Particle.news

Download on the App Store

AI Data Center Boom Faces Bubble Warnings and 'Phantom' Demand

DOE says the grid is unprepared for rapidly rising AI loads.

Overview

  • Goldman Sachs projects data center capacity to grow about 50% by 2027 and electricity use to rise roughly 165% by 2030 to 3–4% of global power, with most incremental supply expected from natural gas.
  • The bank is monitoring for a potential bust if AI demand or monetization lags, even as hyperscalers pour capital into larger, more power-hungry systems.
  • Utilities report speculative multi-site interconnection requests that inflate pipelines, with Oncor citing 552 active projects including 186 GW tied to data centers and AEP noting 24 GW of firm new load plus about 190 GW seeking to connect.
  • DOE estimates U.S. data centers could reach 6.7%–12% of national electricity consumption by 2028 and says the grid is not prepared, while planners face information gaps as roughly 84% of LLM traffic lacks environmental disclosure.
  • Researchers call for shifting from brute-force scaling to resource‑aware design and tighter coordination across chips, software and facilities, as some operators explore on‑site fossil and nuclear generation to meet near‑term demand.