DeepSeek’s HBM claim rattles Samsung Electronics and SK Hynix
Samsung Electronics (005930.KS) and SK Hynix (000660.KS) shares fell 3.5% and 2.2% respectively on 11 September 2026 after reports that DeepSeek's AI models require less high-bandwidth memory (HBM). Investors are concerned about potential reduced demand for HBM, but experts note that the efficiency gains are specific to certain workloads and do not necessarily indicate a lasting change in global HBM demand.
How this was made

The 30-second read
Why it matters
The article explains why the share declines are a reaction to perceived demand risk rather than a confirmed market shift.
Market read
The story highlights a short‑term price reaction for two Korean memory chip makers, with limited broader market implications.
What to watch
DeepSeek's improvements target only KV‑cache; weight and activation memory still require substantial HBM.
Background
DeepSeek announced AI model techniques that lower KV‑cache memory pressure, prompting speculation about reduced HBM demand.
Ticker impact
Samsung Electronics shares fell 3.5% on 11 Sep 2026 after reports that DeepSeek's AI models require less HBM, raising concerns about near‑term HBM demand.
Potential further downside if additional AI efficiency news emerges; rebound possible if HBM demand data proves resilient.
The price move is driven by market speculation, not a confirmed demand shift; sentiment remains fragile.
SK Hynix shares dropped 2.2% on 11 Sep 2026 following the same DeepSeek HBM‑efficiency claim, prompting worries about its HBM sales outlook.
Likely to stay pressured unless the company provides clarification on HBM demand or new orders.
The move reflects reaction to a speculative narrative rather than a concrete order book change.
Market effects
AI‑accelerator memory market may see short‑term volatility; HBM suppliers could face pricing pressure.
South Korean semiconductor stocks likely to underperform in the near term.
Limited to memory‑intensive AI hardware segment; broader market impact minimal.
Counterpoint
Efficiency gains could boost overall AI usage, ultimately increasing total HBM consumption (Jevons paradox).
Key entities
- CompanyDeepSeek
AI model developer reporting reduced KV‑cache memory usage.
- CompanySamsung Electronics
Major HBM supplier whose stock fell on the news.
- CompanySK Hynix
Another leading HBM producer impacted by the same news.



