Are Index Funds Making the Market "Irrational"?

The cleanest expression here is not a direct bet on the memory producer, but on the capex tollbooth created by the race to add HBM capacity. When a concentrated supply chain has to reinvest aggressively just to keep share, the economic rent tends to migrate upstream to equipment and process vendors before it shows up in lasting equity value at the producer level. That favors AMAT more than LRCX: AMAT has broader wafer-fab exposure, while LRCX is much more exposed to the eventual memory normalization that tends to hit orders once pricing cools.

The IPO/ADR mechanism matters because it can create a temporary scarcity premium and a liquidity-event bid that is not the same as durable fundamental upside. In the first days to weeks, the move can be dominated by benchmark demand and momentum; over 1-3 months the real test is whether memory capex guidance stays elevated. If HBM supply expands faster than AI compute demand, the second-order loser is MU and any supplier with memory-heavy revenue, because margins compress just as the market prices in peak profitability.

Contrarian view: consensus is treating HBM as a structural winner, but the more likely regime is still a cyclical boom with a longer duration. The market may be underestimating how fast new supply can erode pricing once all three incumbents are forced to spend simultaneously. The upside case for NVDA is that a tighter memory chain preserves GPU attach rates and gross margin, but that support weakens if memory becomes abundant; the thesis is falsified if HBM pricing rolls over or if memory capex guides down over the next 1-2 quarters.