[Market Focus] Semiconductor Back-End Companies' Stocks Rise on "Memory Supercycle" Outlook

As DRAM semiconductor inventories have dropped to record lows and prices are rising, the long-awaited "semiconductor supercycle" is gaining momentum for the first time in seven years. As a result, the stock prices of semiconductor back-end processing companies, which count Samsung Electronics and SK Hynix among their clients, are showing strong performance.

As of 9:40 a.m. on September 30, LB Semicon is trading at 5,010 won, up 440 won (9.63%) from the previous trading day.

At the same time, Neosem is trading at 10,300 won, up 590 won (6.08%) from the previous day.

LB Semicon is a key partner providing semiconductor back-end processing services (such as bumping and testing) to Samsung Electronics, and it has recently taken on a significant role in Samsung Electronics' CIS (image sensor) and AP (application processor) testing business.

Neosem specializes in semiconductor back-end inspection equipment and has signed a contract to supply SK Hynix with CXL (Compute Express Link) memory testing equipment. Samsung Electronics and Micron are also among its clients.

According to market research firm TrendForce, as of the end of the third quarter, the average inventory held by global DRAM manufacturers stood at 3.3 weeks, marking an all-time low. This is even lower than the average inventory of 3 to 4 weeks during the 2018 semiconductor supercycle.

As the artificial intelligence (AI) market experiences rapid growth, demand for high-bandwidth memory (HBM)-essential for AI accelerators-has surged. Semiconductor companies have begun converting existing DRAM production lines to HBM production, resulting in a sharp decline in general-purpose DRAM output. In addition, the timing of server replacements at data centers has further boosted demand for standard DRAM.

The semiconductor boom is expected to continue for the foreseeable future. This is because all three major memory companies-Samsung Electronics, SK Hynix, and Micron-are focusing their efforts on HBM4, which will be installed in Nvidia's next-generation AI accelerator, Rubin, inevitably limiting the production of general-purpose memory.