How is AI hardware demand affecting data centre design, location, and construction timelines?

AI hardware demand is driving a fundamental redesign of data centers to handle gigawatt-scale operations, up from previous 100MW sites, with a structural shift requiring ground-up overhauls of power, cooling, and network architectures to support high-density, large-scale AI workloads that traditional setups cannot accommodate [1][2][6][9]. This includes escalated needs for storage and memory components, leading to supply chain strains and a focus on data-intensive infrastructure [7][10]. Locations are increasingly influenced by energy constraints, with 3-5 year waits for grid interconnections prompting strategic site selections near power sources and even acquisitions like Amazon's copper mine; however, US states are pushing back with moratoriums on expansions due to strains on energy, land use, and environmental resources, particularly in hubs like Virginia [1][2][11][12]. Construction timelines are severely extended by these demands, with build times of 18-36 months compounded by chip shortages, labor issues, and massive supplier backlogs—such as GE Vernova's orders filled until 2030 and Quanta's $40 billion queue—resulting in only 11-14GW of capacity online despite 40-50GW in development [1][2]. This scarcity risks underinvestment and limits global access to intensive AI computing [1][2].