Sam Altman's AI empire will devour as much power as NYC and San Diego combined (fortune.com)

OpenAI, in partnership with Nvidia, announced a build-out of AI data centers that could draw as much as 10 gigawatts of continuous power on a single project, with additional sites bringing the total to roughly 17 GW already underway—roughly the summer peak demand of New York City (≈10 GW) plus San Diego’s heat‑wave peak (≈5 GW), and comparable to the entire electricity demand of countries like Switzerland and Portugal combined. The rollout includes massive capital commitments—individual sites valued near $50 billion, about $850 billion planned overall, and Nvidia pledging up to $100 billion and millions of its new Vera Rubin GPUs—to keep up with surging demand (ChatGPT usage reportedly rose tenfold in 18 months). The announcement is a watershed for AI’s infrastructure footprint: experts warn the scale could materially affect regional grids (in Texas it’s comparable to ~20% of typical load), strain short‑term capacity, and force reliance on renewables, natural gas, and storage because new nuclear capacity won’t come online fast enough. Beyond carbon, risks include huge water and cooling needs, biodiversity impacts, and rapid hardware turnover producing toxic e‑waste. The technical and societal implication is clear: meeting AI’s compute appetite will require major changes in energy supply, grid planning, regulatory oversight, and sustainability practices—or else tradeoffs between growth and environmental/community costs.