ColdFusion

Why Building AI Data Centres Isn’t Working Anymore

The video examines the exploding $650 billion AI data‑center build‑out slated for 2026 and the longer‑term goal of $9 trillion by 2030, highlighting that a large share of announced projects in the United States are stalled or cancelled. Bloomberg and FT data show only about one‑third of the roughly 140 planned facilities – equivalent to 12 GW of compute – are actually under construction. Supply bottlenecks in power, transformers, GPUs and skilled labor, together with soaring equipment costs, are causing chronic delays, while satellite imagery confirms many sites remain empty. Concrete examples include Microsoft’s Fairwater campus, which is far from completion, and Texas’ $1 billion incentive for the Stargate project that has yet to break ground. Communities report infrasound noise, diesel‑generator emissions, and water‑use spikes comparable to a small city, yet receive little tax revenue because of long‑term abatements. Analysts warn the massive capital outlays may become a sunk‑cost bubble as hyperscalers run out of cash and investors question profitability. Growing local opposition and potential regulatory scrutiny could further slow the sector, forcing a reassessment of AI infrastructure financing.