AI Booms and Industrial Cycles: Capacity Reallocation in the Memory Industry

Abstract

Artificial intelligence (AI) booms can generate industrial price cycles even when frontier demand remains permanently elevated. I develop a two-product model in which high-bandwidth memory and conventional dynamic random-access memory (DRAM) compete for shared productive capacity, while investment arrives through a construction queue. Frontier demand initially diverts resources from conventional production. I derive a condition under which sufficiently responsive investment and costly sustained specialization subsequently expand conventional supply above its initial level. A second result maps installed capacity, inherited allocation, and committed deliveries into the entire supply path over the next commissioning window, conditional on demand. Evidence combines 68 fiscal quarters of Micron financial data, a 14-quarter fixed-supplier DRAM panel, paired business-unit price and shipment disclosures, and dated investment and contract commitments. The paired evidence identifies recipient expansion and donor restriction; joint inference places the financial investment response around two to three quarters after revenue growth. In the normalized permanent-demand experiment, the conventional price peaks 10.22 percent above its initial level and settles 0.81 percent below it. The reversal persists with current-order feedback. A temporary demand expansion generates a larger subsequent supply release, whose amplitude depends strongly on expected duration. Product composition and committed capital jointly determine how a technology boom creates and resolves scarcity across connected markets.

Author abstract from the research manuscript. Typography and line breaks have been normalized; the linked source may show a different manuscript version.