China’s Shift to Advanced Chips Tightens Supply, Powering Samsung’s Q3 Earnings Surge

Samsung Electronics Delivers Q3 Earnings Surprise, Driven by Its DS Division
Memory chip prices rise as Chinese suppliers cut output
China shifts focus to high-value products from AI chips and HBM to CXL

Samsung Electronics achieved an earnings surprise in the third quarter of this year. As Chinese memory chip makers shifted to a high-value strategy focused on artificial intelligence (AI), the supply of DRAM and NAND flash declined, pushing up overall product prices and significantly boosting the performance of Samsung’s semiconductor division.

Samsung Electronics Posts Strong Q3 Results

On October 14, Samsung Electronics announced that its consolidated operating profit for the third quarter reached $8.8 billion, up 31.8% from a year earlier. The figure represents a 158.6% surge from the previous quarter ($3.4 billion) and marks the company’s highest quarterly profit since the second quarter of 2022 ($10.2 billion). The last time Samsung reported an operating profit above the $7.5 billion level was in the second quarter of last year.

Revenue came in at $62.6 billion, up 8.7% year-on-year and 15.3% quarter-on-quarter — the highest in the company’s history. It is also the first time Samsung’s quarterly sales exceeded the $58 billion range. Although detailed figures by business segment were not disclosed, industry sources estimate that the Device Solutions (DS) division, which handles semiconductors, generated about $3.6 billion in operating profit — roughly a tenfold increase from the previous quarter.

The sharp improvement in DS performance was largely driven by higher memory prices amid tighter supply of DRAM and NAND products. As of last month, spot prices for 16Gb DDR5 chips rose 27.5% from the previous month, while 16Gb DDR4 chips surged 38.2%. Despite some suppliers delaying the end-of-life (EOL) process for DDR4 products, supply conditions remained tight, sustaining the upward trend in spot prices.

Inventory levels among major DRAM suppliers also fell noticeably. By the end of the third quarter, DRAM inventories stood at about six weeks for Samsung, two weeks for SK hynix, and two weeks for Micron — decreases of two, one, and two weeks respectively from early Q3. NAND inventories also declined to roughly ten weeks for Samsung, twelve weeks for SK hynix, and eight weeks for Micron.

China Shifts Focus from Standard Memory to Advanced Chips

A key factor behind the recent memory supply shortage lies in the changing production strategy of China’s semiconductor industry, which has long supported global supply of general-purpose memory products. In May, reports emerged that ChangXin Memory Technologies (CXMT), China’s largest memory chipmaker, would gradually halt production of DDR4 chips for servers and PCs by mid-2026. The company was said to have formalized the decision through an end-of-life notice and planned to convert more than 60% of its total output to DDR5 by the end of this year. The move is widely viewed as part of Beijing’s broader effort to achieve semiconductor self-sufficiency amid U.S. export restrictions.

The trend initiated by CXMT is now spreading across China’s memory sector. Investment outlet Ainvest recently reported that Chinese chipmakers are scaling back output of standard DRAM to focus resources on developing high-end AI chips — signaling that China is beginning to actively follow global semiconductor trends.

Leading Chinese tech companies are also accelerating their push into advanced chip technologies. On September 18, Huawei unveiled its next-generation roadmap for the Ascend AI chip lineup at the annual “Huawei Connect” event in Shanghai. The company plans to release the Ascend 950PR and 950DT in the first and fourth quarters of next year, followed by the Ascend 960 in late 2027 and the Ascend 970 in late 2028. These models will feature two types of in-house high-bandwidth memory (HBM) — HiBL 1.0 and HiZQ 2.0.

Around the same time, Alibaba Group’s semiconductor design subsidiary, T-Head, also drew attention. According to a report by the South China Morning Post citing Chinese state broadcaster CCTV, T-Head’s parallel processing unit (PPU) achieved performance levels comparable to NVIDIA’s H20 AI chip designed for the Chinese market. The Alibaba PPU uses HBM2E, a previous-generation high-bandwidth memory technology.

China Enters the CXL Market

CXMT is now expanding beyond HBM into developing DRAM based on the next-generation interface standard, Compute Express Link (CXL). CXL is designed to maximize data processing efficiency by linking CPUs, GPUs, and DRAM within a unified architecture.

Industry experts expect the CXL market to grow rapidly starting next year with the introduction of the CXL 3.0 standard. The new version supports a “fabric” function that can dramatically reduce data center total cost of ownership (TCO). Under the traditional server structure, increasing DRAM capacity required adding CPUs, which created unnecessary costs. By contrast, CXL DRAM modules using the fabric function can connect up to 4,096 CXL devices to a single CPU, depending on workload requirements.

Analysts believe CXMT could find it relatively easier to enter the CXL market than to compete directly in HBM. A CXL DRAM module consists of a DRAM chip and a controller — the system semiconductor that manages communication, data transfer, and resource allocation. Since DRAM itself is a general-purpose product with relatively small performance differences among suppliers, the controller’s capabilities are considered the key source of competitiveness.

Rather than developing its own controllers, CXMT is expected to source them from fabless companies specializing in the field and combine them with its DRAM. One industry source noted, “Given that CXMT’s process technology remains at around the 16-nanometer level, it would be difficult for the company to design controllers in-house. It’s more likely that CXMT will partner with firms like Montage Technology to procure controllers and enter the market that way.”