Cogent at Bank of America conference: ai demand offsets a hard reset

Cogent said wavelength revenue grew 62% year over year and 9.2% sequentially, supported by AI training demand that now represents roughly 25% of the wavelength market. The company sold 10 converted data-center sites for $225 million, or $4.3 million per MW, eliminating about $7 million of annual negative EBITDA drag; it expects similar pricing for 14 additional sites with 54 MW as it works to reduce debt. Offsetting the AI tailwind, Cogent acknowledged its 3% inner-city wavelength share remains well below its 25% target, the mid-2028 target is no longer realistic, and supply bottlenecks—including fivefold memory-price increases and telecom-equipment lead times of up to three years—are delaying deployments and customer acceptance.

Analysis

CCOI’s equity case is shifting from a clean AI-connectivity growth narrative to a balance-sheet execution trade. The remaining asset-sale program could materially extend liquidity runway and remove carrying costs, but assigning the first-sale valuation to the balance is premature: these are non-standard, low-density facilities, and the buyer pool narrows if power upgrades or inference demand prove less durable. Until signed transactions close, the market should capitalize only a discounted portion of the implied proceeds.

Near term, delayed customer acceptance creates a divergence between booked demand, installed capacity and recognized revenue. Management’s reluctance to enforce take-or-pay terms protects a fledgling product franchise but effectively grants customers an option to defer, raising the risk that reported wavelength growth decelerates before supply constraints clear. This is especially consequential for a levered issuer: capex and equipment-cost inflation can absorb operating leverage well before AI volumes convert into cash EBITDA.

The better liquid AI-network beneficiaries may be LUMN and private Zayo rather than CCOI: incumbent density, installed customer credibility and existing routes should capture the first wave of interconnect demand with less go-to-market friction. VZ and T have optionality but also face a capital-allocation test—incremental fiber builds should be judged against wireless and fiber-to-the-home returns, making a broad valuation rerating unlikely absent disclosed contract economics. Hyperscalers retain bargaining power; connectivity is mission-critical but remains a small fraction of their AI capex, limiting suppliers’ ability to pass through equipment inflation.

Contrarianly, AI transport demand may be underappreciated over 6-18 months because geographically dispersed power availability structurally increases data movement. But CCOI’s long-run share aspiration should not drive valuation until service activations and revenue per wave prove that its route-diversity advantage is monetizable rather than simply technically differentiated.

Key Decisions for Investors

• Maintain CCOI as a watch/underweight rather than add on conference commentary. Upgrade only after binding sales for the remaining facilities close and quarterly levered FCF improves materially; thesis is falsified if proceeds are materially below the first transaction or net debt does not decline by the expected amount.
• For 1-3 months, express the AI-connectivity theme through a relative long VZ / short CCOI position, sized modestly. VZ has a stronger balance sheet and can monetize selective enterprise builds; CCOI bears the greater risk from deferred acceptances and refinancing sensitivity. Cover if CCOI reports sustained sequential wavelength growth with accelerating recognized revenue and positive FCF trajectory.
• Monitor CRWV and NBIS data-center commissioning milestones as leading indicators for CCOI order activation, not merely order intake. Delays in GPU delivery, power energization, or site openings are a near-term negative for CCOI revenue recognition but signal persistent bottlenecks that can support pricing for established fiber providers.
• Avoid extrapolating AI-driven transport demand into AMZN, GOOG, META or MSFT earnings. Their network spending is more likely to pressure infrastructure capex and modestly support depreciation than move consolidated margins; a trade becomes actionable only if company disclosures show material leased-line cost escalation or a shift toward owned long-haul fiber.