SK Hynix to Spend $38 Billion on Two New Wafer Fabs, Ramping Capacity Based on Demand
0xBroomberg
SK Hynix's board approved $38.3 billion to build two new fabs in Yongin and Cheongju for HBM and NAND expansion — but capacity won't flood the market at once. Equipment goes in as customers need it, meaning real supply impact won't land until late 2028 at the earliest.
What does this $38.3 billion actually build?
SK Hynix will build one new fab each in Yongin and Cheongju, South Korea, investing roughly KRW 54.3 trillion ($38.3 billion) across both sites, with all spending planned for completion by 2031.
The Yongin Y2 fab gets KRW 35.2 trillion and will produce HBM — high-bandwidth memory, the fast chip-to-chip memory AI processors depend on — plus next-gen DRAM. The Cheongju M17 fab gets KRW 19.1 trillion for NAND flash expansion.
This means → over 60% of the capital is aimed at HBM, not traditional flash. The bet on AI-driven memory demand dwarfs everything else.
How aggressive is the Yongin Y2 timeline?
Y2 covers roughly 341,000 pyeong of floor space, with groundbreaking set for July next year and the first cleanroom — the dust-free production hall inside a fab — going live in June 2029.
Supporting infrastructure for the site — power and water — is already 99% complete. The Y1 fab now under construction expects its first cleanroom operational by February next year.
The bigger signal: SK Hynix has pulled the master plan for all four fabs in the Yongin semiconductor cluster forward from 2045 to 2033 — more than a decade ahead of schedule. This reflects a judgment that AI memory demand is arriving faster and harder than originally expected.
Why build the Cheongju M17 fab next to the old campus?
M17 breaks ground in February next year, with the first cleanroom opening in December 2028 — a faster ramp than Yongin Y2.
Cheongju is SK Hynix's core NAND flash base, home to the M11, M12, and M15 fabs. The new plant plugs straight into existing land, power, and water infrastructure.
In plain terms = building next door to an established campus saves money, saves time, and shares logistics — inherently more efficient than starting from scratch.
Record investment — so why not release all the capacity at once?
SK Hynix is sticking to its "infrastructure first, equipment on demand" playbook: build shells and utilities on schedule, but install production tools only as customer orders and market conditions warrant.
This means → even with buildings ready by 2028–2029, actual wafer output ramps only after tools go in. The real supply impact from this expansion will emerge gradually over several years, not overnight.
The company put it bluntly: in the AI era, the ability to deliver capacity precisely when a customer needs it is itself a core competitive advantage. The race isn't who builds the most — it's who times the ramp best.
How big is the long-term blueprint?
Cumulative planned investment in the Yongin semiconductor cluster: roughly KRW 600 trillion. Additional commitment to the Cheongju base: roughly KRW 100 trillion.
The earliest cleanrooms come online between late 2028 and 2029, meaning at least three-plus years separate today's board approval from actual product shipments.
This reflects a bet not on a short-term cycle but on a conviction: AI-driven memory demand growth is a decade-scale structural trend, one worth matching with an equally long capital-expenditure horizon.
Content is for reference only, not financial advice.