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

Memory prices are surging and have driven outsized gains in Micron and SK Hynix, but the article flags the inherently cyclical DRAM market. The current upside is attributed to AI-driven demand for HBM, with supply constrained by EUV equipment bottlenecks and longer capacity buildouts. Micron expects earnings to peak at $170.70/share in FY2028 (then fall to $121.77 in FY2029 and normalize around ~$50), while SK Hynix management says supply won’t catch up to demand until at least 2030; both firms are signing new 3–5 year long-term contracts, including SK Hynix’s reported $750B deals with $500B from Nvidia, which is positioned as extending the cycle.

Analysis

This is less a normal memory cycle than a supply-constrained pricing regime inside AI infrastructure. The economic rent should accrue to the HBM supplier with the strongest allocation discipline and best customer lock-in, so the relative winner is the name with the cleanest AI mix, not the one with the most legacy DRAM volume. That makes SKHY the sharper equity expression than MU; ASML is a quieter beneficiary because the bottleneck is now tool access and advanced packaging throughput, which keeps incremental supply expensive and slow. The secondary loser is the laggard memory incumbent with weaker HBM credibility: if it has to chase share with capex and lower pricing, it absorbs the worst combination of margin pressure and capital intensity.

Near term, the market can still rerate the group on every contract disclosure and every upward analyst revision, but the bigger catalyst is whether lead times and allocation remain tight into the next earnings season. The key falsifiers are a faster Samsung ramp, materially better HBM yields, or a sudden slowdown in hyperscaler capex; any of those would turn today’s scarcity premium into a classic oversupply setup. Over 6-18 months, the cycle likely cools only when EUV, clean-room, and advanced packaging capacity finally catches up, so the risk is not an immediate collapse but a delayed normalization that compresses multiples before profits actually roll over.

The contrarian miss is that long-dated supply contracts reduce volatility, but they also cap some upside convexity versus a pure spot market. That means the best trade is relative value, not blanket long semis: HBM-rich names should outperform commodity DRAM, but the whole basket may be over-owned if investors extrapolate peak margins too far out. For NVDA, the issue is less demand destruction than shipment friction and potential margin timing, which can matter even in a still-strong demand environment.

Key Decisions for Investors

• Long SKHY / short SSNLF for the next 1-3 months: best relative expression of HBM share gains versus a weaker legacy-memory mix; stop if Samsung commentary or capex points to a faster-than-expected HBM catch-up.
• Add ASML on pullbacks rather than chase the momentum: the trade works on a 6-18 month horizon as a bottleneck beneficiary, but upside is more backlog-led than spot-price led, so expect slower multiple expansion than SKHY.
• Avoid buying MU aggressively at current levels; use it as a hedge against a basket long in SKHY if you need sector exposure. MU is still a beneficiary, but its contract structure implies less upside convexity than the pure HBM winners.
• If holding NVDA, consider trimming into strength or overlaying short-dated call overwrites into earnings season: HBM scarcity can delay unit shipment timing even while AI demand stays intact, which can create temporary margin noise.
• Set a watch item on Samsung HBM yield/capacity commentary and hyperscaler capex guidance; if either improves materially, rotate out of the memory beta and keep only the quality leader exposure.