Current institutional consensus postulates that the proliferation of artificial intelligence will precipitate a compounding strain on global energy infrastructure. The most sophisticated of these forecasts are not naive linear extrapolations: agencies such as the IEA construct multiple demand scenarios spanning hardware efficiency, adoption rates, and supply-chain bottlenecks. Yet across that scenario range they share two structural commitments. They treat AI as an energy load whose magnitude is uncertain but whose relationship to the surrounding economy is exogenous, held ceteris paribus even as the scenarios vary the load itself; and they aggregate that load to national or global totals, holding the spatial topology of where it physically lands outside the frame. This paper argues that the adaptive responses most likely to bend the AI energy trajectory act precisely on those two excluded dimensions. Examining three under-modeled feedback loops — topological decentralization (edge computing), material-science transition (photonic processing), and macroeconomic demand destruction — it shows that these vectors introduce severe, non-linear volatility into infrastructural planning, and that the institutional baseline is best understood not as wrong but as a corner solution in which the adaptive variables are switched off. The central analytical pivot is the Jevons paradox: whether physical constraints can cap the rebound effect before it converts efficiency gains into new, geographically dispersed baseloads.This version reframes the paper as a decision-oriented synthesis rather than a rival forecast. It introduces a three-ledger framework connecting data-center electricity demand, distributed grid stress, hardware-efficiency rebound, and macroeconomic demand displacement, and adds a decision map of leading indicators for monitoring which AI-energy regime is emerging.
Alfredo De Joannon (Sat,) studied this question.