As the government pushes to build a new semiconductor cluster in the Honam region, including an advanced memory front-end fab, to meet surging demand for memory chips in the age of artificial intelligence (AI), debate continues inside and outside the industry over whether the plan is feasible. In particular, in the case of Samsung Electronics, questions have been raised about the need for additional investment in a Honam cluster because the company already plans to run the Yongin cluster while operating "monster" production plants such as P4 and P5 at its state-of-the-art Pyeongtaek campus.
According to the semiconductor industry on the 1st, there is no disagreement that the spread of AI will lead to long-term growth in demand for high bandwidth memory (HBM) and high-capacity server DRAM. However, views are split between the analysis that the mega clusters being built by Samsung Electronics and SK hynix across southern Gyeonggi alone can meet a significant portion of demand, and the view that a new base is needed to diversify the supply chain over the long term.
◇ The formula for the memory industry rewritten by AI
Behind the government's emphasis on securing new bases is the structural change in the memory industry triggered by AI. The spread of Generative AI services is driving a surge in the amount of memory installed per server, and HBM in particular remains in short supply. Market research firm Omdia projected that the global memory chip market will expand from $200 billion in 2025 to $800 billion in 2030, quadrupling in five years.
The industry sees memory emerging as a key bottleneck in the AI era as HBM supply fails to keep pace with the spread of AI infrastructure. In fact, as global Big Tech corporations expand their AI infrastructure, competition to secure cutting-edge memory is intensifying.
Structural changes in the memory cycle are also becoming visible. Kim Yang-paeng, a senior fellow at the Korea Institute for Industrial Economics & Trade (KIET), said, "As the share of HBM and AI Server Memory expands, the demand base for memory has become far more diverse than in the past," adding, "The memory cycle has not disappeared completely, but the DRAM market now differs in nature from the past, when it centered on PCs and smartphones."
There is also analysis that, as long-term supply contracts expand among Big Tech corporations and demand grows for customized products, efficient operation of existing advanced production lines and technology upgrades are becoming more important than large-scale capacity additions aimed merely at boosting output, as in the past.
◇ Even with exploding AI demand, "Yongin and Pyeongtaek are enough"
However, many in the industry say the outlook for rising demand does not immediately translate into the need to build a new cluster. That is because the scale of investment already underway by Samsung Electronics and SK hynix is far larger than in the past.
At the Yongin national industrial complex, where Samsung Electronics will invest about 360 trillion won, a total of six advanced semiconductor fabs are planned. SK hynix also plans to invest 122 trillion won to build four large fabs at the Yongin semiconductor cluster. Factoring in P4 and P5, the key expansion lines at Samsung Electronics' Pyeongtaek campus, production capacity is expected to grow further. In particular, P5 is planned as a three-story structure with six clean rooms and is evaluated as a super-large line with about 1.5 times the production space of the existing P4.
A Samsung Electronics official said, "Our memory production base expansion has been pursued with Pyeongtaek P4 and P5 as the priority, and the principle of investing in line with market conditions and demand remains unchanged."
An official at a market research firm said, "P4 and P5 are the completion of the Pyeongtaek campus," adding, "If both lines go into full operation, advanced DRAM production capacity could expand to 2.5 to 3 times that of competitors." The official added, "P4 and P5 are expected to be in full operation as early as 2027 or by 2028 at the latest, and by then, even if demand for AI infrastructure continues, memory supply and demand are likely to gradually find balance."
The industry believes that if the 10 fabs in Yongin and the Pyeongtaek expansion plans are all completed, even if memory demand in the AI era grows faster than expected, a significant portion can be met by expanding the existing mega clusters alone. A semiconductor industry official said, "It is true that memory demand in the AI era is surging, but the production capacity planned in Yongin and Pyeongtaek is on a scale incomparable to the past," adding, "It is reasonable to first complete the existing mega clusters as planned and then decide on additional investment after watching the actual pace of demand growth."
The government is also accelerating the creation of clusters in the greater Seoul area. It has moved to shorten approval periods for the Yongin national industrial complex and support the construction of infrastructure such as power and water. The industry sees the early establishment of the Yongin cluster as a key variable that will determine the competitiveness of the domestic memory sector.
◇ The 1,100 trillion-won blueprint, and what it takes to make it real
On the other hand, some argue that, from a long-term perspective, Korea should diversify regional bases and preemptively broaden the national industrial foundation. They say that if production facilities become overly concentrated in the greater Seoul area amid intensifying global competition for semiconductor dominance, risks such as natural disasters or large-scale blackouts could grow.
Park Jae-geun, president of the Society for Semiconductor & Display Technology (endowed chair professor at Hanyang University's School of Integrative Electronics), said, "Because building semiconductor infrastructure typically takes around seven years, it is necessary to prepare industrial complexes and basic infrastructure in advance," adding, "Taiwan also builds infrastructure first and then executes investment in line with actual demand." He added, "While it is necessary to prepare infrastructure in advance, it is desirable to decide on actual fab investments after watching future demand and market conditions."
There are also many hurdles to clear for a Honam front-end fab to become a reality. Even the Yongin cluster now underway is facing difficulties with land compensation and the construction of infrastructure such as power and water, and there are concerns that if the Honam cluster proceeds simultaneously, national infrastructure resources and specialized talent could be spread thin.
Logistics efficiency is also a variable. If, as the government plans, a front-end fab is built in Honam and back-end facilities are set up in Cheonan and Onyang in South Chungcheong, wafer transfers will be unavoidable. The industry is concerned that logistics expenses could rise and supply chain operating efficiency could decline during transfers between production bases.
A semiconductor industry official said, "Although this memory upcycle is likely to last more than five years, considering the front- and back-end processes and the partner ecosystem, it is more efficient to first complete the existing mega clusters as planned."
With an investment blueprint of 1,100 trillion won on the table, the industry says quantitative verification is needed during execution to determine whether a Honam front-end fab can establish itself as a competitively viable production base. Observers say it is necessary to consider not only the outlook for AI demand growth, but also the expansion capacity of existing mega clusters in the greater Seoul area, the supply of specialized talent, and logistics efficiency between the front- and back-end processes.