AI is described as weightless digital technology, yet it depends on enormous physical inputs. A single ChatGPT query consumes roughly 10 times the energy of a Google search. An AI-generated video? Approximately 10,000 times — equivalent to charging a smartphone 119 times.

Data center capacity is now measured in megawatts, not terabytes. Energy has become the single largest operational cost — and hyperscalers are responding with unprecedented vertical integration into power generation itself.

Water consumption tells a parallel story. Data centers rank among the top 10 water-consuming industries in the United States. Water-based cooling saves up to 40% of energy compared to conventional air conditioning — creating tension between energy efficiency and water scarcity.

Between September 8 and October 15, 2025, AI infrastructure firms signed deals worth over $500 billion — more than the GDP of most countries. This capex contributed roughly 1.1–1.2% to US GDP growth in the first half of 2025, exceeding Internet infrastructure spending during the dot-com era.

Energy constraints are now driving chip design decisions — closing the loop from power generation back to semiconductor competition, and creating a new geography where AI capacity concentrates wherever energy is cheap and abundant.