With demand for artificial intelligence (AI) expanding, prices of memory semiconductor products are rising sharply. If high bandwidth memory (HBM), a core component of AI servers, drove performance improvement at domestic semiconductor corporations this year, forecasts say profitability of general-purpose DRAM such as DDR5, LPDDR, and GDDR will stand out next year. As Samsung Electronics and SK hynix diverge in their investment strategies amid a memory boom, SK hynix's HBM-centric strategy is being tested as to whether it is the best choice for mid- to long-term profitability.

According to market research firm DRAMeXchange on the 16th, the price of DDR5 16Gb has recently climbed to the mid-$10 range, and server DDR5 module prices have risen in tandem. The industry says the price per gigabyte (GB) of general-purpose DRAM has risen to around $12–$13, significantly narrowing the gap with the estimated unit price of HBM4 (around $15 per GB).

The physical HBM4 from SK hynix is on display at the 27th Semiconductor Exhibition (SEDEX 2025) held at COEX in Gangnam-gu, Seoul./Courtesy of News1

General-purpose DRAM is also gaining presence in terms of profitability. Global investment bank UBS said in a recent report that, given HBM's high manufacturing cost, the gross margin of general-purpose DRAM could temporarily exceed that of HBM in 2025–2026. As capacity expansion concentrates on HBM, supply of traditional DRAM is tightening, strengthening the pricing power of the DDR family.

Korea's securities sector shares a similar view. KB Securities research center head Kim Dong-Won said, "With rising prices of general-purpose DRAM, DDR5 margins next year are likely to overtake HBM3E (5th-generation HBM), reversing profitability," adding, "With server demand solid, a deepening shortage of general-purpose memory used in smartphones, laptops, and cars means next year's semiconductor corporations' results will be driven by general-purpose DRAM." Some analyses also suggest the operating margin of certain general-purpose DRAM products could rise to around 70%.

This shift in the profitability structure is being reflected in earnings forecasts. KB Securities put Samsung Electronics' 2026 operating profit at about 82 trillion won and SK hynix at 74 trillion won. Some securities firms, including Kiwoom Securities, cite upper-end estimates of around 100 trillion won for Samsung Electronics and around 90 trillion won for SK hynix. The securities industry says Samsung Electronics, with a relatively higher share of general-purpose DRAM revenue, could enjoy a greater operating leverage effect in a rising price phase.

The two companies' strategies are clearly different. SK hynix is maintaining a strategy that centers new capacity investments on HBM. According to foreign media and industry data, SK hynix internally reviewed the possibility that a shortage in general-purpose DRAM supply could continue through 2028, but the company has set a policy to prioritize newly introduced Low-NA EUV (extreme ultraviolet) equipment for HBM and advanced packaging processes. Observers say the choice emphasizes securing technological leadership in AI memory and the lock-in effect with strategic customers.

However, despite its high price, HBM faces structural profitability limits, critics say. That is because applying foundry (contract manufacturing) processes to the base die and undergoing complex stacking and packaging entails high cost burdens and significant production yield risks. By contrast, general-purpose DRAM has stable processes, so price increases are directly reflected in profit, the securities sector commonly explains.

Samsung Electronics is investing in HBM4 (6th-generation HBM) while also pursuing a "mix strategy" that flexibly allocates capacity to general-purpose and mobile DRAM such as GDDR7 and LPDDR5X. The market says, "HBM is a product of significant strategic importance, but the short- to mid-term results of memory corporations ultimately depend on the product mix between general-purpose DRAM and HBM."

An industry official said, "HBM is essential for securing technological leadership in the AI era, but short- and mid-term results are still heavily influenced by prices of general-purpose DRAM and NAND," adding, "The key to evaluating SK hynix's HBM-centric strategy will be how much it translates into actual revenue."

※ This article has been translated by AI. Share your feedback here.