SK Hynix announced Friday it will invest 54 trillion Korean won, roughly $38.1 billion, to build two new memory chip manufacturing plants in South Korea, marking one of the largest single semiconductor investments in history as the company rushes to keep pace with explosive AI demand.
The board approved a $24.9 billion DRAM and high-bandwidth memory fab called Yongin Y2, the second of four chip factories planned for the Yongin Semiconductor Cluster about 20 miles south of Seoul. Groundbreaking is set for July 2027, with the first cleanroom targeted for production by June 2029. The facility will span 279 acres of floor space and focus on HBM, the high-speed memory that powers Nvidia’s AI accelerators.
A second fab in Cheongju will receive $13.2 billion and manufacture NAND flash storage chips. SK Hynix said existing power and water infrastructure at the Cheongju campus will accelerate construction, with the first cleanroom expected to open in December 2028. The company recently introduced HBF, a new NAND technology that stacks flash circuits in a manner similar to HBM.
The investment is part of a broader 430 trillion won long-term plan to expand SK Hynix’s manufacturing capacity. The company is the world’s second-largest memory chip maker after Samsung and the primary supplier of HBM to Nvidia, whose AI GPUs depend on the technology for storing and accessing massive model data at high speed.
Industry analysts warned that memory prices are unlikely to ease before the end of 2028, even with the new capacity coming online. “With demand growing even faster than planned capacity, memory prices are unlikely to soften before the end of 2028,” Counterpoint Research analyst Neil Shah told CNBC. PC builders and consumers may wait even longer for relief, as manufacturers allocate the majority of production to data center customers chasing higher margins.
The announcement comes amid a broader memory shortage that has pushed lead times beyond 2028 for some products. SK Hynix is simultaneously ramping HBM4 production to supply Nvidia’s next-generation Vera Rubin platform, which is expected to begin commercial shipments in the third quarter of 2026. Samsung and Micron are also racing to qualify their own HBM4 offerings.
SK Hynix’s bet underscores a central tension in the semiconductor industry: whether the AI infrastructure buildout will sustain demand at current levels long enough to justify hundreds of billions in new fab construction. Memory makers are betting yes, even as some analysts warn of a potential bubble if AI spending plateaus.
CNBC – SK Hynix to invest $38 billion building new memory chip plants SiliconANGLE – SK hynix approves $38B+ investment in two new memory fabs
discussion