent’s Wavelength network will compete well in the market for numerous reasons:

1. It brings unique routes to the market, improving diversity and network resilience for customers.

2. It brings more national capacity to the market.

3. It will bring more national CNDC ubiquity.

4. It likely will be cheaper than competitors on a like-for-like basis.

5. It can install faster.

Cogent should be able to win significant market share on dimensions #1 through #4.  We think it will also grow the overall market by adding new, diverse routes that were never available before. However, #5, powered by Cogent’s network transformation, should allow the company to gain share much more rapidly than otherwise.  Imagine if another provider quotes 9 months and Cogent quotes 2 weeks for a new contract.  These are not exclusive routes given the need for route diversity, but Cogent should win a large share of competitive situations by offering efficient and predictable provision timing.  Additionally, faster installations will allow Cogent to clear its backlog of 2,400 contracts quickly and efficiently once rapid provisioning is available across the network. 

Hopefully this Insight helped illustrate Cogent’s unique network strategy, which is built upon efficiency and specialization in growth end markets, and which drives much of the company’s go-to-market and share-gaining strategies.  It is in Cogent’s DNA to assemble the most efficient infrastructure layer and operate the lowest-cost services layer, and passing through those savings to customers by offering the best value in its markets. 

Cogent’s network transformation is not a trivial project.  It requires about half of Cogent’s workforce to execute thousands of unique and coordinated work efforts, some of which are technically difficult and others of which require scheduled customer outages to complete.  However, Cogent has absorbed other companies and has transformed its networks before, and every time it has emerged with more customers, more revenue, and a better network.  While difficult, the Cogent team eagerly confronts and accepts these challenging transformations to generate more value for Cogent.  If the company can achieve its medium-term revenue guidance, this transformation should be the most valuable of them all.