According to a research report released by Goldman Sachs on October 16, memory chip manufacturer ChangXin Memory Technologies (CXMT) has made progress in a new-generation DRAM memory cell architecture, sparking concerns across the global memory chip industry about the future competitive landscape.

Since the relevant news emerged, shares of major memory chip companies including Samsung Electronics, SK hynix (SKHY.US), and Micron Technology (MU.US) have fallen sharply, with the combined market capitalization of six major memory manufacturers evaporating by approximately $738 billion.

However, Goldman Sachs believes that although CXMT is rapidly expanding production capacity, the market may have overestimated its ability to catch up with South Korean leaders in advanced DRAM and high-bandwidth memory (HBM) in the short term.

On October 10, reports disclosed that CXMT is advancing a new DRAM memory cell architecture called 4F². The company's president also stated plans to launch new server memory products based on the 4F² architecture by the end of 2026. This news triggered a sharp correction across the global memory chip sector.

According to Goldman Sachs statistics, between October 1 and 15, Samsung Electronics shares fell 20%, SK hynix dropped 25%, Micron Technology declined 11%, SanDisk (SNDK.US) fell 11%, Kioxia dropped 26%, and CXMT itself also declined 15%. The combined market capitalization of these six companies shrank by approximately $738 billion.

Goldman Sachs noted that CXMT's technological progress was not the only reason for the memory stock decline. During the same period, the market was also affected by AI infrastructure financing risks, credit market concerns, and U.S. inflation data exceeding expectations. Notably, NAND flash manufacturers Kioxia and SanDisk, which have little connection to 4F² technology, also suffered significant sell-offs, while DRAM spot prices continued to rise during this period. This suggests that the memory stock correction may reflect more of a shift in market sentiment rather than a clear deterioration in industry fundamentals.

4F² Architecture Expected to Increase Memory Density, but Mass Production Still Faces Challenges

4F² is one of the important development directions for next-generation DRAM technology. Traditional DRAM typically uses a 6F² memory cell layout, while the 4F² architecture optimizes the memory cell structure to reduce the footprint of a single memory cell to 2F×2F, enabling more memory cells to be integrated within the same chip area and increasing memory density.

Theoretically, all else being equal, the 4F² memory cell area is approximately one-third smaller than 6F², helping to increase storage capacity per wafer. Goldman Sachs believes that the 4F² architecture holds particular significance for memory manufacturers affected by advanced semiconductor equipment export restrictions. Due to restrictions on obtaining advanced extreme ultraviolet (EUV) lithography equipment from ASML (ASML.US), manufacturers find it difficult to rely entirely on shrinking process feature sizes to increase DRAM density, and the 4F² architecture provides an alternative technological path.

CXMT plans to combine vertical surrounding gate transistors (VSG) and wafer-to-wafer (W2W) hybrid bonding technology to address memory cell leakage issues and allow memory cells and control circuits to be manufactured separately.

However, this technical approach also introduces new manufacturing challenges, including consistency of vertical transistors, charge retention capability of storage capacitors, interference between dense wiring, and the high-precision alignment required for hybrid bonding. Goldman Sachs warned that these technical challenges could drive up manufacturing costs and reduce yields, thereby offsetting some of the benefits from increased memory density.

The firm believes that based on currently disclosed information, CXMT's 4F² product appears to still be in the design phase, with the goal of entering the sample stage by the end of 2026. Even if progress can be made as planned, subsequent customer certification and high-yield mass production will still face greater challenges.

Samsung and SK hynix Also Pursuing 4F², Advanced Process Gap Remains Significant

It is worth noting that CXMT is not the only manufacturer advancing 4F² technology. According to local media reports cited by Goldman Sachs, Samsung Electronics has successfully manufactured functional sub-10nm DRAM chips using the 4F² architecture and vertical channel transistor (VCT) technology, and plans to complete development of 10a DRAM based on the 4F² architecture by the end of 2026, conduct customer certification in 2027, and achieve mass production in 2028.

SK hynix has also incorporated the 4F² architecture into its next-generation DRAM technology roadmap and verified the read/write functionality of related technology through a test array in July 2026.

Goldman Sachs believes that the key to determining the commercial value of 4F² technology is not whether it can be achieved in the laboratory, but whether competitive yields, costs, and economic returns can be achieved in large-scale production. From the current level of technology, South Korean memory manufacturers still maintain their lead.

Goldman Sachs pointed out that Samsung Electronics and SK hynix currently concentrate their DRAM production capacity mainly on the 1b process node and are gradually transitioning to the more advanced 1c node. In comparison, CXMT currently focuses primarily on 1z node products and is expected to continue relying mainly on this process through 2027. As advanced DRAM processes increasingly depend on EUV lithography technology, the lack of such equipment may limit CXMT's pace of advancing to advanced nodes such as 1b and 1c and bring additional cost pressures.

CXMT's 2028 Capacity Expected to Grow 77%, but Still Insufficient to Meet Domestic Demand

Despite the gap in advanced technology, Goldman Sachs acknowledges that CXMT's progress in capacity expansion may exceed previous market expectations. The firm estimates that by the end of 2026, the monthly DRAM wafer capacity of Samsung Electronics, SK hynix, and CXMT will reach 770,000, 660,000, and 290,000 wafers respectively. By the end of 2028, CXMT's monthly capacity is expected to further grow to 513,000 wafers, an increase of 77% from the end of 2026.

During the same period, Samsung Electronics's monthly capacity is expected to grow 21% to 930,000 wafers, and SK hynix's to grow 17% to 770,000 wafers. This means CXMT's capacity growth rate will be significantly faster than that of the two South Korean leaders, and the capacity gap between them is expected to gradually narrow.

However, Goldman Sachs emphasized that even under the above aggressive capacity expansion assumptions, by 2028, CXMT is still expected to meet only about 50% of domestic DRAM demand and 40% of HBM demand. Therefore, a considerable portion of China's memory chip demand will still need to rely on other suppliers, and Samsung Electronics and SK hynix are still expected to obtain important business opportunities in the Chinese market in the future.

HBM Technology Barriers Are Higher, Making It More Difficult for CXMT to Catch Up

Driven by rapidly growing AI demand, HBM has become one of the core areas of competition in the global memory chip industry. Goldman Sachs believes that compared with traditional DRAM, CXMT will face greater challenges in narrowing the gap with South Korean leaders in the HBM field.

HBM competitiveness depends not only on the DRAM chip itself but also on advanced packaging technology and the manufacturing capability of the base logic die at the bottom of the memory stack. As HBM technology continues to upgrade, base logic dies are gradually adopting more advanced foundry processes. For example, the base logic die of Samsung Electronics's HBM4 product already uses a 4nm process.

Goldman Sachs pointed out that if CXMT hopes to develop competitive next-generation HBM products, domestic foundries also need to have the capability to mass-produce advanced-process base logic dies. However, because advanced semiconductor manufacturing equipment remains subject to export restrictions, the supply capability of related base logic dies may become an important bottleneck for CXMT in catching up with global HBM leaders.

Goldman Sachs Bullish on Samsung Electronics and SK hynix, Believes Market Pessimism Is Excessive

Regarding the recent sharp correction in global memory stocks, Goldman Sachs believes that the market may have correctly recognized the competitive pressure brought by accelerated capacity expansion among memory manufacturers, but at the same time overestimated their ability to achieve advanced technological breakthroughs in the short term.

The firm noted that Samsung Electronics's current stock price corresponds to expected price-to-book ratios of approximately 1.2 times and 0.9 times for 2027 and 2028 respectively, while SK hynix's are approximately 1.4 times and 0.9 times respectively. In Goldman Sachs's view, the two companies' expected 2028 price-to-book ratios have fallen below 1 time, meaning the market has effectively priced in a pessimistic scenario in which the memory industry may enter a downturn in 2028. But this differs significantly from Goldman Sachs's judgment that the memory industry's prosperity is expected to remain at a high level for a relatively long period.

The firm believes that from a fundamental and valuation perspective, the current risk-reward ratio for Samsung Electronics and SK hynix is attractive. Goldman Sachs maintains a "Buy" rating on both companies, with a 12-month target price of 500,000 Korean won for Samsung Electronics common stock and 3.5 million Korean won for SK hynix.

However, the firm also cautioned that although the relevant stock valuations have become attractive, catalysts to drive a rapid short-term rebound in share prices may still be limited.