SK Hynix's HBM Empire Powers 65% US Revenue — Is This the Must-Own AI Stock?

SK Hynix is trading less like a cyclical memory name and more like a toll collector on the AI buildout, but that premium only works while HBM remains the bottleneck. The first-order winner is obvious, yet the second-order effect is that NVDA also benefits if HBM supply is expanding just enough to keep GPU shipment growth unconstrained; any memory shortage that slows accelerators would show up first in NVDA lead times, not just in Hynix revenue.

The bigger competitive read-through is to SSNLF and MU. If Samsung closes the HBM gap faster than expected, the scarcity rent that supports SKHY’s margins can compress quickly; MU participates in the broader DRAM upcycle but is less exposed to the AI scarcity premium, so it should lag if the market starts valuing HBM share rather than generic memory beta. Over 6-18 months, the main risk is self-inflicted: the industry is likely to overinvest into capacity once pricing stays firm for a few quarters, which usually marks the top of the cycle.

The consensus is missing that revenue concentration is not diversification by geography; it is a concentrated bet on U.S. hyperscaler capex and one platform ecosystem. The near-term catalyst is HBM4 qualification and any capex commentary from cloud customers over the next 1-3 months; the falsifier is a slowdown in NVDA order momentum or a meaningful share gain announcement from Samsung. If the stock keeps rerating without evidence of sustained HBM shortages, the move is getting ahead of fundamentals rather than discounting a durable moat.