Largest US Grid Will Curtail Power to Big Data Centers During Shortages Starting 2027

Skynet Chance (-0.02%): Physical energy constraints on large compute clusters act as a mild external brake on unbounded AI scaling, marginally reducing the odds of runaway capability growth outpacing oversight. The effect is small since curtailment is compensated, occasional, and easily worked around with on-site generation.

Skynet Date (+0 days): Power interruptions and higher energy costs slightly slow the buildout of frontier training infrastructure, nudging any risk-relevant capability milestones marginally later. The delay is minor because operators can self-generate and curtailment events are limited in frequency.

AGI Progress (-0.01%): Grid capacity is emerging as a hard bottleneck on the compute scaling that underpins current AGI trajectories, and this is the first major US grid to formalize cutting off AI data centers. It does not affect algorithms or research, only the physical substrate for scaling.

AGI Date (+0 days): Rising electricity prices, curtailment risk, and the need for costly on-site or diesel generation add friction and expense to data center expansion, modestly pushing large-scale training capacity later. Deep-pocketed frontier labs can absorb these costs, limiting the magnitude of the delay.