Samsung and SK Hynix Announce 800 Trillion Won Southwestern Chip Cluster: Profound Restructuring of Global Memory Supply Chain

What Happened

On June 29, 2026, South Korean President Lee Jae-myung attended a government event and officially announced the national semiconductor and AI industry strategy. Samsung Electronics and SK Hynix pledged to build two new semiconductor manufacturing plants each in Gwangju, southwestern South Korea, and South Jeolla Province, with a total investment of 800 trillion won (approximately $518 billion). In addition, the government and local authorities jointly allocated an additional 5-20 trillion won for local infrastructure construction, and separately planned 81 trillion won to build an advanced packaging cluster in Chungcheong Province. After the announcement, Samsung and SK Hynix stock prices fell on the same day, as the market questioned the ability to absorb such a large-scale production capacity.

Why It Matters

This investment is the largest single-industry plan in South Korean history, exceeding the combined value of all recent expansion projects by TSMC and Intel. For the semiconductor industry, it means that global DRAM and NAND Flash capacity will increase dramatically over the next 5-10 years, and it will directly impact the supply-demand balance of high-bandwidth memory (HBM). More critically, this strategy aims to turn southwestern South Korea into the second-largest chip manufacturing hub after Gyeonggi Province, fundamentally changing the geographical layout of the country's semiconductor industry.

In-Depth Industry Analysis

Technology Roadmap Impact: Memory Capacity Surge and HBM Landscape Shift

Currently, Samsung and SK Hynix collectively hold over a 90% share of the HBM market. The new fabs will primarily produce DRAM and NAND Flash, a significant portion of which will be used for HBM. According to current plans, when all four fabs are fully operational between 2030-2032, global DRAM capacity could increase by more than 30%, and HBM capacity could even double. This will have a profound impact on the cost structure of AI training hardware:

• Declining AI Training Costs: After the easing of HBM supply tightness, the price per GB of memory in GPU clusters will enter a downward trend, and the total training cost for large models is expected to drop by 15%-25%.
• Technology Roadmap Competition: The capacity expansion by Samsung and SK Hynix will force Micron to accelerate the construction of its domestic plants in the U.S. and may trigger a new round of specification competition between HBM3E and HBM4. At the same time, the new fabs will introduce advanced processes such as EUV and hybrid bonding, further narrowing the gap with logic chip manufacturing.

Supply Chain Impact: Formation of a New Ecosystem and Challenges to the Existing Landscape

#### Upstream Equipment and Materials### Supply chain impact: Formation of a new ecosystem and challenges to the existing landscape

#### Upstream equipment and materials

Each fab's construction cycle typically takes 3-5 years, with the first two years being the period of intensive equipment procurement. Equipment vendors such as ASML, Applied Materials, Lam Research, and KLA will receive a large number of orders. In particular, ASML's EUV lithography machines — the new plant is expected to require 30-50 EUV tools, with a total value close to $20 billion. Suppliers of photoresist, specialty gases, and silicon wafers will also benefit. However, if domestic Korean materials companies (such as Dongjin Semichem and SK Materials) cannot simultaneously expand production, they will face increased risk of import dependence.

#### Midstream manufacturing and packaging/testing

The 81 trillion won packaging cluster in Chungcheong Province directly extends the industrial chain to the back end. This poses direct competition to packaging and testing companies such as ASE and Amkor — Samsung and SK Hynix are building vertically integrated capabilities from wafer manufacturing to advanced packaging. In particular, Samsung's HBM business is gradually adopting in-house panel-level packaging (PLP) technology, which may reduce its reliance on outsourced packaging and testing.

#### Downstream AI infrastructure

Cloud service providers (AWS, Microsoft, Google) and enterprise AI customers will be the biggest indirect beneficiaries. Increased HBM supply will shorten GPU server delivery times and lower the pricing of AI instances in cloud computing. However, in the short term, the spot market must be wary of overcapacity risks — if AI demand growth slows, a large inventory of DRAM could depress prices and harm memory makers' profitability.

Reshaping the competitive landscape: A game among three countries and four players

#### South Korea vs. Taiwan vs. the United States

• South Korea's investment is essentially a struggle with Taiwan's TSMC for dominance in AI chip manufacturing. TSMC leads in advanced logic processes, but South Korea is countering through HBM, a "companion chip." In the future, every AI chip will need to be paired with HBM. If South Korea controls HBM capacity, it can grasp the key node in the AI supply chain.
• The United States is promoting domestic manufacturing through the CHIPS Act, but Korean companies' cost advantages and policy environment remain attractive. Micron, as the only U.S. memory manufacturer, will face dual pressure: a flood of capacity from Korea and labor competition from within the U.S.
• China: South Korea's capacity expansion will further squeeze the market space for Chinese memory manufacturers (such as Changxin Memory Technologies and YMTC). Export controls on advanced equipment have already restricted China's access to EUV, and the EUV investment in South Korea's new plant will widen the generational gap with China. However, China may accelerate self-developed HBM or shift to domestic alternative technologies.

#### The internal race between Samsung and SK Hynix

Competition between these two companies in the HBM field has become intense. The new plants are close in location, but their technology generations may differ — Samsung is more inclined to prioritize the 1c nm DRAM process, while SK Hynix may invest more in TSV (through-silicon via) and hybrid bonding. Over the next five years, the market shares of the two companies will move from the current dynamic equilibrium toward more intense price wars.

Regional industrial chain restructuring: The transformation of southwestern provinces

(Note: The text after this heading is not provided in TEXT_TO_TRANSLATE, only in CONTEXT_AFTER, so we do not translate it.)Gwangju and Jeollanam-do have historically not been semiconductor hubs, lacking skilled FAB engineers and supporting facilities. The Korean government plans to attract talent from the Seoul metropolitan area through tax incentives, housing subsidies, and industry-university cooperation. However, according to reference reports, critics argue that local infrastructure (electricity, water, transportation) remains insufficient to support such a massive project.

If successful, the southwest region will form a new semiconductor industry belt, driving the agglomeration of upstream materials, components, and logistics services, reducing Korea's over-reliance on Gyeonggi Province for semiconductors. However, it should be noted that Korea's long-standing shortage of manufacturing labor persists—semiconductor engineer salaries have risen by over 20% annually for three consecutive years, and new factories may further increase labor costs.

Investment Perspective: Short-Term Bearish and Long-Term Value

Stock prices fell on the day of the announcement, reflecting market concerns over massive capital expenditures. Samsung and SK Hynix's capital expenditures will increase significantly over the next decade, putting pressure on free cash flow. However, from a long-term perspective, controlling HBM production capacity means locking in core resources for the AI era. Referencing the historical semiconductor cycle, large-scale capacity expansion typically goes through 2-3 years of price declines before entering a super cycle driven by explosive demand growth.

Institutional investors need to focus on three key indicators: 1. Construction permits and environmental review progress: Delays will postpone capacity release. 2. HBM average selling price (ASP): If new capacity causes ASP to drop by more than 30%, profitability will be under pressure. 3. Downstream AI capital expenditure: If the growth rate of AI investment by Microsoft, Google, etc., slows down in 2027-2028, it could lead to severe overcapacity.

Long-Term Outlook: Possible Evolution from 2026 to 2036

• 2026-2028: Factories break ground, equipment orders increase, and the market begins to digest expectations of new capacity. HBM prices decline moderately.
• 2029-2031: The first batch of factories starts production, DRAM supply becomes excessive, and prices may fall by 40%-50%. Samsung and SK Hynix's return on capital declines, but their market share further solidifies.
• 2032 and beyond: AI applications become fully pervasive (autonomous driving, robotics, AGI), HBM demand surges again, and new capacity exactly meets demand. The Korean southwest becomes the world's densest memory manufacturing hub.

Risks that cannot be ignored: Geopolitical tensions may lead to a subsidy race under the CHIPS Act, and the U.S. may restrict Korean HBM from flowing back to the U.S. through tariffs or export controls. At the same time, China may accelerate independent HBM R&D, weakening Korea's first-mover advantage.

ConclusionSamsung and SK Hynix's southwestern chip cluster represents the largest single capacity investment in the history of the global semiconductor industry. It is not merely an expansion of production by the companies, but a strategic deployment by the South Korean government attempting to convert its memory dominance into leadership in AI infrastructure. In the short term, the market worries about overcapacity and profit dilution; in the long term, the concentration of HBM production will make South Korea an irreplaceable link in the AI computing supply chain. For industry observers, the key over the next five years is not whether capacity can be built, but whether demand can match this massive supply expansion.