Summary

Cogent’s legacy business is a deliberately narrow “dumb pipe” whose advantage is operational design, not exclusive technology. Its leased-fiber network reaches roughly 98% of global IP addresses from only about 3,400 buildings, concentrating on skyscrapers representing just 10 basis points of US corporate-building count but 11% of floor space. Pre-wiring every floor lets Cogent install dedicated one-gigabit service in nine days on average, versus potentially 90 days for an incumbent.
The Sprint wireline acquisition paired a deeply unprofitable services business with a scarce nationwide fiber network that Cogent effectively got paid to take. Cogent “paid a dollar” and is receiving $700 million from T-Mobile because removing Sprint Wireless as anchor tenant left the operation burning more than $300 million—and perhaps $400 million to $500 million—of annual cash flow. Chan views the inherited services as a necessary evil: eliminating negative-margin revenue can shrink sales while raising EBITDA because “you turn off $100 of revenue, but you turn off $200 of cost.”
The market’s skepticism is deserved because management repeatedly mapped Cogent’s internet playbook onto wavelengths and missed the resulting milestones badly. Management expected roughly a $100 million wave run rate about a year before the recording but missed badly; as Andrew put it, the miss “rounds to a 100% miss.” Chan argues management ultimately chose correctly to finish the complete network instead of diverting resources into bespoke early installations.
Cogent’s wave proposition is less about impossible technology than delivering route diversity, capacity and timing with a radically better operating model. Customers routinely multisource connectivity because a roughly $25,000 annual wavelength can protect a data center containing $1 billion of GPUs; Cogent can therefore add “another nine of resilience” without displacing Lumen or Zayo. Its preconfigured network is designed to provide consistent delivery within 30 days in an industry where on-time performance within 30 days may average only 10%–20%.
The data-center sale has slipped because Cogent initially sought premium pricing for unfinished shells, then decided to spend $100 million making them more turnkey. Chan’s reconstruction is that the portfolio might have fetched roughly $250 million as-is but potentially $700 million after investment; the process remains awkward because Cogent will accept bids for the whole portfolio, subsets or individual facilities. His concern is that value-maximizing CEO Dave Schaeffer might hold a $500 million offer for 18 months merely to seek $550 million.
Schaeffer is simultaneously the thesis’s central asset and its largest governance risk. Chan calls him brilliant and uniquely capable of “disrupting [telecom] in slow motion,” but Andrew’s pushback is that the “smartest man in the room” combined with corporate leverage and Schaeffer’s leveraged commercial-real-estate portfolio is a familiar setup for trouble. Chan attributes the CEO’s persistent stock sales to lender pressure after Silicon Valley Bank’s collapse and concedes that “it looks really bad.”
At roughly $48, the equity is a levered call on whether wavelengths become a real growth engine, not a conventional cheap-telecom trade. Andrew’s example used roughly $2 billion of debt and $2 billion of market value—about a $4 billion EV, although the transcript also says “$400 million”—against Schaeffer’s roughly $500 million 2028 EBITDA target, or about 8x on the $4 billion interpretation. Investors Chan speaks with model more than $10 of free cash flow per share; given Cogent’s payout policy, he raised the possibility of a $12 dividend. If waves fail, IPv4 addresses, data centers and a potentially multibillion-dollar fiber network may provide fundamental support, but Chan warns shareholders could still “eat a lot of downside” before those assets are realized.

Deep dive

1. Cogent built a growth business by accepting that telecom is a commodity

Chan’s high-level description is intentionally plain: legacy Cogent is “just an internet provider,” assembled from distressed telecom assets bought during the early and mid-2000s and rebuilt into a narrow, scalable global network.
Schaeffer’s contrarian 1999 premise was that voice, video and other specialized networks were not differentiated: the internet would become the lowest-cost network and everything would ride over it. Cogent therefore embraced being a “dumb pipe” decades before that became consensus.
Rather than own every strand, Cogent historically leased dark fiber from 377 providers, connecting metropolitan hubs through redundant rings and linking cities through long-haul routes. From roughly 3,400–3,500 buildings, Chan said it can reach about 98% of global IP addresses.

2. Pre-wiring lets Cogent sell commodity bandwidth as a superior service

Cogent targets skyscrapers averaging roughly 450,000 square feet and 50 floors—only 10 basis points of US corporate-building count, but approximately 11% of the addressable office space. The strategy is to “outpunch” the footprint rather than pursue ubiquity.
Spending about $10,000 per floor to pre-wire each building’s riser means a customer on floor 44 can be activated in nine days on average, against perhaps 90 days and construction-like charges from an incumbent that must pull new cable.
Chan contrasted Cogent’s roughly $700 monthly one-gigabit symmetrical dedicated service with an incumbent T3—which he described as about 45 megabits per second—potentially costing $1,000. Cable’s shared hybrid fiber-coax architecture can also create peak-hour congestion that a dedicated Cogent connection avoids.

3. T-Mobile paid Cogent to absorb Sprint’s stranded wireline operation

Cogent bought Sprint Wireline for $1 and is receiving $700 million from T-Mobile. Andrew also flagged roughly $500 million of IPv4 addresses, while the real prize is what Chan called the oldest fiber network in the United States, built nationwide.
The economics collapsed after T-Mobile migrated Sprint Wireless—the network’s anchor customer—elsewhere, leaving external customers unable to support the infrastructure. The stranded business was burning more than $300 million, and possibly $400 million–$500 million, of annual cash flow.
Before signing, Cogent reportedly mapped when every costly wholesale contract could be terminated and where off-net customers could move onto Cogent’s own facilities. That explains the counterintuitive financials: consolidated revenue declines while EBITDA rises as low- or negative-margin contracts disappear.

4. Management’s transit analogy created a large credibility gap

Chan’s criticism is direct: Schaeffer repeatedly used Cogent’s internet-transit history as the best available data set for forecasting wavelengths, even though “that mapping has been incorrect the entire time.” Schaeffer’s response to Chan was effectively, “I don’t have the data, so what do you want me to do?”
Management expected connecting the top 50–100 data centers to produce roughly a $100 million annual wave run rate. Yet the business remained small even after reaching 329 data-center endpoints, demonstrating that wavelengths lack transit’s strong 80/20 concentration and require far more route permutations.
The projected ability to install 500 waves monthly also ignored customer behavior. Buyers conditioned by six-to-13-month provisioning cycles neither believed Cogent could deliver in 30 days nor organized procurement around that speed, even after the technical capability existed.
Andrew’s market evidence: investors bought the anticipated Sprint inflection in Q2 and Q3 2024, helping take the stock from roughly $50 to $80, only to see it round-trip to about $50 by June 2025. “He’s been late on everything,” Chan conceded.

5. The wave network competes on diversity and certainty, not magic

Lumen, Zayo and others can provide wavelengths; Cogent has not invented an impossible service. The distinction is a modern, preconfigured network using the newest technology for wavelengths, without the incumbents’ “hodgepodge” of legacy voice, video, hardware, customers and cost structures.
These contracts are non-exclusive. A data center holding $1 billion of NVIDIA GPUs and depreciating perhaps $100 million monthly will readily pay around $25,000 annually for another independent wavelength if it adds “another nine of resilience.”
Installation certainty may matter more than raw speed. A former Zayo operations employee told Chan that the best on-time-delivery rate within 30 days he had seen was 50%, versus an industry norm of roughly 10%–20%; Cogent’s disruptive promise is, “When we give you a date, we can deliver on that date as promised.”
Chan said price comes last among the purchasing criteria and emphasized route diversity, resilience and dependable delivery. Cogent’s capital-light architecture nevertheless lets it supply those attributes at nearly zero incremental cost, improving its free-cash-flow economics relative to field-intensive incumbents.

6. Data-center monetization turned into a retrofit before a sale

Chan believes Schaeffer initially hoped to sell unfinished “shells” at AI-frenzy prices, like houses with plans but “no kitchen, no cabinets, no flooring.” When bids disappointed, Cogent committed about $100 million to retrofit the portfolio toward turnkey condition.
His illustrative economics were stark: perhaps $250 million for the assets as-is versus around $700 million after spending $100 million. Cogent now entertains whole-portfolio, partial-portfolio and single-site bids, making comparisons across overlapping offers and different completion levels unusually difficult.
Most sites are edge facilities around five megawatts, not giant AI-compute campuses. Chan nevertheless sees value in proximity to emerging secondary-market clusters; the roughly 2.5-megawatt Cheyenne facility was reportedly in demand because Microsoft and Meta facilities sit nearby.

7. Schaeffer’s brilliance does not cancel the key-person and leverage risks

Chan has known Schaeffer for 12–13 years and regards him as an unusually insightful operator whose Spartan offices, bad coffee and ruthless parsimony fit the business. Schaeffer chose telecom partly because legacy incumbents’ sunk costs let him exploit them while “disrupting it in slow motion.”
Andrew’s pushback—left unresolved—is that Schaeffer reportedly has roughly 40 direct reports, is a 70-year-old founder who “is Cogent,” and has repeatedly missed external milestones. Chan expects Schaeffer ultimately to sell the company and leave outright rather than remain under a buyer.
Schaeffer’s stock sales reportedly fund accelerated principal repayment on commercial real estate whose estimated value fell from $1.1 billion to $600 million while loan-to-value rose from about 50% to 65%. He owns roughly four million Cogent shares, takes no salary and receives around $16 million of annual dividends, mostly return of capital, allowing him to defer personal taxes.

8. Peak leverage makes the dividend a wager on near-term acceleration

Andrew cited headline leverage near 6.6x and a quarterly dividend of roughly $1 per share—a near-10% yield at $48—after 54 consecutive quarterly increases. His challenge: suspend it, delever and repurchase stock instead of preserving a potentially founder-friendly payout.
Chan described Cogent as “overlevered but not risky”: net debt is temporarily elevated by Sprint-network and data-center capex, while EBITDA is depressed by Sprint Core’s negative opex and a separately disclosed $50 million–$60 million operating expense that should bleed away through the end of 2026.
If no assets sell and waves still are not accelerating six to 12 months later, Chan expects Schaeffer and the board to revisit the payout, probably reducing rather than eliminating it. A recent $600 million issuance refinancing $500 million due in 2026 illustrates why a buyback, however accretive, would look aggressive today.

9. Moore’s-law deflation is buffered by contract structure and traffic mix

Andrew’s structural concern is that cost per delivered bit falls roughly 30% with Moore’s law; if usage stops rising, customers may demand the same 25-gigabit service for $500 rather than accept a free upgrade to 50 gigabits at $1,000.
In wholesale contracts, Chan said volume and price move together: without traffic growth, prices do not deflate as sharply. Traffic had been sequentially flat for two quarters and only 8%–9% higher year over year—around its historical low—yet still remained positive.
Chan thinks human-attention-driven growth has matured as streaming penetration rises and small mobile screens reduce the benefit of ever-higher resolution. His possible next leg is AI-driven machine-to-machine traffic, where usage becomes largely “disassociated from the human engagement,” though he framed that as a belief, not certainty.
Cogent’s standing offer to undercut wholesale competitors by 50% helped drive its progress toward approximately 25% share but is being used less aggressively. Its subtler tactic lifted customer-to-customer traffic from about 50% to 75%–80% by identifying routes where Cogent could eliminate an unpaid intermediary and split the savings.

10. Valuation works only if operating growth reaches the equity

Andrew’s simple math used $2 billion of debt and $2 billion of market value, which implies roughly a $4 billion enterprise value, against Schaeffer’s roughly $500 million 2028 EBITDA target, or 8x a three-year-forward number. The transcript also says “$400 million” for enterprise value, so that figure is internally inconsistent. The resulting valuation is not obviously cheap beside mature telecom and cable comparables.
Chan’s upside case is per-share cash flow: investors he speaks with model more than $10 of free cash flow per share, continued dividends in the interim and eventually perhaps a $12 annual payout. At today’s sub-$50 price, that would be a growth company carrying a roughly 25% dividend yield on cost.
Andrew’s pushback—worth keeping—is that much of this equity return comes from leverage amplifying a still-unproven rise from roughly $350 million of trailing EBITDA. Chan agreed: the stock is “basically just a call on, is this business going to work or not?”

11. The technical risk has receded; monetization is now the decisive test

A strong year for another provider might add 20 data centers; Cogent reportedly added around 700 in one year. Chan’s analogy is Starlink launching “a full constellation in one shot and it’s unloaded”—ubiquitous capacity arrives before the customers.
Amazon reportedly tested roughly 425 circuits and bought an entire line of fiber on the scarce Seattle-to-Chicago route after the network “passed with flying colors.” Chan cited one example of turning on 10 terabytes of capacity between endpoints within 30 days and elsewhere contrasted 20 terabits in 30 days; the transcript does not resolve that unit and quantity discrepancy. His categorical conclusion is: “The network risk is gone. It’s behind us.”
What remains is sales execution by Cogent’s seasoned strategic-account team. Chan thinks 90% route uniqueness, reliable delivery and basic demand for another resilience path could produce 15%–20% share, before considering any further advantage over slower incumbents.
Dark-fiber sales could bring a large upfront check—Chan illustrated perhaps $500 million nationwide plus $25 million annual maintenance—but waves might monetize the same network at roughly $1 billion of revenue. If demand is genuine, Cogent should retain the fiber and rent lit capacity into hyperscaler and AI growth instead.

YouMind Founder 玉伯: Waiting for the Tree to Bloom Vol.63 Tencent Music Is Being Too Generous: A Discussion of the Ximalaya Acquisition