OpenAI GPT-6 Astra boosts Samsung, SK hynix on AI memory demand

The rally in Samsung Electronics and SK hynix shares following the GPT-6 Astra launch points to a critical shift in AI memory demand. The new model’s agentic workflows, which involve complex, multistep tasks, consume substantially more memory, particularly HBM chips. This increased consumption is not just about processing more data but about the shift from token consumption to delegating entire tasks, as noted by Eugene Investment & Securities analyst Son In-jun. This change directly impacts the need for greater memory capacity and bandwidth, making HBM chips a more significant cost component for advanced AI models. The immediate impact is a rebound for Korean memory giants, whose stock performance had softened due to investor concerns about AI infrastructure spending. However, the underlying issue of memory chip shortages is becoming more acute. KB Securities’ Kim Dong-won reports that memory inventories at Samsung and SK hynix fell to less than 10 days’ worth in the third quarter, suggesting a potential critical limitation in available supply next year. This indicates that while demand is robust, the supply side faces significant constraints that could affect future growth. OpenAI’s development of its in-house AI chip, Jalapeño, which incorporates HBM4, further underscores the strategic importance of advanced memory. While OpenAI has not yet named its memory supplier, the move reflects a broader industry trend towards custom solutions that maximize memory efficiency for complex AI workloads. For Asian chipmakers, this means sustained pressure to innovate and scale HBM production, even as the market faces potential supply bottlenecks.