South Korea’s benchmark KOSPI index retreated below the psychologically significant 7,000 level on Friday, dragged lower by fresh selling in the country’s two dominant memory chipmakers after Chinese AI startup DeepSeek unveiled a model architecture that markets interpreted as a potential threat to high-bandwidth memory demand.
The KOSPI closed down 1.76% at 6,909.91, interrupting what had been a tentative recovery from July’s sharp sell-off and ending the week on a sour note. Samsung Electronics and SK Hynix each fell more than 3% on the session, extending a prolonged period of underperformance that has left both stocks more than 25% below their respective all-time highs.
Daily price swings of 5% or more have become a recurring feature of trading in both names, with realised volatility now running near levels last observed during the 2008 global financial crisis and the acute market stress of the 2020 COVID-19 pandemic. For a benchmark index as heavily weighted toward these two companies as the KOSPI, the ripple effects are immediate and pronounced.
Driving The Markets Lower
The immediate trigger for Friday’s move was an announcement from DeepSeek, the Chinese AI startup that has repeatedly unsettled global semiconductor markets this year, that its latest model employs a causal encoder-decoder architecture designed to operate with significantly less high-bandwidth memory and solid-state storage than competing designs.
Top Australian Brokers
- Pepperstone - Top Australian broker - Read our review
- eToro - Invest in ASX and international shares - Read our review
The implication markets drew was straightforward and damaging: if frontier AI models can be built and run more efficiently, the insatiable appetite for HBM chips that has underpinned the memory sector’s recovery thesis could prove less durable than previously assumed.
Samsung and SK Hynix are the world’s two largest producers of HBM, the specialised memory stacks that sit at the heart of the AI accelerator chips built by Nvidia and others. Any credible signal that AI model developers are engineering around memory constraints rather than simply buying more chips carries direct revenue implications for both companies. DeepSeek’s announcements have now triggered at least two significant sell-off episodes in Korean chip stocks this year, establishing a pattern that has made markets acutely sensitive to any efficiency-related news from the AI software layer.
Leverage Unwind
The scale of retail selling has been striking even by the standards of a market known for active domestic participation. South Korean retail traders have offloaded more than US$10 billion worth of Samsung and SK Hynix shares this month alone, while simultaneously unwinding positions in leveraged exchange-traded funds tied to the two chipmakers.
The speed and magnitude of these outflows point to a broader loss of confidence among domestic retail participants, many of whom accumulated positions during the sector’s AI-driven rally and are now exiting under sustained pressure.
Valuations: A Study in Contrasts
Against the backdrop of selling pressure and elevated volatility, the valuation picture for both companies has become increasingly difficult to ignore. Samsung currently trades at approximately 2.7 times book value, while SK Hynix commands around 5 times book.
Both figures compare starkly with the Philadelphia Semiconductor Index, which trades at roughly 11 times book. On a forward earnings basis, the divergence is equally pronounced: Samsung and SK Hynix are each priced at approximately 4 times forward earnings, against a global semiconductor peer group multiple of around 19 times.
Structural Risks Remain
The bear argument, however, is not simply about DeepSeek. It reflects a broader concern that the AI infrastructure buildout, which drove memory chip demand to exceptional levels, may be entering a phase of architectural optimisation rather than raw hardware accumulation. Hyperscalers and cloud providers have strong incentives to reduce per-model compute and memory costs, and the software industry has consistently found ways to do more with less over time. If that trend accelerates, the capital expenditure assumptions that underpin consensus earnings forecasts for Samsung and SK Hynix may need to be revised downward.
What Lay Ahead?
The near-term technical picture for both stocks remains fragile. Neither Samsung nor SK Hynix has managed to establish a sustained base above key support levels following July’s sell-off, and the recurrence of large daily moves suggests that price discovery is still ongoing. The 7,000 level on the KOSPI has now been tested multiple times in recent months, and Friday’s close below it will be watched closely by technical analysts as a potential signal of renewed index weakness heading into the following week.
Longer term, the investment case hinges on whether the structural demand for advanced memory in AI workloads proves resilient enough to offset the efficiency gains being demonstrated at the model architecture level. The inventory data and the transition to next-generation memory formats suggest the fundamental supply-demand balance is not as broken as current prices might imply. But until markets see concrete evidence that HBM order books remain intact through the next product cycle, the path of least resistance for both stocks is likely to remain volatile and sentiment-driven.
