View of Samsung Electronics’ Giheung campus. Source:Samsung Electronics

Samsung Electronics’ semiconductor business fell short of expectations in the third quarter of this year. Its operating profit was about half that of competitor SK hynix, which leads in high-bandwidth memory (HBM).

On October 31, Samsung Electronics reported that it recorded third-quarter sales of 79.1 trillion won and an operating profit of 9.18 trillion won. The Device Solutions (DS) division, responsible for semiconductors, posted revenue of 29.27 trillion won and an operating profit of 3.86 trillion won. Revenue was up 2.5 percent from the previous quarter, but operating profit dropped by about 40 percent.

Initially, the DS division was expected to post an operating profit in the mid-5 trillion won range for the third quarter. However, after preliminary results were announced on the 8th, market analysts lowered their expectations to the 4 trillion won range. Even this fell short, amounting to just 55 percentof SK hynix’s 7.3 trillion won. This increases the likelihood that Samsung Electronics’ semiconductor division will be overtaken by SK hynix in annual operating profit for the first time.

Samsung attributed the decline in semiconductor performance to one-time expenses and the impact of exchange rates due to a weaker dollar. Although it did not disclose specific details, one-time expenses are estimated at around 1.2 trillion won.

Most of the losses were incurred in the non-memory businesses, such as foundry (semiconductor contract manufacturing) and system LSI (design), which are relatively new business areas. Losses in these divisions are estimated to be in the mid-to-high 1 trillion won range. Samsung Electronics’ foundry division has not posted any notable results due to declining yield rates (percentage of usable products) and difficulties in securing clients.

The memory business performed relatively well, generating a profit of nearly 7 trillion won. “Driven by demand for artificial intelligence (AI) and servers, we saw significant sales growth in HBM, DDR5 (a type of double data rate DRAM), and server SSDs (solid-state drives) compared to the previous quarter,” Samsung stated.

HBMs, which are built by stacking DRAM, have been SK hynix’s stronghold in recent years. Semiconductor design companies like NVIDIA use HBMs to support data processing in chips for AI computation. Samsung Electronics’ lagging performance is largely due to its loss of leadership in the HBM supply competition.

However, Samsung began mass-producing its fifth-generation HBM3E in the third quarter of this year and is reportedly supplying it to AMD and other clients. For sixth-generation HBM4 production, Samsung has also opened up the possibility of collaborating with foundry competitor TSMC.

Deliveries of HBM3E to NVIDIA are also expected by the end of the year. “We have now completed critical steps in quality testing for our major customer (NVIDIA),” said Kim Jae-joon, executive vice president of Samsung’s DS and Memory Business Divisions, during the earnings call. “We expect to expand (HBM3E) sales in the fourth quarter.”

Capital expenditure for the DS segment is projected to reach 47.9 trillion won for the year, a slight decrease from last year’s 48.4 trillion won, due to a decision to scale back foundry investments. “This year’s CAPEX is expected to decrease, and we plan to proceed carefully and efficiently with profitability in mind,” said Song Tae-joong, vice president of the foundry business development team at Samsung Electronics.

Memory investment is expected to be similar to last year. However, the company plans to reduce the share of legacy (general-purpose) memory for mobile and PC applications and focus on high-value-added products such as AI and high-performance computing. “Demand for high-capacity and high-performance products is expected to increase due to data center investments related to AI,” Samsung Electronics stated, “and we plan to expand the share of advanced processing.”

※This article has undergone review by a professional translator after being translated by an AI translation tool.