South Korea has spent three decades building the world’s most concentrated memory chip industry in one metropolitan corridor around Seoul. On Monday, the government announced that it has run out of room there — and that the AI boom is demanding a second one, built faster and further away than anything the country has attempted before.
At a public briefing at the Blue House presidential office in Seoul on June 29, President Lee Jae Myung announced three AI mega-projects with Samsung Electronics Executive Chairman Lee Jae-yong and SK Group Chairman Chey Tae-won. The centrepiece is a semiconductor investment programme worth 800 trillion won ($518 billion), under which Samsung and SK Hynix will each build two new memory chip fabrication plants in South Korea’s southwestern region — an area selected for its available land, power supply, water access, and alignment with President Lee’s regional development priorities. The announcement is the largest single government-backed semiconductor commitment in South Korean history, and one of the largest national chip investments ever publicly announced anywhere in the world.
The semiconductor announcement is one of three mega-projects. The second, led by SK Group, GS Group, and Naver, is a 550 trillion won AI data centre programme targeting 8.4 gigawatts of initial capacity and 18.4 gigawatts by 2035. The third is a physical AI initiative to develop robots, autonomous systems, and humanoid machines for manufacturing, logistics, defence, and healthcare. Across all three programmes, SK Group Chairman Chey said the group will invest a total of 1,100 trillion won — confirming the scale of commitment that makes this not just a semiconductor story but a comprehensive national AI infrastructure pivot.
Key Developments
- Samsung Electronics and SK Hynix announced 800 trillion won ($518 billion) for semiconductor investment on June 29, with each company building two new memory chip fabs in South Korea’s southwest — four new facilities total.
- South Korea’s Industry Minister said the plan is to double DRAM output within five years; global memory market tracker Omdia projects the sector will quadruple to $800 billion by 2030, driven by HBM, AI servers, and data-intensive computing.
- SK Group alone commits 1,100 trillion won across all three mega-projects, including 400 trillion won in the southwest semiconductor cluster and 550 trillion won for AI data centres (18.4GW capacity by 2035).
- Samsung and SK Hynix shares both fell on the announcement — Samsung down ~5%, SK Hynix initially down 6% before recovering. Investors had expected a larger Samsung-alone figure after the Maeil Business Newspaper reported a 1,000 trillion won Samsung-only programme.
What Happened
According to the Washington Post’s wire coverage of the Blue House event, President Lee said South Korea “must secure the core elements of AI faster than any other country,” identifying semiconductors, physical AI, and AI data centres as the “triple axis” for Korea’s economic leap forward. Samsung Executive Chairman Lee Jae-yong named Gwangju — a major city in the southwest — as the leading candidate for Samsung’s next semiconductor investment site, citing its advantages in electricity, water supply, workforce availability, and infrastructure incentives. Gwangju and the South Jeolla provincial government will also contribute between 5 trillion and 20 trillion won toward the project, President Lee confirmed, and an additional 81 trillion won is planned for a chip packaging cluster in the Chungcheong area near Seoul.
As CNBC’s detailed breakdown confirmed, the southwest fab programme is designed to address a constraint that industry observers have been tracking for years: South Korea’s Yongin and Pyeongtaek semiconductor clusters, where most of the country’s current DRAM and HBM production is concentrated, “have already reached their limits,” in President Lee’s words. SK Group Chairman Chey acknowledged publicly that SK Hynix’s Yongin cluster took nine years to prepare from announcement to groundbreaking, underscoring why the government and the chipmakers are moving now rather than waiting for existing sites to overflow further. “Demand visible as of today remains solid. Even if investment continues, it will be difficult to fully resolve the supply shortage,” Chey said.
The Mechanism: Why Memory Can’t Scale Fast Enough
The structural pressure behind Monday’s announcement is not difficult to trace. High-bandwidth memory, the chip variant that Nvidia’s AI accelerators require and that SK Hynix currently leads in supplying, has seen demand outstrip supply for more than 18 months. Micron Technology, the other major HBM supplier, disclosed in its most recent earnings that its entire 2026 HBM supply is fully contracted to AI chip customers before the year began. The result is the same supply constraint that is simultaneously pushing up the price of consumer devices across the industry and limiting how fast the AI infrastructure buildout itself can proceed — the memory shortage is not just a consumer problem but a bottleneck on the AI capability expansion that is driving the investment cycle in the first place.
Market tracker Omdia projects the global memory market will quadruple from $200 billion in 2025 to $800 billion by 2030, driven by HBM, AI server deployments, and data-intensive computing. Building the fab capacity to serve that market, however, requires commitments made five to nine years before the output arrives, given the planning, permitting, construction, and technology qualification timelines involved. The same memory shortage that is straining Apple’s and Microsoft’s supply chains — sending MacBook and Xbox prices up 18 to 33 percent in a single week — is what makes the long planning horizons of fab investment both commercially compelling and urgently necessary.
The Backstory: Building the Second Chip Belt
Monday’s announcement extends a pattern of South Korean semiconductor investment commitments that has been building since the AI infrastructure boom accelerated in 2024. Earlier in June, Nvidia signed sweeping AI deals with SK Hynix, SK Telecom, Naver, and Doosan Group, confirming that NVIDIA purchases “billions and billions of dollars” from SK Hynix annually and that volume will grow substantially. That customer relationship is the commercial context for SK Hynix’s southwest expansion: the company’s Yongin cluster broke ground six years after announcement, and the southwest commitment is an attempt to accelerate that timeline by engaging government support, pre-permitted sites, and regional incentives from the start.
Samsung’s position in this announcement is more complicated. The company has been lagging SK Hynix in advanced HBM technology: SK Hynix is the leading supplier of HBM3E to Nvidia, while Samsung has been investing heavily to narrow the gap. Several of the global chip customers that had been concentrated on TSMC and SK Hynix are now exploring Samsung as a second source for advanced chip manufacturing, adding commercial urgency to Samsung’s investment commitment. The announcement comes after the Maeil Business Newspaper reported on Friday that Samsung Group would announce a 1,000 trillion won investment programme spanning the next decade, causing Samsung’s stock to initially rise before falling approximately 5 percent on Monday as the actual announcement landed with a different structure than markets had anticipated.
Reactions
President Lee framed the mega-projects explicitly in terms of national competitive urgency rather than regional development alone: “We must secure overwhelming production capacity in advance through large-scale new investments, including in the southwestern region.” The three mega-projects — semiconductors, physical AI, and data centres — are designed to be mutually reinforcing: the data centre investments provide domestic demand for the memory chips the new fabs will produce, while the physical AI ecosystem creates a customer base for the inference and training infrastructure the data centres will house. Industry Minister Kim Jung-kwan called semiconductors “Korea’s last chance to escape low growth,” a framing that positions the entire programme as a structural economic intervention rather than simply a capital allocation decision by two large companies.
The Dispute: Ambition Meets Infrastructure Reality
The practical constraints on the southwest expansion are well-documented and have not been resolved by the announcement. SK Hynix’s own Yongin experience demonstrates the pattern: nine years elapsed between announcement and groundbreaking, primarily due to water supply, land compensation, environmental permitting, and community impact negotiations. The southwest sites face the same categories of constraint in a region without the existing supplier ecosystem that Yongin and Pyeongtaek have built up over decades. Attracting and retaining the semiconductor engineers required to staff four new leading-edge fabs in a region that is not currently a chip industry centre is an additional challenge that the investment commitment does not automatically solve.
There is also a technology risk dimension. The southwest fabs are intended to produce advanced memory including next-generation HBM and AI-optimised DRAM variants. The process technology required to produce those chips in 2029 or 2030, when the first new fabs might begin output, does not yet exist in its final commercial form. If Samsung’s HBM technology gap with SK Hynix does not close meaningfully before the southwest fabs come online, the commercial rationale for Samsung’s half of the investment becomes harder to defend in investor returns terms. The broader AI infrastructure investment cycle that justifies these commitments is itself being questioned by the BIS and other financial analysts who warn that the current capex pace is outrunning near-term revenue generation capacity across the AI sector.
What Happens Next
President Lee pledged to appoint a dedicated Blue House official to oversee the mega-projects and to personally lead their execution — an unusual degree of executive-level ownership for an industrial investment programme. The first concrete milestones to watch are site selection confirmations for each company’s two southwest fabs, permitting timelines, and whether the government delivers the regulatory streamlining, infrastructure pre-investment, and worker accommodation support that makes southwestern South Korea competitive with Yongin and Pyeongtaek for engineering talent. The parallel OpenAI IPO situation — where a company with enormous private valuation faces uncertainty about whether public markets will support AI-era investment levels — is the financial risk context in which these commitments will be tested over the next two to three years.
Why It Matters
South Korea’s 800 trillion won semiconductor commitment is the largest single national chip investment announcement in the AI era, and it is a direct response to a global memory shortage that is now visibly affecting consumers, constraining AI infrastructure expansion, and concentrating geopolitical attention on HBM supply chains. The decision to build a second production belt in the southwest rather than simply expanding existing sites in the Yongin-Pyeongtaek corridor reflects both a land and power constraint and a political choice to distribute the economic benefits of the semiconductor boom more broadly across the country. Whether the investment produces its intended output — doubled DRAM capacity, global HBM leadership, and a new industrial anchor for the southwestern economy — will depend on execution quality, technology timing, and whether the AI infrastructure demand that justifies the investment continues at the pace that today’s commitments assume it will.
Sources
Washington Post (AP wire); CNBC; Korea Herald; Korea Times; Yahoo Finance (Reuters); Electronics For You; Benzinga.
