These 2 AI stocks are getting crushed: Goldman Sachs says buy the dip

Samsung is down roughly 27% over the past month and SK Hynix about 36% as investors reassess memory-price prospects amid Chinese competition and shifting hyperscaler AI budgets. The selloff signals a cooling in AI-linked memory demand assumptions and increases downside risk to sector earnings visibility.

Analysis

This looks more like a regime shift from scarcity pricing to cyclical pricing than a one-off de-rating. The key market mechanism is that AI-related memory no longer gets a blank check multiple once investors believe hyperscaler capex is becoming more selective; in that case, earnings beta to spot DRAM/HBM pricing reasserts itself and the highest-multiple names get hit first. SK Hynix likely carries the most operating leverage to any slowdown in HBM pricing, while Samsung has more balance-sheet and product diversification, so the selloff is compressing the AI premium faster than the underlying fundamentals may justify.

Second-order effects matter: cheaper memory is a mild margin tailwind for server OEMs, handset assemblers, and any downstream hardware name with memory-heavy BOMs, but that benefit only matters if end-demand holds. If the current move is really a capex reset at the hyperscalers, then the demand shock will outweigh the input-cost relief and the whole AI hardware stack gets multiple compression. In the next 1-3 months, watch memory contract pricing, inventory days, and guidance from US cloud vendors; those are the catalysts that can either validate the de-rating or trigger a sharp squeeze.

The contrarian view is that the market may be extrapolating a cyclical pause into an AI demand collapse. HBM is still structurally important for training and increasingly for inference, so the long-run story is not broken unless hyperscaler budgets roll over for multiple quarters. That said, the 6-18 month risk is that Korea’s export-sensitive equity complex and won stay under pressure if memory pricing weakens, which argues for relative-value rather than outright bottom-fishing until estimates reset.

Key Decisions for Investors

• Short SKHYV on any bounce or via 1-3 month put spreads; best expression if you believe HBM pricing and hyperscaler capex are still rolling over. Falsify the trade if contract pricing stabilizes or a major cloud vendor raises AI capex guidance.
• Pair trade: long SSNLF / short SKHYV over 3-6 months. Samsung has more diversification and should outperform if the issue is a broad memory downcycle rather than an AI secular break. Risk is that Samsung’s HBM catch-up disappoints and relative multiple gap closes the other way.
• Use SMH or SOXX puts instead of single-name exposure if you want to express a broader read-through to semiconductor multiple compression without Korean ADR liquidity risk. Time horizon: next earnings season into early 2026.
• Set an alert on next-quarter memory pricing and inventory commentary; if pricing turns up before the next guidance cycle, cover shorts quickly, as the market is likely pricing a deeper downturn than fundamentals warrant.