South Korea invests $550B to prevent AI 'RAMageddon'

What happened?

On June 29, 2026, Samsung Electronics and SK Hynix, the world's two largest memory chip manufacturers, jointly announced an investment plan exceeding $550 billion to build new semiconductor fabrication plants in South Korea. The goal is to dramatically increase production of high-bandwidth memory (HBM) and DDR5, critical components for AI accelerators used by companies like NVIDIA, AMD, and Intel. The initiative, reported by TechCrunch, aims to avoid a scenario dubbed 'RAMageddon' — a severe memory shortage that could stifle AI growth. This announcement comes after HBM demand already exceeded supply by 15% in 2024, according to TrendForce data, with projections for 2025 indicating a 20% deficit. The joint investment is equivalent to roughly 30% of South Korea's GDP in 2025, underscoring the magnitude of the commitment. Samsung plans to allocate $300 billion to a new complex in Pyeongtaek, while SK Hynix will invest $250 billion in an expansion of its Icheon campus. Both companies expect the new factories to begin mass production by 2029, just as HBM demand is forecast to double from 2026 levels.

Why is this important?

Demand for memory chips for AI has skyrocketed over the past two years. Large language models (LLMs) and AI training systems require enormous amounts of HBM, which offers superior bandwidth for moving data between GPU and memory. According to industry estimates, HBM demand could outstrip supply by 30% by 2027 if manufacturing capacity is not expanded. This massive investment not only secures supply but also solidifies South Korea's dominance in the semiconductor market, which already controls over 70% of global memory share. To put it in perspective, in 2023, the combined R&D and capex investment of Samsung and SK Hynix was approximately $80 billion; this plan quintuples that figure in a single announcement. Additionally, the South Korean government has approved $20 billion in tax incentives and streamlined environmental permits to accelerate construction. The initiative also responds to growing competition from China, which has increased its memory investment by 40% annually since 2024, according to the Semiconductor Industry Association (SIA).

Market and user implications

In the short term, the investment could stabilize memory prices, which have risen up to 40% year-over-year due to shortages. AI companies and data centers will be able to plan their deployments with greater certainty. For example, NVIDIA, which consumes 60% of global HBM production for its H200 and B100 GPUs, has already signed long-term supply agreements with both manufacturers. In the long term, overcapacity could lead to a price war, benefiting consumers but pressuring manufacturer margins. Historically, each memory overinvestment cycle (such as 2018-2019) has resulted in price drops of 30-50%, followed by market consolidation. Moreover, this move reduces dependence on Taiwan in the chip supply chain, diversifying production geographically. Currently, Taiwan produces over 90% of advanced logic chips, but its share in memory is marginal; South Korea already dominates that segment. However, risks include construction delays, water and energy shortages (South Korea faces seasonal water deficits in its industrial zones), and a potential AI demand bubble. If AI adoption slows, overcapacity could be devastating for manufacturers.

What readers should know

This is not just a hardware investment; it is a geopolitical bet. South Korea aims to position itself as the 'chip arsenal' for AI, leveraging tax incentives and deregulation approved by its government. In parallel, the United States has pushed the CHIPS and Science Act, allocating $52 billion to domestic manufacturing, but memory remains a weak point. Companies like Micron Technology (US-based) have also announced expansions, but their capacity is much smaller. Investors should monitor HBM shipments from Samsung and SK Hynix to clients like NVIDIA, which are leading indicators of sector health. According to TechCrunch, analysts expect Samsung's HBM revenue to grow 150% in 2026 compared to 2025, reaching $40 billion. However, the plan's success depends on execution: building semiconductor fabs takes years and requires specialized equipment from companies like ASML, which already has 18-month lead times for its EUV lithography machines. In summary, this is the boldest move in memory industry history, and it will define AI's ability to scale over the next decade.