At a glance: On June 29, 2026, South Korean President Lee Jae Myung announced a combined 1,350 trillion won — approximately $880 billion — commitment from Samsung Electronics and SK Hynix, the world’s two largest memory chipmakers, toward new semiconductor fabs and AI data center capacity through 2035. Bloomberg confirmed the figure the same day; CNBC and Al Jazeera reported it as a government-coordinated private capital commitment, not direct state spending. Including AI data centers from SK Group, GS Group, and Naver, the total AI infrastructure commitment across all companies exceeds $1 trillion.

The AI infrastructure arms race that began with data center announcements in the United States has reached South Korea — and the scale of the response is notable. Samsung and SK Hynix held roughly 65% of the global DRAM market combined in Q2 2026 (39% and 26% respectively, per Counterpoint Research), and the two together account for the large majority of HBM production that feeds AI accelerators.

What’s Being Built

CryptoBriefing and Al Jazeera reporting provides a breakdown of the $880B Samsung-SK Hynix commitment — the fab and data-center pieces below sum to the $880B headline figure; the packaging-cluster investment further down is additional spending on top of it, not a third slice of the same total:

Four new semiconductor fabrication plants — $518 billion (800 trillion won)

Samsung will build two fabs; SK Hynix will build two. All four will be located in southwest South Korea, deliberately dispersing development beyond Seoul’s existing chip corridors into the Gwangju and South Jeolla province region. This location choice is partly logistical — targeting underutilized power resources — and partly political (see below).

AI data center capacity — $362 billion

The data center buildout targets 8.4 gigawatts of AI compute capacity, with a completion target of 2029. This is additive to existing South Korean data center infrastructure — a purpose-built AI-specific capacity layer.

Additional chip-packaging cluster — $52.5 billion (on top of the $880B above)

Advanced packaging — the technology that integrates memory and logic chips into the compact modules inside AI accelerators — gets its own dedicated cluster investment in the Chungcheong region. Reporting puts this packaging-cluster commitment at roughly 81 trillion won, which converts to the $52.5B figure — this is additional to, not part of, the $880B fab-and-data-center total, reflecting the importance of High Bandwidth Memory packaging to the AI chip supply chain.

When combined with separate AI data center pledges from SK Group, GS Group, and Naver, Al Jazeera reports South Korea’s total AI infrastructure commitment across all companies tops $1 trillion through 2035.

The Timeline Acceleration

The most operationally significant aspect of the announcement is speed. CryptoBriefing reported that the new fabs’ buildout timeline has been pulled in by approximately a decade from previous plans — projects originally slated for completion in the mid-to-late 2040s are now targeting the mid-2030s, a schedule shortened by as much as 12 years for some fabs.

The cause: AI-driven demand for advanced memory solutions is growing faster than capacity projections made even three years ago. HBM (high-bandwidth memory) chips are the specific product driving the urgency. SK Hynix is currently Nvidia’s primary HBM supplier, holding roughly half or more of global HBM revenue share through 2025-2026 — though Samsung and Micron also supply HBM qualified for Nvidia’s H100, H200, and Blackwell accelerators, so the supply base is not exclusive to SK Hynix. Samsung is competing aggressively for a larger share of that market with its HBM4 generation.

The fab investment is not speculation about future AI demand — it is a response to demand that AI infrastructure buyers are already placing.

The Government’s Role

This is predominantly private capital being coordinated, not direct government spending. CNBC described the government’s role as that of orchestrator rather than funder: creating regulatory frameworks, providing infrastructure support, and setting regional development incentives that make the southwest construction sites viable.

The Information framed it as a state-directed strategy that avoids the model of direct subsidy used by the US CHIPS Act — South Korea is betting that market incentives, combined with Samsung and SK Hynix’s existing market dominance, are sufficient to attract the capital without the roughly $52 billion in direct manufacturing and R&D incentives ($39B for manufacturing incentives plus $11B for R&D, with a further $75B in loan/loan-guarantee authority) that the CHIPS and Science Act provided to the US Commerce Department.

Regional government contributions — reported at roughly 5 to 20 trillion won ($3.2 to $13 billion) depending on province — are part of the package, but represent a small fraction of the total commitment.

President Lee’s Framing

President Lee Jae Myung announced the commitment personally, framing it around South Korea’s competitive position in the AI era. His stated priority: “We must secure the core elements of AI faster than any other country.” He characterized the initiative as part of what he called the “triple axis” of semiconductors, physical AI, and data centers — the infrastructure layer he believes determines which country leads the next phase of the technology economy.

Al Jazeera noted the political controversy: opposition parties criticized the southwestern fab location, alleging the site selection favors a traditional stronghold of President Lee’s Democratic Party over commercial viability. The administration rejected this framing, citing power availability and regional economic development rationale.

The Geopolitical Context

The South Korea announcement sits alongside competing national semiconductor commitments that have reshaped the global chip manufacturing landscape since 2022:

  • United States: CHIPS and Science Act ($39 billion in direct manufacturing incentives + $11 billion for R&D, plus separate authorization for up to $75 billion in loans/loan guarantees); TSMC, Samsung, Intel, and Micron all announced or broke ground on US fabs
  • Japan: Government subsidies attracting TSMC’s Kumamoto fab (opened 2024, with second-fab construction following); a third Kumamoto facility has been reported as under consideration by TSMC, but had not been formally announced as of this writing
  • Taiwan: TSMC’s domestic capacity expansion alongside its global diversification
  • European Union: European Chips Act targeting 20% of global chip production by 2030 (up from roughly 10%)

CryptoBriefing reported that South Korea’s investment is explicitly positioned as a national response to this competitive landscape — particularly to the US CHIPS Act, which has attracted foreign investment into American manufacturing while potentially drawing capacity away from South Korean sites.

The HBM focus is the strategic differentiator. Logic chips (GPUs, CPUs) are contested by TSMC, Samsung, Intel, and GlobalFoundries across multiple manufacturing nodes. HBM is different: it requires memory-specific manufacturing expertise, and South Korea’s two incumbents hold commanding market share. The $880B commitment doubles down on that structural advantage rather than attempting to compete head-on in leading-edge logic with TSMC.

What This Means for AI Infrastructure

HBM availability has become a genuine bottleneck in AI infrastructure buildout. Nvidia’s Blackwell GPU production has been capacity-constrained by HBM supply — every H100, B100, and GB200 requires HBM3e or HBM4 stacks that only SK Hynix, Samsung, and Micron currently manufacture at scale. SK Hynix remains the clear leader by revenue share, but Micron has grown rapidly, reaching roughly 21% of the global HBM market by Q4 2025 (up from 9% a year earlier) and rivaling Samsung for second place — so Micron is no longer reliably the smallest of the three.

The four new fabs and accelerated timeline are a supply-side response to this bottleneck. If the buildout proceeds on the announced schedule — mid-2030s completion — it roughly aligns with the timeline when current projections suggest AI training cluster scale may require substantially more memory bandwidth per accelerator than today’s HBM3e generation provides.

For AI infrastructure buyers (hyperscalers, AI labs, enterprises building on-premise AI capacity), the South Korea announcement signals that the memory chipmakers are making capital bets that AI-specific memory demand remains elevated through the decade — not a temporary spike.

Sources