JPMorgan Maintains Overweight Ratings on SK Hynix and Two Other Memory Chipmakers
nashnova research
JPMorgan reiterated Overweight ratings on SK Hynix, Samsung Electronics, and Kioxia, arguing that a ~45% drawdown from 2026 highs has made valuations attractive — though a lack of near-term catalysts may keep shares range-bound.
Down 45% — why is JPMorgan still bullish?
Memory stocks have pulled back roughly 45% from their 2026 highs, pushing 2027 forward P/E below 3.5×.
This means → the market has already priced in concerns over AI-financing sustainability, HBM spec downgrades, and a cycle peak — all at once.
JPMorgan's three-month preference order: SK Hynix > Samsung > Kioxia, with price targets of $260, ₩400,000, and ¥43,334 respectively.
Q3 earnings beat expectations — but each company has its own headache?
Q3 2026 results came in above consensus: EPS and margins beat despite FX headwinds, driven by pricing gains, successful HBM price negotiations, and smooth shipments.
Samsung's weak spot is the absence of a near-term shareholder-return event; Kioxia's is soft Q4 guidance. In plain terms = Samsung lacks a dividend catalyst, and Kioxia's next-quarter outlook disappointed.
JPMorgan views both as temporary — Samsung's shareholder-return catalyst is expected to return in January 2027, and Kioxia's Q1 2027 growth should run above seasonal norms.
Demand is splitting — who is spending and who is pulling back?
B2B: Hyperscale cloud operators (Microsoft, Amazon, Google and peers) continue to grow AI capex steadily; consensus estimates for next year have been revised upward.
B2C: PC and smartphone segments face greater headwinds.
This reflects a market where AI infrastructure investment remains the core demand engine, while consumer electronics has yet to pick up the baton.
DRAM pricing beat again — and HBM4 is already shipping?
Q3 DRAM average selling price rose 23% quarter-on-quarter, above JPMorgan's prior estimate of high-teens to 20%.
Q4 bit-shipment guidance added another surprise: vendors guided double-digit QoQ growth, pointing to roughly ~20% YoY bit-shipment growth by year-end 2026.
On HBM4 — the fourth generation of high-bandwidth memory, faster and denser than HBM3E — vendors have begun shipping, and expect HBM4 to surpass HBM3E in bits by Q3 and in revenue by Q4. JPMorgan forecasts a 96% rise in HBM ASP in 2027.
Could memory actually sell out?
Some vendors say 2027 memory supply is effectively sold out, with demand expected to exceed supply beyond 2027.
DRAM supply-demand is projected to stay tight through 2028; over 75% of 2027 fiscal-year capacity is already locked in.
This means → if these calls prove right, memory makers' pricing power will strengthen for at least two more years, with room for margins to expand further.
Is NAND really safe from a price downturn?
Despite market fears of a NAND price downcycle as early as Q4, JPMorgan sees no near-term signs of price cuts or a downcycle.
Kioxia's weak December-quarter guidance stems from pull-forward demand in the September quarter, with some orders pushed to the March quarter — a company-specific timing issue, not an industry inflection.
Structural support comes from enterprise SSD (eSSD) demand driven by KV-cache offloading — a technique that moves part of an AI model's data from expensive DRAM to cheaper flash storage — making cost-effective NAND solutions attractive to cloud clients optimizing AI workloads.
Long-term agreements (LTAs) continue to expand: Samsung and SK Hynix have extended some contracts past 2030, and Samsung targets LTA coverage of 60%–70%. In plain terms = more and more capacity is locked in by long-term contracts, giving vendors greater earnings stability through the next cycle.
市场有风险，内容仅供研究参考，不构成投资建议。