AI memory demand lifts Samsung, SK Hynix 2027 outlook

This isn't a cyclical recovery—it's a structural squeeze. HBM consumes three to four times the wafer area of standard DDR5 per die. As Samsung and SK Hynix tilt capacity toward HBM3E/HBM4, conventional DRAM and NAND supply elasticity gets compressed. That is the real driver behind the Q4 2026 shortage, not mere demand growth. On the technical cascade, HBM's 3D stacking demands extreme TSV and hybrid-bonding yields, locking equipment vendors into longer order cycles. But consumer weakness is a hard cap: LTA price corridors prevent contract prices from fully reflecting spot tightness, capping upside before it materializes. Competitively, Micron has entered NVIDIA's HBM supply chain, yet the 2027 HBM4 production window is the true inflection. If Samsung establishes a generational lead in 1c-node and hybrid bonding, it reclaims pricing power; otherwise, SK Hynix's HBM head start erodes within 18 months. CXMT and YMTC anchor pricing at the low end, structurally suppressing global ASP elasticity over time. 12-to-24-month outlook: H1 2027 is the sweet spot. But if AI capex decelerates on weaker LLM monetization, H2 2027 risks a 2023-style inventory unwind. LTAs are shifting from volume-price linkage to volume-lock with price corridors—reducing earnings volatility, provided AI demand doesn't cliff.