# SK hynix Plans Four Fabs in Yongin Amid AI Memory Shortages

By John Kim

Canonical URL: https://www.tokenpost.com/news/business/32277
Published: 2026-10-15T04:37:34.000Z
Updated: 2026-10-15T04:37:34.000Z
Section: Business

> Kim estimates SK hynix’s first Yongin fab could eventually process about 350,000 wafers a month. Samsung Electronics is also expanding in Pyeongtaek, while Micron is investing in facilities in Idaho and New York.

Samsung Electronics and SK hynix plan 10 fabs in Yongin, South Korea, as shortages lead some PC and smartphone makers to adopt Chinese supplier CXMT and raise questions about whether future production can keep pace with AI demand.

SK hynix plans four facilities in the Yongin semiconductor cluster. Samsung has six fabs planned at a separate Yongin site, including logic-chip production, and is also expanding in Pyeongtaek; Micron is investing in new facilities in Idaho and New York.

Kim Hyung-joon, head of the Next-generation Intelligent Semiconductor Foundation, estimates the memory supply gap at 30% to 40%. He contrasts that with past supply-demand imbalances of about 2%, which could trigger sharp price swings. Kim estimates monthly DRAM capacity at about 720,000 wafers for Samsung, 550,000 for SK hynix and 300,000 to 400,000 for Micron.

Kim estimates the first SK hynix fab in Yongin could eventually process about 350,000 wafers a month, more than 60% of the company’s current capacity. He also estimates CXMT holds about 10% of the global DRAM market. CXMT says roughly 60% of its output is for export.

Severe shortages have led PC and smartphone makers to start using CXMT memory. Kim views that adoption as a competitive risk for established suppliers. More capacity could give Samsung, SK hynix and Micron room to allocate additional production to conventional DRAM and pursue buyers that adopted [CXMT memory](<https://www.tokenpost.com/news/people/30930>) during the shortage.

The effect of new capacity would unfold over years. Construction, equipment installation and production qualification take time, and manufacturers can open cleanrooms and production lines in stages as yields improve. Kim says companies can adjust the timing of each phase as market conditions change.

Demand is another variable. AI developers are working to make models use less memory, but Kim expects data-center growth to offset or exceed those savings. He sees electricity supply as a potentially larger constraint on AI expansion than chip availability. If demand does not absorb added production, Kim warns, the market could face oversupply and lower prices.



## Links in this article

- [CXMT memory](https://www.tokenpost.com/news/people/30930)
