South Korea MSIT National AI Computing Center — 2025 Work Plan + KRW 1.46tn GPU Procurement

Mechanism

MSIT's 2025 Work Plan designated the National AI Computing Center as a priority infrastructure initiative, committing the government to: (1) establish the center via PPP structure with major domestic cloud providers; (2) revise data-center siting and permitting regulations to accelerate construction timelines; and (3) procure sovereign compute capacity on a multi-year budget trajectory from KRW 2 trillion toward KRW 4 trillion through 2030.

The July 2025 first tranche (KRW 1.46tn / ~USD 1.1bn) was allocated to three operators:

Total first-tranche GPU count: ~13,136 units. Government-owned subpool earmarked for universities and AI research institutes. Private operator share supports commercial AI workloads with preferential access for domestic AI firms.

Structural distinctiveness from related KR instruments

This action occupies a distinct layer in Korea's AI policy stack:

• AI Basic Act (2025-01-21): Horizontal AI governance framework — risk tiering,

high-impact AI obligations, trustworthiness standards. Regulatory layer only; no compute procurement.

• K-Chips Act (2023-03-31) / Semiconductor Special Act (2026-01-29): Fiscal and

regulatory incentives for semiconductor fab investment. Targets production of chips, not procurement of compute capacity.

• National Power Grid Expansion Special Act (2025-03-25): Addresses the electricity-

infrastructure bottleneck — grid siting and permitting for AI-datacenter + fab load. Upstream enabler; this action is the demand-side complement.

• This action: Sovereign compute procurement — directly acquires GPU capacity for

domestic AI model training, research, and public-sector AI workloads. Korea becomes a buyer of NVIDIA supply at scale, reducing dependence on commercial cloud API access for nationally sensitive AI development.

Downstream implications

• NVIDIA (NVDA): Direct demand signal. B200 and H200 allocation confirms Korea as

a Tier-2 sovereign buyer alongside Japan (NEDO compute programme), UAE (G42), Saudi Arabia (Humain/KACST), and France (ANDIL). KRW 1.46tn ≈ USD 1.1bn in one tranche.

• Samsung Electronics HBM: NVIDIA B200 and H200 use HBM3E; Korean government

procurement indirectly anchors Samsung (and SK Hynix) HBM demand domestically.

• Korean cloud oligopoly: Naver/NHN/Kakao become quasi-public compute utilities for

AI. Risk: market distortion if government-backed compute undercuts private AI-cloud pricing.

• Geopolitical read-through: Korea joins the 2025 sovereign-AI-compute race alongside

Japan, Singapore, France, Saudi Arabia, and UAE. Reinforces NVIDIA's leverage in bilateral tech diplomacy — Korea's HBM export-control sensitivity (Samsung/SK Hynix to Huawei) is now partially offset by its own NVIDIA-dependent infrastructure stack.

• KRW 4tn 2030 trajectory: If delivered, Korea would operate ~50,000 government-

accessible GPUs by 2030 — competitive with Japan's NEDO programme and ahead of most European national compute programmes.

Open questions

• Whether the MSIT data-center siting regulation revision (committed in work plan) will

materially accelerate permitting in tandem with the Power Grid Expansion Special Act.

• AMD / domestic GPU (SAPEON, Rebellions) penetration in future tranches — current

tranche is 100% NVIDIA.

• Whether Kakao's tranche (no explicit government-owned subpool) creates a

preferential-access governance gap relative to NHN/Naver.

• Integration with the K-Cloud national cloud sovereignty initiative and whether the

compute center will be designated a national critical-infrastructure node under the AI Basic Act's forthcoming Presidential Decree.