This episode argues that today’s AI exuberance fits a familiar pattern: bubbles misallocate capital on the way up but leave behind productive infrastructure that powers the next S-curve. We revisit the “boom-bust-build-out” cycle and apply it to compute and the grid, note why “good enough” AI latency could flip capex behavior, and push back on modeling everything in gigawatts—useful for planning, risky for strategy. On companies: ARM’s quarter was solid, with rising royalties/CSS stickiness and a strongly implied first-party chip effort complemented by the DreamBig memory-controller acquisition to improve AI-era CPU roles. Qualcomm benefits from a higher-ASP Android cycle, nurtures auto/smart-glasses adjacencies, and eyes DC inference. On supply: Elon hedges silicon with foundry allocations while fab-building talk reads as negotiating leverage; memory/storage stay tight with longer contracts and measured adds. Net: optimism about what survives the eventual correction, caution on GW-only thinking, and watchpoints around ARM’s vertical creep, Qualcomm’s DC push, and supply-chain discipline.