Memory chip scarcity keeps AI thesis intact: RBC Wealth Management

The incremental alpha is not in the scarcity story itself, but in its duration: a 6-12 month bottleneck means memory pricing can stay tight long enough to re-rate the entire AI supply chain, not just the top-line winners. The cleanest beneficiaries are the HBM leaders with the best qualification status and yield curves — effectively Micron and SK Hynix — while Samsung remains a laggard risk if it cannot convert capacity into shipped wafers fast enough.

Second-order, this shifts leverage away from GPU designers and toward component suppliers that sit on the critical path. If HBM remains constrained, AI system builders may face slower shipment ramps or higher bill-of-materials costs, which can cap margin expansion even as demand stays healthy; that matters most for names with the most aggressive AI-capex expectations. The more interesting setup is buying the suppliers on technical dislocations, because the flow-driven weakness in Korea looks disconnected from the fundamental backdrop.

The contrarian risk is that the market treats every supply shortage as durable, when memory cycles usually self-correct faster than consensus expects once capex, yield, and qualification catch up. The falsifier is a visible step-up in supply: HBM lead times normalizing, Hynix/Micron capex commentary implying faster output, or Samsung closing the qualification gap. On a 1-3 month horizon, the trade is still supported; on a 6-18 month horizon, the setup becomes much more two-sided as overbuild risk returns.