[Edaily Reporter SOYEON KIM ] "An increase in investment or the number of new factories does not necessarily lead to a proportional increase in supply."

Kim Woo-hyun, Chief Financial Officer (CFO) of SK hynix, made this remark during a recent third-quarter earnings conference call regarding the recent spate of memory capacity expansions. Although semiconductor companies are embarking on large-scale expansions of their production facilities, the actual supply of memory is unlikely to grow as quickly as the scale of investment. This is because the technical complexity of High-Bandwidth Memory (HBM) is increasing, the construction of new factories and procurement of equipment take a considerable amount of time, and companies are gradually adjusting their production capacity to maintain high profitability.

In the past, the memory industry experienced repeated cycles of oversupply and price crashes during boom periods, as semiconductor companies competitively expanded capacity. The memory downturns of 2019 and 2022–2023 are prime examples. Typically, a cycle of booms and busts has emerged, repeating every three to four years. While some have recently raised concerns about oversupply due to large-scale capital investments, many analysts argue that attention should be focused on structural factors limiting supply growth, alongside the expanding demand for artificial intelligence (AI) infrastructure.

① As HBM Performance Increases, Supply Shortages Persist

A key factor is the technical complexity of HBM. As HBM performance increases, it requires more production resources. Next-generation HBM4E and HBM5 are being developed with larger die areas and more metal interconnect layers than previous generations. As the die area increases, the number of chips that can be produced from a single wafer of the same size decreases. More metal interconnect layers mean additional circuit formation processes, which increases the time required for wafer production and the burden on equipment.

The stacking process unique to HBM is another factor that complicates production. HBM involves stacking multiple DRAM chips vertically and connecting them via through-silicon vias (TSVs) and other methods. Unlike standard DRAM, this requires an additional advanced packaging process to stack and connect the chips, and defects occurring during the stacking process can impact the production efficiency of the final product. For the current flagship product, the 12-layer HBM3E, the yield rate is reportedly around 50 percent.

The three global memory giants are internally aware of these structural limitations. Sanjay Mehrotra, CEO of Micron, recently stated, “As the product lineup shifts from HBM3E to HBM4 and HBM4E, the burden of wafer consumption is increasing, which is constraining supply growth.”

② Even After Building Factories, Mass Production Takes a Long Time… Securing Equipment Is Also a Variable

The fact that it takes longer than in the past to move from the construction of a new factory to actual mass production also constrains supply expansion. SamsungElectronics projected that memory supply will remain tight through 2028, even if capital expenditures increase significantly next year. Kim Jae-jun, Executive Vice President of SamsungElectronics, said, “Given the long lead time from new fab construction to wafer production, it will be difficult for next year’s increase in capital expenditure to make a substantial contribution to bit production by 2028.”

Building a semiconductor factory does not mean production can begin immediately. Mass production is only possible after securing, delivering, and installing cutting-edge equipment, followed by trial operations, process stabilization, yield optimization, and quality verification by customers. Vice President Kim noted, “As lead times for equipment orders increase, it is becoming even more difficult to secure equipment in a timely manner.” The production capacity and delivery schedules of equipment suppliers have emerged as key variables determining the pace of capacity expansion for memory companies.

A prime example is the Dutch semiconductor equipment manufacturer ASML. ASML projected that it would ship approximately 65 extreme ultraviolet (EUV) lithography systems—used in cutting-edge semiconductor production—this year. The company is pursuing plans to expand its production capacity to about 85 units next year and to about 110 units by 2028. However, it stated that current production capacity plans will likely be insufficient to meet customer demand in 2027–2028. This means that even if semiconductor companies expand their production facilities, securing key equipment could become a stumbling block to scaling up mass production.

③Build the Factory First, Get the Equipment Later… A Shift in Expansion Strategy

The supply management practices of semiconductor companies are also changing. The strategy involves proactively constructing new factories while adjusting equipment installation and production line operations to align with market demand. The goal is to reduce the risk of oversupply by separating factory construction from the actual expansion of production capacity.

SK hynix plans to accelerate the construction schedule for the “shells”—the structural frameworks of new factories such as the Yongin Semiconductor Cluster and M17—while opening the cleanrooms, the sterile environments necessary for semiconductor production, in phases. CFO Kim Woo-hyun stated, “We will proactively prepare the underlying infrastructure in anticipation of medium- to long-term demand, but we will adjust key resources—such as equipment installation and mass production expansion—that lead to actual increases in supply based on demand.”

SamsungElectronics also plans to secure cleanrooms proactively and then adjust the timing of equipment installation in response to changes in demand. The strategy is to increase supply by considering demand and profitability, rather than activating all production lines at once, even if the factories are built in advance.