Cogent Communications (CCOI) investor relations material

Data center sales and strategy

• Recently sold 10 data centers with 53 MW of installed power for $225 million to I Squared Capital, helping reduce leverage and enabling refinancing of 2027 notes.

• Net proceeds from the sale are expected to have minimal tax impact due to available NOLs and bonus depreciation.

• Remaining 14 data centers have 56 MW of gross inbound power (about 40 MW usable), and the company is open to various sale structures, including single or portfolio transactions.

• Data centers are considered non-core, generating only about 3% of revenues, and the company is willing to monetize additional facilities if attractive offers arise.

• I Squared Capital plans to use the acquired data centers for AI inference, leveraging their distributed footprint for low-latency applications.

Capital expenditures and equipment costs

• CapEx in the last quarter was $46 million, higher than expected, driven by DRAM shortages, supply chain disruptions, and increased competition from hyperscalers.

• Equipment vendors have implemented multiple price increases, making the previous $25 million per quarter CapEx target unrealistic for the near term.

• The company expects equipment costs to eventually revert to historical declines, but elevated spending will persist for at least the next year.

• Capital as a percentage of revenue remains around 10-12%, and price declines for internet bits have moderated industry-wide.

Wavelength business performance and outlook

• Wavelength business grew 90% year-over-year from a small base, now holding 3% market share after 18 months of active sales.

• The company offers a robust portfolio with over 1,100 carrier-neutral data centers enabled for Waves, providing unique routes and high reliability.

• 90% of routes have no competition on the right of way, and the network is more reliable than competitors due to deeper, armored cable placement.

• AI and hyperscaler demand is driving market growth, with 25% of Wave buyers being hyperscalers; training uses wavelengths for low latency, while inference is expected to be more distributed.

• Customer feedback on Waves is positive, with high marks for reliability, speed to install, and breadth of footprint.