📰 Key Takeaways

Samsung and SK Hynix announced an investment of approximately $51.8 billion to build four new memory fabs in southwestern Korea — a region traditionally underserved in semiconductor investments — plus an additional $520 million to build an HBM (High-Bandwidth Memory) packaging center in central Korea. The plan was officially unveiled during a national investment briefing chaired by President Lee Jae-myung, with SK, GS, Naver, and other tech and energy giants simultaneously committing to invest $35.6 billion in AI data centers by 2035. Combined, Korean industry has pledged over $90 billion in AI investments.

Samsung separately announced it will invest over $170 trillion won in the next decade, with approximately 42.5 trillion won concentrated in the southwestern Honam region, selecting Gwangju — about 300 km from Seoul — as the location for a new fab, and establishing an AI data center at the southernmost tip of the peninsula in Haenam. Preferential conditions for electricity, water resources, and workforce were key factors in the location selection. Lee Jae-myung noted that the existing chip hubs in Yongin and Pyeongtaek have “reached capacity limits,” necessitating a dispersal strategy to the southwest, and declared 2026 as the critical year for Korea to establish its status as an “irreplaceable” industrial power. This wave of massive investment comes against the backdrop of a severe global memory chip shortage, industry dubbed “RAMageddon,” driven by explosive demand from rapid AI infrastructure expansion. Samsung, SK Hynix, and Micron have all been enjoying record orders as a result.

💬 JudyAI Lab Perspective

Korea’s $90+ billion AI investment commitment is driven by the memory shortage crisis the industry calls “RAMageddon” — this isn’t just industry news; it’s a signal that the next phase of AI infrastructure competition is officially underway.

Samsung and SK Hynix’s choice to build new fabs in the southwest rather than expand existing facilities in Yongin and Pyeongtaek shows that “geographic dispersion” has become the practical solution for hardware supply chains. The original text directly states that the two major chip hubs have “reached capacity limits” — this is a point worth taking seriously for AI builders. The inference APIs we rely on require massive HBM to function; whether HBM supply can keep pace with AI infrastructure expansion directly impacts future compute availability and pricing. Samsung, SK Hynix, and Micron are currently enjoying record orders, but these new fabs won’t come online until the early 2030s — that gap is a structural pressure the entire industry will face together.

If you’re planning long-term costs for AI applications, you can now start tracking the HBM capacity release schedule — it reflects compute pricing trends earlier than any pricing announcements.

📅 Original Source

🔗 Further Reading