Overview
- In mid‑June 2026, cities and regulators imposed pauses or stricter reviews on new facilities, including Seattle’s one‑year moratorium and Dublin‑area approval suspensions, because of concerns about grid strain and local resource impacts.
- A UN University report projects that by 2030 data centres could use about 945 TWh of electricity per year, have a water footprint comparable to the basic needs of 1.3 billion people, and require more than 14,500 km² of land.
- Industry and engineers are proposing integrated 'grid‑to‑chip' designs that combine higher‑voltage distribution, fewer power conversions and cooling matched to GPU heat to cut energy losses and manage extreme rack densities.
- Analysts warn that technical fixes and modular prefabrication speed deployment but do not remove binding limits; Gartner projects 40% of AI sites will face power constraints by 2027 and many planned facilities are already stalling.
- If the U.S. and other markets cannot secure predictable power, water and permitting timelines, investment and large AI clusters may shift to jurisdictions that can deliver faster approvals and infrastructure, with effects on jobs, local taxes and where models are trained.