Key Takeaways:
SK hynix will spend 54 trillion won ($38 billion) on two new fabs in South Korea, betting that AI memory demand is a structural shift rather than a cyclical boom.
SK hynix approved 54 trillion won ($38 billion) for two new memory fabs in South Korea, locking in DRAM and NAND capacity through 2031 as it bets AI demand is a structural shift, not a cyclical boom.
"By proactively securing production capabilities, we aim to establish ourselves as a key partner in AI infrastructure," an SK hynix official said.
The board approved 35.2 trillion won for the Yongin "Y2" DRAM fab and 19.1 trillion won for the Cheongju "M17" NAND fab. Y2 breaks ground in July 2027 with its first cleanroom opening in June 2029 for HBM and next-generation DRAM; M17 starts in February 2027 with a December 2028 cleanroom. Omdia projects DRAM and NAND demand growing at a 19 percent CAGR through 2030.
The outlay executes a master plan to invest 600 trillion won in the Yongin cluster and 100 trillion won in Cheongju, with the cluster's four-fab completion pulled forward to 2033 from 2045. SK hynix is racing Samsung and Micron to secure HBM supply for Nvidia and other AI customers.
Yongin Y2 anchors HBM and next-gen DRAM
Y2 is the second of four fabs planned for the Yongin Semiconductor Cluster, a 1.26 million pyeong (4.16 million m²) site in Gyeonggi province. The fab spans 341,000 pyeong (1.13 million m²) and will produce HBM — the high-bandwidth memory stacked atop Nvidia's AI accelerators — alongside other next-generation DRAM. Phase 1 power and water infrastructure needed through Y2 operation is 99 percent complete, letting construction proceed on schedule.
The investment also funds a business support building, an integrated R&D center consolidating product testing and analysis, and auxiliary infrastructure including substations and water treatment. Y1, the cluster's first fab, is on track to open its first cleanroom in February 2027.
Cheongju M17 extends NAND for enterprise SSDs
SK hynix chose Cheongju for its NAND expansion because the site offers the fastest construction timeline. The campus already houses the M11, M12, and M15 fabs, with established power and water infrastructure that lets M17 link into existing operations. The 206,000 pyeong (680,000 m²) fab targets surging demand for enterprise SSDs and KV cache storage in AI inference, with applications expected to broaden as agentic and physical AI roll out.
Capacity staged to demand, not dumped at once
SK hynix is not releasing all capacity at once. The company plans to build fabs on the master schedule while staging cleanroom expansions and equipment installation to match customer demand, a strategy that maximizes capital efficiency. Investment runs through October 2031 for Y2 and April 2031 for M17.
The scale of the bet is significant for the memory market. SK hynix, the dominant HBM supplier, is expanding as rivals Samsung and Micron also pour capital into AI memory. Nvidia has tapped SK hynix and Samsung for HBM4 in its Vera Rubin platform, excluding Micron. With Omdia forecasting 19 percent annual demand growth through 2030, the three suppliers are racing to lock in capacity before customers sign multi-year deals — memory makers have already sealed 2027 supply agreements with no room for new buyers.
For investors, the question is whether the market has priced in the build-out. SK hynix's infrastructure-first, equipment-on-demand approach means the 54 trillion won commitment does not immediately translate into operating capacity; the first new cleanrooms arrive only in late 2028 and mid-2029. That timing, and whether AI memory demand holds through the decade, will determine whether this capex cycle pays off or leaves the industry with excess supply.
This article is for informational purposes only and does not constitute investment advice.
