SK Hynix Bets $38 Billion on AI Memory Chip Boom

SK Hynix has pledged 54 trillion Korean won ($38.1 billion) to expand memory chip manufacturing in one of the largest single-company capital commitments in semiconductor history. The investment is split between two flagship facilities: 35.2 trillion won for the Yongin Y2 fabrication plant and 19.1 trillion won for the Cheongju M17 NAND facility.

The announcement signals that AI-driven demand for advanced memory is not a short-term trend but a structural shift reshaping where and how the world's largest chipmakers allocate capital. Analysts at Morgan Stanley and Goldman Sachs have both raised price targets for SK Hynix following the announcement.

HBM: The Chip AI Can't Get Enough Of

High-Bandwidth Memory (HBM) has become the critical constraint in building AI accelerators. Nvidia's H100 and B200 GPUs each require multiple HBM stacks, and data center operators are ordering at a pace that has left the entire industry supply-constrained since early 2025.

SK Hynix is currently the leading supplier of HBM3E, the most advanced production variant, with Micron and Samsung both racing to close the gap. The Yongin Y2 plant — which breaks ground in July 2027 — is specifically designed to produce HBM and next-generation DRAM at volume, with first cleanroom output expected by June 2029.

The scale of the facility is unprecedented: Y2 will span multiple cleanroom bays with total installed capacity that analysts estimate will effectively double SK Hynix's HBM output by 2030.

AI Investment Cycle Shows No Sign of Slowing

The investment comes as hyperscalers — Amazon, Google, Meta, and Microsoft — continue to expand data center capacity at record rates. Each new cluster of AI servers requires thousands of GPU units, each of which relies on HBM stacks that only a handful of companies in the world can produce.

Analysts note that demand is expected to outpace supply through at least end of 2028, keeping memory prices elevated and margins strong across the sector. Samsung and Micron have both announced their own multi-billion dollar expansions in response to the same demand signal.

For SK Hynix, the timing is deliberate. By locking in capacity now, the company aims to maintain its lead in HBM while the AI infrastructure buildout accelerates — a window that CEO Kwak Noh-jung has described as a once-in-a-generation opportunity.

The Cheongju M17 NAND Facility

The second component of the investment — 19.1 trillion won for the Cheongju M17 NAND plant — targets the enterprise and data center storage market. Construction begins in February 2027, with output planned for 2030.

NAND flash demand is also rising, driven by AI inference workloads that require fast local storage at scale. While NAND margins are thinner than HBM, the sheer volume of storage required by large language models during training and deployment makes it a strategically important product line.