SK hynix ramps new chip plant for AI memory demand

SK hynix will start ramping up mass production of advanced DRAM at its M15X. Most output allocated to high-margin products, including HBM4, and some used for high-performance server memory.

The company plans to raise monthly wafer input from about 10,000 now, with the line expected to handle 30,000 to 50,000 by year-end and as much as 80,000 by next year.

M15X is an expansion of the existing M15 site that cost about 20 trillion won (US$13.6 billion) and uses extreme ultraviolet tools for 1b DRAM production.

Once fully ramped, M15X is expected to lift SK hynix’s DRAM wafer capacity by 10% to 15% and help bridge supply before its Yongin fab starts operating.

🔗 Source: Chosun Daily

🧠 Food for thought

Implications, context, and why it matters.

The M15X ramp-up fits into a multi-billion dollar plan

SK hynix is ramping M15X with a record 11.95 trillion won (US$8 billion) order for about 24 to around 30 extreme ultraviolet lithography systems from ASML, a Dutch chip equipment maker, to support M15X and the planned Yongin semiconductor cluster 1.
Analysts view the purchase as a bid to secure hard-to-get tools while ASML works through a long backlog 2.
In Cheongju, which also hosts M15X, the company is putting 19 trillion won (US$13 billion) into an advanced packaging facility for the complex assembly of High Bandwidth Memory (HBM) chips 3.
SK hynix pulled forward the opening of the first fab, or chip factory, at Yongin by three months to February 2027, which leaves M15X as the bridge for supply growth in the near term 4.

This memory supercycle may not follow past cycles

In earlier upswings, a build-out like M15X raised investor worries about a later price crash in the cyclical DRAM market, since past downturns brought steep price and revenue drops 5.
AI demand may behave differently because it stays elastic, so lower memory prices can unlock more AI use cases and create fresh demand instead of topping out like PCs or phones 5.
Shifting capacity toward HBM tightens supply for conventional memory, and SK Group chairman Chey Tae-won said the squeeze could last four to five more years through 2030 6.
Hyperscalers, large cloud computing companies, are moving from one-year contracts to multi-year supply agreements to protect supply chains during the crunch 4.