Via tomshardware.com
The South Korean chipmaker is betting big on AI-driven demand for high-bandwidth memory, cementing its dominance over Samsung and Micron
SK Hynix just committed $720 billion to constructing the largest network of memory fabrication plants ever built. The investment, concentrated in South Korea, is the company’s answer to a simple question: who’s going to make all the memory chips that AI needs?
The numbers behind the bet #
On August 7, SK Hynix’s board approved roughly 54 trillion won, approximately $38 billion, as the first major tranche of spending. That money will fund construction of the Yongin Y2 fab for DRAM and high-bandwidth memory (HBM) production, along with the Cheongju M17 NAND fab.
Construction on both facilities is set to begin in 2027, with cleanrooms expected to open between 2028 and 2029.
In July, SK Hynix raised $26.5 billion through a US listing on the Nasdaq, marking the largest American depositary receipt offering ever by a foreign company. That capital injection is earmarked for both new facility construction and continued R&D in semiconductor technology.
SK Hynix’s market capitalization has now surpassed $1 trillion.
Why memory chips matter more than ever #
SK Hynix currently commands 58% of the global HBM market as of the first quarter of 2026. Samsung and Micron each hold 21%.
South Korean President Lee Jae-myung has been pushing for accelerated capacity expansion from both SK Hynix and Samsung as part of a national initiative worth roughly $22 billion. The broader goal is ambitious: double South Korea’s memory production capacity within five years.
Competitive dynamics and market implications #
These new fabs won’t produce chips until 2028 or 2029 at the earliest. That means the current supply-demand imbalance for HBM is likely to persist in the near term.
HBM is significantly more complex to manufacture than standard DRAM, requiring advanced packaging techniques that stack multiple memory dies vertically. That complexity creates a natural moat: even with unlimited capital, a competitor can’t simply build a fab and start producing competitive HBM overnight.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our