美国人工智能雄心与核电现实相撞

The race to dominate artificial intelligence is no longer just about algorithms and semiconductors. It is increasingly about electricity. As AI data centers multiply across the United States, policymakers and energy analysts are warning that America’s power infrastructure — particularly its nuclear capacity — may not be growing fast enough to support the next phase of technological expansion. AI workloads, especially large-scale model training and inference, demand enormous and continuous power. Hyperscale data centers already consume more electricity than many small cities, and projections suggest that demand could accelerate sharply through the end of the decade. Energy officials say that without firm, carbon-free baseload generation, the U.S. …