•Samsung and SK Hynix to build four fabs in southwest Korea
•SK targets 15GW data centre capacity by 2035 in state-coordinated push
The fact
South Korea has launched the "Korea Great Leap 3" programme, a national investment framework worth more than $576bn spanning semiconductors, AI data centres and physical AI. President Lee Jae Myung unveiled the plan in Seoul alongside Samsung Electronics Chairman Lee Jae-yong and SK Group Chairman Chey Tae-won.
Under the plan, Samsung and SK Hynix will each build two semiconductor fabrication plants in southwest Korea — centred on Samsung's confirmed Gwangju site, a former air force base — shifting production capacity away from the Seoul metropolitan area where power and water infrastructure is nearing saturation. Samsung has committed 2,655 trillion won and SK Hynix 2,100 trillion won in domestic long-term investment, with the government pledging support on power, water, land, infrastructure and training of 7,000 AI chip specialists.
The government aims to double DRAM production capacity within five years and has set an AI data centre investment target of 550 trillion won by 2029, rising to more than 1,000 trillion won by 2035, with a total capacity goal of 18.4GW. SK Group alone targets 15GW of AI data centre capacity by 2035.
On the day of the announcement, Samsung Electronics and SK Hynix shares fell 4.86% and 1.68% respectively.
The Assessment
This programme marks a shift toward a state-coordinated model of AI infrastructure development, where semiconductor manufacturing and data centre capacity are planned as a unified national system rather than separate industrial layers. The integration of chip production, memory output and AI compute infrastructure into a single long-term framework led by industrial policy is unusual for a market that has historically relied on chaebol-driven investment cycles.
South Korea's strategic repositioning within the global AI supply chain consolidates its leadership in memory chips — particularly HBM and DRAM — while extending into downstream compute infrastructure. From a BTW perspective, this matters because HBM and DRAM are the physical-layer inputs for global AI compute systems. When the world's dominant memory producers coordinate capacity expansion through state policy, it directly shapes the global AI supply constraint timeline.
Execution risks remain tied to energy supply, water availability, logistics constraints and skilled labour bottlenecks — challenges already flagged by industry experts for greenfield fab construction outside established industrial clusters.
What to Watch
Whether hyperscalers participate in Korea's AI data centre investments, whether power grid approvals keep pace with planned expansion, and whether DRAM capacity growth aligns with global AI demand without creating pricing pressure.