S15.1 · Telecom & Connectivity
Licensed mobile operators facing slowing mature-market ARPU growth, with AI repricing capacity demand rather than disrupting the core service model.
Wireless network operators sell licensed mobile voice and data service over spectrum, to consumers and enterprises. Start with the number that gets misquoted: the global mobile industry contributed $11.3T to global GDP in 2025 (GSMA Mobile Economy 2026), but that figure is downstream economic impact, not carrier revenue. Operator service revenue itself has historically tracked closer to $1.1T (GSMA Intelligence), and secondary reporting conflates the two constantly. The segment's real story is the gap between two businesses sitting on one balance sheet: durable, contracted infrastructure underneath, and a churn-exposed, capital-hungry retail service on top.
Service revenue growth in mature markets has slowed to 2-3% as the average-revenue-per-user uplift from 5G migration fades; Africa and South Asia still add subscribers at a high-single-digit clip. Revenue sits in North America, China and Europe; subscriber growth sits in Asia-Pacific and Africa. Within a country, three to four operators typically hold more than 90% combined share. Globally the market stays fragmented, because spectrum is licensed and non-fungible country by country — consolidation logic runs in-country, not across borders.
Operators buy spectrum, towers, network equipment and fiber backhaul upstream, and serve mobile virtual network operators, enterprise IoT deployments and consumers downstream. Net of capex, the retail business earns thin 30-40% EBITDA margins, carries churn risk, and is gated by spectrum licensing. The tower and fiber infrastructure carved out of the same balance sheets runs on long-dated, escalator-based contracts — a different economic animal entirely, and the two should never be read as one business.
AI hits mobile operators on the demand side, not the labor or product side. Rising AI-driven data traffic modestly reprices the network capacity operators must plan and build for; it does not restructure the business model. Agentic AI is compressing customer care — billing and support opex lines are shrinking — but that is an efficiency gain the retail P&L needed anyway, not a change in the competitive moat.
Two adjacency moves carry the current activity, and their money works differently. Fixed-mobile convergence bundling protects the base: tying mobile and broadband subscriptions together cuts churn, defending existing revenue rather than adding any. Direct-to-device satellite partnerships add coverage — and dollars — in geography operators could not otherwise reach, with no additional spectrum or tower capacity required, and those partnerships are already live. AI-RAN capacity economics, where inference workloads run on radio access network infrastructure, sit on a 2-5 year horizon.