DCI DCI provides the high-capacity, low-latency optical transport required to link hyperscale, regional and edge data centres supporting distributed AI training and inference.14 As AI workloads drive sustained east-to-west traffic flows between compute locations, DCI becomes a foundational enabler of scale and performance. Demand is driven by hyperscalers, graphics processing unit (GPU) vendors and enterprises running distributed AI models that require deterministic, high-throughput interconnects. This demand will intensify as AI becomes a more persistent, ambient and continuous exchange of tokens between compute nodes.15 IDC projects that global DCI revenue will grow at a 10.9% CAGR, reaching $5.5 billion by 2029.16 For CSPs, the DCI opportunity is shaped less by service differentiation and more by infrastructure scarcity. Ownership of long-haul fibre routes and power-dense sites creates sustained pricing power as demand for AI-driven interconnect outpaces supply. Monetization is therefore anchored in multi-year capacity contracts and premium service level agreements (SLAs) for guaranteed latency and availability, with use- based and SLA-tiered models emerging as traffic patterns mature. CASE STUDY 2 Verizon partnering with Amazon Web Services to provide fibre for AI Verizon Business and Amazon Web Services (AWS) announced a significant new Verizon AI Connect agreement to build long-haul, high-capacity fibre pathways linking multiple AWS data centre locations. Under the deal, Verizon will construct resilient, low-latency fibre routes designed to support the performance and reliability demands of advanced AI workloads, especially generative AI and other data-intensive applications. The expanded fibre network strengthens Verizon’s strategic relationship with AWS and reflects the growing need for dedicated high-performance infrastructure to scale cloud-AI services globally. CASE STUDY 3 Cellnex: activating existing infrastructure for AI-optimized edge colocation Cellnex is harnessing its extensive infrastructure footprint – including tower sites, fibre networks and existing data centre facilities – to deliver AI-optimized colocation at the edge. By repurposing carrier-neutral edge data centres18 and modular sites, Cellnex enables low-latency, scalable compute closer to users. Partnerships, such as the one with Vapor IO, accelerate deployment of edge compute for real- time AI workloads, including vision systems, robotics and autonomous applications.AI-optimized colocation AI-optimized colocation adapts traditional data centre hosting to AI workloads’ power, cooling and latency requirements. By retrofitting existing sites with higher-power density, cooling and low-latency interconnects, CSPs can enable enterprises to deploy GPU clusters and AI servers within telco- operated facilities, bringing operators into the AI infrastructure supply chain while monetizing underused real estate. Demand is driven by enterprises seeking compliant, power-dense environments for AI training and inference without the cost, time and permitting complexity of building new facilities. IDC projects the global colocation services market will grow from $56.4 billion in 2024 to $81.2 billion17 by 2029 (7.2% CAGR), with AI-ready capacity becoming increasingly scarce in regions where power and permits are constrained. CSPs’ right to play is strongest in regional and edge colocation rather than hyperscale campuses. While global providers are likely to remain dominant, telcos’ fibre-connected sites, local compliance positioning and enterprise relationships support a differentiated role. Monetization follows fixed-term colocation contracts with premium pricing for AI-specific capabilities such as cooling density and latency assurance, unlocking incremental value from existing property portfolios. The Strategic Role of Telecom Providers across the AI Value Chain 11

The Strategic Role of Telecom Providers Across the AI Value Chain 2026