President Lee Jae-myung is unveiling one of the largest coordinated industrial investment programmes in history, with Samsung and SK Hynix anchoring a chip and AI push that puts Seoul on a collision course with Taiwan and the United States.
The number itself is staggering. South Korea is expected to announce more than 1,000 trillion won, roughly $651 billion, in semiconductor, AI and robotics investment today, with President Lee Jae-myung presiding over what his office is calling a national "great leap" event. Samsung Group is pledging 1,000 trillion won alone over the next decade, including a possible 300 trillion won push to build chip factories in the country's underdeveloped southwest. SK Group is adding its own targets on top. Jay Y. Lee and Chey Tae-won, the chairmen of both conglomerates, are among the business leaders attending alongside representatives from LG Electronics, HD Hyundai Robotics and Korea Electric Power Corp.
The centrepiece isn't the headline figure. It's Gwangju and South Jeolla province, a region that has historically lagged the Seoul metropolitan area in industrial development. Samsung Electronics and SK Hynix are each expected to build four to five semiconductor fabs in the Gwangju area, transforming it into a chip hub the government says will rival anything in the existing industrial belt. The presidential office says the package will also cover AI data centers, batteries, displays and physical AI including robotics, with the government separately detailing support measures for power, water, land, infrastructure, workforce training and housing.
If you want to understand why this matters beyond South Korea's borders, look at what Samsung and SK Hynix actually make. The two companies together dominate global production of high-bandwidth memory, the chips that sit inside every serious AI accelerator Nvidia ships. SK Hynix is already projected to hold roughly 70% of the HBM4 market for Nvidia's next-generation Rubin platform, according to UBS. Samsung is spending $73 billion on capital expenditure in 2026, a 22% increase over the prior year, pushing hard on 2nm-class processes and vertical integration that combines memory, logic and packaging under one roof. The southwest investment accelerates timelines that were originally planned for the 2040s into the mid-2030s, because AI-driven memory demand is outrunning every prior forecast.
The piece of this that tends to get buried is advanced packaging. TSMC currently dominates CoWoS packaging, the process that lets HBM stacks sit close enough to logic dies to function as a unified AI compute unit. TSMC's CoWoS capacity is projected to reach over 125,000 wafers per month by end of 2026, up roughly 79% from 2025. That's an enormous amount of leverage over anyone who wants to ship AI hardware at scale. SK Hynix is directly challenging that position through a partnership with TSMC on HBM4 development, while simultaneously building its own advanced packaging plant in Cheongju at a cost of 19 trillion won. Samsung is pursuing the same end from a different angle, betting on in-house vertical integration to cut out intermediary packaging steps entirely. The Gwangju investment supercharges both efforts.
This is also a geopolitical calculation, not only an industrial one. South Korea watched the US CHIPS Act mobilize over $50 billion in domestic semiconductor subsidies, and Taiwan's government has spent years treating TSMC as a national strategic asset. Lee's government is making a comparable move, using Samsung and SK Hynix the way Washington uses Intel and the way Taipei uses TSMC. The difference is scale: at $651 billion, this dwarfs the US CHIPS Act commitments and represents the largest coordinated government-industry AI and chip commitment outside the United States and China. Samsung is also expected to separately announce what would be Korea's largest AI data center in Asan, South Chungcheong province, on July 2.
Whether a government-directed model can match the speed of US private-market capital is the real open question. American hyperscalers, Microsoft, Google, Amazon and Meta, committed a combined $320 billion in AI infrastructure spending for 2025 alone, driven by competitive pressure rather than industrial policy. South Korea's approach is top-down: the state sets the direction, the conglomerates execute, and the government backs it with infrastructure support. That model built the Korean semiconductor industry in the 1980s and 1990s. Whether it's fast enough for an AI cycle that moves in quarters rather than decades is something Seoul is betting it can find out.
For the global AI supply chain, the implications are immediate. If Samsung and SK Hynix pull off even half of what's being announced today, the bottlenecks in HBM supply that have constrained AI hardware shipments through 2025 and into 2026 get materially looser in the second half of this decade. Nvidia, AMD and the hyperscalers designing their own silicon all benefit from more memory supply competing for their orders. TSMC faces a credible packaging rival for the first time. And an industrial region of South Korea that has been left behind for decades gets a reason to believe it won't be left behind again.
Also read: China's GLM-5.2 matches a banned US AI on cybersecurity tasks and there's no export order that can stop it • Momenta's Hong Kong IPO prices at HK$295.60 as Chinese autonomous driving bets on software margins over profits • Micron Technology briefly overtook Meta and Tesla in market value after revenue quadrupled on AI memory demand