Analysis
South Korea's largest technology companies have committed more than $550 billion to expand memory-chip production, in a coordinated push to relieve the worldwide shortage that the industry has nicknamed 'RAMageddon,' according to TechCrunch. The headline allocations include roughly $518 billion for four new memory fabrication plants in the country's southwestern Honam region, a $52 billion high-bandwidth-memory (HBM) packaging hub in the central region, and a further $356 billion earmarked for AI data centers through 2035 by SK, GS, Naver and others.
RAMageddon is the supply-side mirror of a story that has been building for months. As the world's three dominant memory makers -- Samsung, SK Hynix and Micron -- reallocated wafer capacity toward HBM, the premium stacked DRAM that sits beside Nvidia's AI accelerators, conventional DDR5 and LPDDR memory went scarce and prices surged. The shortage has already pushed device makers, including Apple, to raise prices and even, by some reports, to seek unconventional supply from blacklisted Chinese suppliers. Korea's answer is to build its way out.
The scale is staggering even by semiconductor standards. Samsung outlined a roughly $1.7 trillion (2,655 trillion won) decade-long plan, and SK Group a roughly $1.4 trillion (2,100 trillion won) medium-to-long-term roadmap. President Lee Jae-myung framed 2026 as the year South Korea must cement itself as an 'irreplaceable' industrial power, casting the government's role as enabling companies to invest 'without losses' -- industrial policy explicitly aimed at owning the AI memory layer.
“Samsung outlined a roughly $1.7 trillion (2,655 trillion won) decade-long plan, and SK Group a roughly $1.4 trillion (2,100 trillion won) medium-to-long-term roadmap.”
The competitive and geopolitical implications are significant. Memory is one of the few segments where a handful of players control nearly all global supply, and concentrating yet more capacity in Korea deepens a strategic chokepoint at a moment when the US, China, Japan and the EU are all pouring subsidies into domestic chipmaking. It also raises the stakes for Micron and for China's CXMT, which is racing to scale conventional DRAM as Western buyers hunt for alternative sources.
The catch is timing. Fabs take years to design, build and qualify -- existing facilities in Yongin and Pyeongtaek are said to have already hit their limits -- so the new capacity will not meaningfully ease prices until 2027 and beyond. In the meantime, the memory squeeze remains a live tax on everyone building hardware, from hyperscalers to robotics and consumer-device startups, and a windfall for the incumbents enjoying near-doubled DRAM pricing.
The bear case is the memory industry's brutal cyclicality: massive coordinated capacity additions have historically arrived just in time to turn shortage into glut, crushing prices and margins. A half-trillion-dollar build-out predicated on AI demand staying vertical is a bet that the boom doesn't cool. What to watch: groundbreaking and qualification timelines for the new fabs, where DRAM and HBM contract prices settle into 2027, and whether other governments respond with memory-specific subsidies of their own.
