Puru Saxena argues that falling AI token prices paradoxically increase total computing demand through the Jevons paradox, as cheaper inference unlocks new agentic workflows and automation. While unit costs decline, total GPU, HBM, and infrastructure spending remain elevated due to surging volume, supporting AI hardware makers and infrastructure stocks despite margin pressures on hyperscalers.