South Korea's policy chief Kim Yong-beom defended a 10-year national AI strategy on July 6, saying semiconductor, AI data-center and physical-AI projects should continue beyond one administration. UPI reported that Kim tied the strategy to long-term investment plans from Samsung Electronics and SK hynix, with government support focused on electricity, water and land for memory-chip supply. Yonhap's June 29 coverage adds the scale behind the push: a planned $650 billion AI data-center buildout by 2035, including 8.4 GW by 2029. For practitioners, the signal is capacity planning; Korean power, land, cooling and chip policy may affect memory availability, data-center siting and physical-AI deployment timelines.
This is an infrastructure story before it is a political one. For AI and data teams, South Korea's 10-year framing matters because memory supply, data-center power, water, land and regional industrial policy can determine whether training and inference capacity is available at the moment demand arrives.
What happened
UPI reported that Kim Yong-beom, President Lee Jae Myung's chief of staff for policy, said South Korea's three mega-projects in semiconductors, AI data centers and physical AI should be treated as a 10-year national strategy that continues beyond the current government. Kim linked the plan to long-term investment by Samsung Electronics and SK hynix, and described the government's role as supporting electricity, water and land so memory-chip supply can keep pace with the AI era.
Industry context
Yonhap's June 29 report gives the capacity backdrop: the science ministry said South Korea would push more than a quadrillion won, about $650 billion, into AI data centers by 2035, with an initial 550 trillion won to build 8.4 GW by 2029 and a later expansion through 2035. The Korea Times and Yahoo-linked coverage also show that the chip boom is feeding debate over public investment, tax windfalls and a future response fund, but those are policy discussions rather than settled appropriations.
For practitioners
The practical takeaway is to track South Korea as a compute and memory-capacity market, not only as a chip exporter. If the public-private buildout proceeds, it could influence HBM supply assumptions, non-capital-region data-center siting, power procurement, and physical-AI deployment options for robotics and manufacturing workloads.
What to watch
Watch for binding budgets, permits, utility commitments and company-specific capex disclosures from Samsung, SK hynix, Naver and other participants. The strategy is ambitious, but execution will depend on grid capacity, water access, regional construction timelines and whether future governments keep the same priority.
Key Points #
- 1Kim Yong-beom framed semiconductors, AI data centers and physical AI as a 10-year strategy beyond one administration.
- 2The plan links government infrastructure support to long-term Samsung Electronics and SK hynix memory-supply investment.
- 3Practitioners should track power, water, land and permit execution because those constraints shape real compute capacity.
Scoring Rationale #
The story is notable because it links South Korea's AI, semiconductor and physical-AI infrastructure ambitions to a multi-administration strategy and concrete utility constraints. It remains below major-impact range until binding budgets, permits and company-specific execution milestones are clearer.
Sources #
Public references used for this report.
[01upi.comSouth Korea policy chief defends 10-year AI strategy](https://www.upi.com/Top_News/World-News/2026/07/06/kim-yong-beom-national-strategy/8501783384178/)
02en.yna.co.krScience ministry vows quadrillion-won investment in AI data centers
View 4 more sources #
04Chipmakers face pressure to share excess profits despite investment pledgekoreatimes.co.kr05South Korea to funnel AI chip tax windfall into public investmentfinance.yahoo.com06Kim Yong-beom previews major semiconductor investment announcementkhan.co.kr07Korea President clarifies aide's AI dividend postbloomberg.com
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.