#淡马锡投资SK海力士并推进日本新厂 #Gate广场中秋团圆局 This is a pretty aggressive move—Temasek is investing while SK Hynix expands its production capacity in Japan, with the two sides fitting together perfectly: one is filling its AI memory exposure, while the other is strengthening its overseas manufacturing base.

First, Temasek’s investment
Reports say it plans to invest directly in Samsung and SK Hynix through its internal team, viewing memory chips as “the most undervalued link” in the AI value chain, with the goal of raising the share of AI-related assets to 15% over the next five years[6][8]. However, Temasek later responded that it had invested in both companies more than two years ago and did not confirm plans for a new round. More accurately, the signal is strong, but the claim of a “first major bottom-fishing investment” has not been confirmed.
Now, the new Japanese plant
SK Hynix is evaluating the construction of a memory chip plant in Japan, reportedly with Miyagi Prefecture as a candidate location. The investment could reach tens of trillions of won, equivalent to hundreds of billions of dollars. Officials still say that “no decision has been made.”
Why Japan? The logic is very practical:
AI demand is overwhelming: SK Chairman Chey Tae-won said data center memory chips are in a shortage of roughly 20%–30%, and expanding production in South Korea alone is still insufficient.
Japan has a complete supply chain: materials, equipment, components, and customers are all clustered there, enabling faster communication and coordination due to the shorter distances.
It can also diversify risks: By not putting everything in South Korea, shocks from earthquakes, logistics disruptions, export controls, and other factors will be less damaging. Viewed together, the two moves look more like a grand strategy: Temasek is buying SK Hynix’s future profits, while SK Hynix’s move into Japan is competing for a ticket to AI-era production capacity. If the Japanese plant is built, SK Hynix could not only secure its HBM advantage, but also potentially lift its NAND market share to 36% through cooperation with Kioxia, surpassing Samsung’s 25%.
But the risks are also serious
Memory is highly cyclical, and once prices fall, expanded capacity can quickly become a burden.
US export controls, South Korea-Japan reviews, and local policies could all delay the project.
Japanese plants have high costs, and cultural integration will not be easy either.
Put simply, this is not an ordinary financial investment, but a bet by sovereign capital on the second half of the AI hardware race: whoever controls advanced memory and stable production capacity will have the upper hand.⚠️$SK Hynix