Is the Memory Supercycle Peak Near for Micron and SK Hynix?

AI-driven memory demand is keeping DRAM upcycle momentum: Micron and SK Hynix are benefiting from a structural shift to HBM, with memory supply still “undersupplied.” SK Hynix signed long-term contracts worth $750B (including $500B with Nvidia), and Micron has non-cancellable deals with 16 customers running through 2030 (about 40% of revenue), which should smooth the cycle versus past supercycles. Despite peak-earnings expectations (Micron EPS peaking around $170.70 in FY2028 before declining), management sees DRAM supply not catching up to demand until at least 2030, supporting a longer runway for these stocks.

Analysis

The market is probably underestimating how much the profit pool is shifting upstream from generic DRAM into scarce HBM capacity and, just as importantly, into the toolchain that gates that capacity. That helps MU/SKHY now, but the bigger second-order winner is ASML: when both logic and memory are competing for the same EUV bottleneck, memory pricing becomes less about “a cyclical commodity” and more about a constrained industrial capacity market. The real loser is any buyer whose AI rollout economics depend on falling memory costs; if HBM stays tight, cloud and server OEM gross margins get squeezed before end-demand does.

The consensus mistake is treating long-term contracts as purely bullish. They stabilize revenue, but they also reduce spot upside and create a false sense of permanence: if the supply chain finally catches up in 12-24 months, the adjustment can be abrupt because the contract book masks underlying demand softness until it doesn’t. The key falsifier is a visible easing in packaging/EUV bottlenecks or a step-up in capex from Samsung/Micron that narrows the HBM shortage faster than 2030 guidance implies.

Near term, the trade is momentum-plus-short-squeeze in the memory complex; over 6-18 months, the better expression is owning the constrained enablers rather than the most cyclical silicon. NVDA benefits from HBM scarcity as long as it preserves AI system pricing power, but if memory inflation starts to slow shipments, NVDA can still digest it better than the memory names themselves because it owns the higher-value layer. The risk-reward is best on relative value, not outright beta, unless one is explicitly playing for another earnings revision cycle higher.

Key Decisions for Investors

• Long SKHY / short SSNLF for 3-6 months: express HBM leadership versus a late-mover catch-up story; thesis breaks if Samsung closes the HBM gap or wins share in next-gen packaging.
• Add on pullbacks to MU only as a tactical long, not a core hold: the setup is strongest into revisions over the next 1-2 quarters, but trim if management commentary starts implying margin normalization earlier than consensus.
• Overweight ASML on weakness as a lower-beta beneficiary of the same capacity bottleneck; best held 6-18 months as long as EUV remains the gating factor for both logic and memory expansion.
• Use NVDA as a relative winner, but prefer it over the memory suppliers only if AI capex remains intact; watch for any sign that memory inflation is delaying rack deployments or forcing OEM discounts.
• Set a watchlist alert for Samsung/Micron capex and HBM yield commentary: if either shows a meaningful step-up, fade the memory rally because the market is likely pricing a shortage that could normalize faster than expected.