JPMorgan Decodes South Korea's "Three Mega Projects": Memory Capacity to Double in Five Years as Samsung and SK Commit a Combined 47.55 Trillion Won
N.R. Finch
South Korea, Samsung, and SK Hynix launched a KRW 4,755 trillion semiconductor mega-investment plan targeting a doubling of memory capacity within five years and pulling Yongin fab mass-production forward by up to 12 years — but the capacity takes time to land, and the near-term memory shortage remains unresolved.
Where does KRW 4,755 trillion go?
Samsung Group commits KRW 2,655 trillion, with KRW 2,100 trillion for semiconductors. SK Group commits KRW 2,100 trillion, with KRW 1,000 trillion for AI infrastructure.
This means → both giants are betting on the same direction: memory + AI compute, which together account for the vast majority of total spending.
The government branded the strategy "3S+1F" — Speed, Stronghold, Spearhead, Full Support. In plain terms = Seoul is treating semiconductors as a national-security priority, not just industrial policy.
How is Samsung allocating its budget?
The largest slice — KRW 1,650 trillion — goes to the Yongin fab cluster and existing sites. Next is a potential new manufacturing hub in Gwangju at KRW 400 trillion.
The Cheonan/Onyang HBM back-end packaging line — where high-bandwidth memory chips are stacked and packaged — gets KRW 56 trillion. Next-gen display in Asan gets KRW 67 trillion.
This reflects Samsung's priority ranking: advanced-node wafer capacity > packaging > new ventures, with core resources concentrated on fab manufacturing.
What does SK's AI-infrastructure bet mean?
SK is repositioning data centers from "storage" to "compute generation," targeting 15 GW of compute in two phases: Phase 1 at 5 GW, Phase 2 adding 10 GW by 2035.
This means → SK is no longer just selling memory chips — it wants to build AI factories itself, shifting from selling components to selling compute.
In plain terms = SK used to ship chips to others who built data centers. Now it is building the data centers too, aiming for a bigger share of the value chain.
How much earlier will the Yongin fab come online?
The original timeline was 2045–2047. The new target is 2033–2040 — a pull-forward of 7 to 12 years.
Four new fabs will be built in the southwestern region at a combined KRW 800 trillion. The Chungcheong HBM packaging plant adds KRW 81 trillion. R&D and talent spending totals KRW 30 trillion over 15 years.
This means → Korea is trying to get the capacity built while AI demand is still surging, rather than breaking ground after the peak has passed.
Why did the market sell the news?
On the announcement day, Samsung Electronics fell nearly 5% and SK Hynix dropped 1.7%.
Apple's MacBook and Microsoft's Xbox have already raised prices due to memory shortages. Apple is reportedly planning to source memory chips from China to cope with tight supply stretching into next year.
In plain terms = this capacity will land gradually over five years. It does nothing for the shortage right now — and the market is pricing the near-term pain, not the long-term blueprint.
What is the key test for this mega-investment?
The Korean government projects the global memory market will quadruple within five years — if that holds, even a doubling of capacity may not be enough.
Whether the Yongin fab can hit mass production around 2033 on the new schedule is the single most important milestone for validating the entire plan.
This reflects a deeper question: the return on mega-investment depends on whether the AI demand curve stays steep, not just on how many fabs get built.
Content is for reference only, not financial advice.