Dell'Oro projects that approximately 200GW of new data center power will be needed globally between 2026 and 2030, increasing total capacity from 50-60GW in 2025 to approximately 250GW in 2030, with 60-70% concentrated in North America. Power availability has become the primary hurdle for industry growth and revenue recognition—power, licensing, site interconnectivity, and customer readiness are collectively determining when equipment can be accepted and revenue recognized. In other words, the bottleneck for AI infrastructure has shifted from chip procurement to power deployment, and equipment vendors' revenue trajectory is constrained by grid progress. To address this constraint, third-party research firm Dell'Oro predicts the industry will gradually reduce its reliance on the traditional power grid: on-site power generation could reach approximately 30% of total power deployment in the next five years, as hyperscale vendors pursue speed to market and attempt to circumvent regulatory bottlenecks in grid expansion.
At the network architecture level, Dell'Oro believes that cross-regional expansion solutions can partially alleviate power constraints by allocating AI computing power across geographically dispersed campuses. They judge that Arista's Jericho-based architecture has a clear competitive advantage in long-distance deployments and is a driving force for TAM expansion and new customer acquisition. HPE Juniper, Cisco, and Nokia also maintain a strong position.