DRAM Bottleneck: Key Growth Engine in the AI Era

Companies at bottleneck points within the semiconductor industry tend to see their value increase due to rising product prices and improved performance.

Companies located at bottleneck points can expect positive effects, including rising product prices and improved performance.

As GPUs, specialized in parallel processing, gained prominence, the AI accelerator market grew rapidly, but early semiconductor designs lacked effective data pathways.

To solve these data traffic issues, High Bandwidth Memory (HBM) emerged, beginning to alleviate data processing bottlenecks.

The vertical stacking structure of HBM acts as a factor that simultaneously increases the demand for general-purpose DRAM, becoming central to AI data processing.

As AI inference and agent AI technologies make the ability to store large-scale data and retrieve it as needed increasingly important, the related memory capacity is growing exponentially.

Samsung Securities projects long-term DRAM demand growth at 22% annually, while the wafer capacity growth rate for the top three memory companies remains at approximately 15% per year.

With 70% of this capacity focused on HBM production, the wafer capacity growth rate for general-purpose DRAM is only around 5%, indicating a deepening structural supply shortage.

During past downturns, companies postponed expansion, leading to a shortage of key infrastructure like cleanrooms, and new fab expansions by Samsung Electronics and SK Hynix are scheduled for after 2027.

HBM uses 3 to 5 times more wafers than general DRAM of the same capacity, meaning that as HBM production increases, wafer supply for general-purpose DRAM production decreases.

The strengthening of legal enforceability for long-term supply contracts and the introduction of advance payment and cancellation penalty clauses have ensured the sustainability of profits.

The strengthening of legal enforceability for long-term supply agreements (LTAs) and the introduction of advance payment and cancellation penalty clauses are ensuring the sustainability of profitability for semiconductor companies.

With the establishment of a three-strong semiconductor system comprising Samsung Electronics, SK Hynix, and Micron, supply and demand control capabilities have improved, enhancing overall industry stability.

To resolve AI inference bottlenecks, server DRAM to store cache data is also essential, suggesting that investment opportunities exist where these bottlenecks occur.

Reflecting these market conditions, Shinhan Asset Management listed the 'SOL Global DRAM Semiconductor Plus ETF' on the domestic market on October 7, adopting a strategy of concentrated investment in key global memory companies.

The remaining 25% of the portfolio is allocated to selected NAND, storage, and memory equipment companies with strong market dominance and competitiveness.

This ETF consists of a total of 10 stocks, diversifying investments across leading companies in the US, Japan, and Korea, while weighting major players to pursue both stability and growth.

Previously, there was no way to directly invest in global DRAM within pension accounts, but with the launch of this ETF, investors can benefit from deferred taxation, reducing the effective tax rate to around 5%.