The Memory Supercycle Is Back in Motion — and Samsung and SK Hynix May Be the Stocks With the Most Torque

The article argues the AI-driven memory chip supercycle is still running, with Micron up 565% in the past year and analysts reportedly seeing upside to $1,000 per share. It highlights SK Hynix as a high-bandwidth memory leader with 72% operating margins and Samsung as another leveraged play on the boom. For investors seeking exposure, it also points to the iShares MSCI South Korea ETF (EWY), which is up 189% over the past year and has 25%+ exposure to Samsung and SK Hynix.

Analysis

This is less a simple memory-cycle trade and more a supply-chain bottleneck trade with a duration problem: the market is still underestimating how long AI infrastructure can keep absorbing incremental DRAM/HBM capacity before supply catches up. That tends to favor the highest-quality beneficiaries of the constraint first, but it also means the second-order winners may be the firms with the strongest customer lock-in and the most direct exposure to AI server buildouts, not just the cheapest memory names. Nvidia’s relevance is indirect but important: if HBM remains tight, it reinforces Nvidia’s ecosystem power because system-level scarcity shifts bargaining power toward the GPU platform owner and away from downstream buyers.

The biggest risk is not a cyclical collapse so much as a sequencing shift. If capacity additions arrive faster than AI demand growth over the next 6-12 months, the margin peak will likely be compressed rather than punctured, which is worse for late entrants who pay peak multiples for peak earnings. A second-order risk is capital allocation: once memory makers are flush with cash, aggressive capex can sow the seeds of the next downcycle even while reported fundamentals still look pristine.

The contrarian view is that the trade may be too focused on named winners and not enough on the persistence of scarcity pricing. If HBM/SSD pricing normalizes but AI server volumes keep growing, the pure-play memory names could still outperform the broader semis while absolute returns flatten. The market may also be underappreciating cross-asset spillovers: a strong Korea-beta proxy can act as a more diversified way to express the theme than chasing single-name upside after a huge run.

Key Decisions for Investors

• Add NVDA on 3-6 month weakness as a beneficiary of persistent HBM scarcity; use it as a lower-beta way to stay long the AI supply bottleneck, with the thesis that constrained memory keeps the platform owner structurally advantaged.
• Prefer a basket approach over single-name momentum chasing: build exposure via Korean memory leaders or EWY on pullbacks over the next 1-2 months, but size smaller than a direct GPU long because the trade is more dependent on future multiple expansion than on near-term earnings surprise.
• Sell upside calls against existing memory exposure into strength over the next 4-8 weeks; after a 12-month vertical move, implied upside capture is likely worse than implied downside from any capex/capacity headline.
• Pair long NVDA / short a mature semiconductor hardware beneficiary with more limited AI content over 6-12 months; the second-order edge is that memory scarcity should widen platform winners’ moat while commoditized hardware names face margin pass-through pressure.
• If using EWY as a proxy, treat it as a tactical trade only: enter on broad Korea market weakness, set a 10-12% stop, and take profits on any additional multiple re-rating because the ETF dilutes the purity of the memory thesis.