SK海力士，DRAM扩产800%

Core Viewpoint - SK Hynix is significantly increasing its production capacity for sixth-generation 10nm DRAM, betting on the AI inference application market, with plans to raise monthly production from approximately 20,000 wafers to between 160,000 and 190,000 wafers, representing an increase of 8 to 9 times, which will account for over one-third of its total DRAM capacity [1][2]. Group 1: Production Capacity Expansion - SK Hynix plans to enhance its advanced memory chip production capacity, focusing on the 1c DRAM technology node, with expectations to add 140,000 wafers per month at its Icheon plant, which is seen as a minimum increase [3]. - The company has achieved a yield rate of over 80% for 1c DRAM, which is primarily used for manufacturing the latest general DRAM products like DDR5, LPDDR, and GDDR7 [3]. Group 2: Market Demand Shift - The strategic shift of SK Hynix is driven by the transition of AI applications from training to inference, leading to a surge in demand for cost-effective general DRAM, which is expected to grow at a rate comparable to HBM [5]. - Major tech companies, including NVIDIA, Google, and OpenAI, are developing custom AI accelerators that integrate a significant amount of general DRAM, indicating a shift in market demand structure [5][6]. Group 3: Financial Outlook - Analysts predict that SK Hynix's facility investment will exceed 30 trillion KRW next year, a substantial increase from the estimated 25 trillion KRW this year, with operating profit expected to surpass 70 trillion KRW, setting a historical record [2][9]. - The company has successfully negotiated a price increase of over 50% for HBM4 to above $500 per chip, which is expected to bolster its profitability [8][9]. Group 4: Competitive Landscape - Samsung is also expanding its 1c DRAM production capacity, aiming to reach 200,000 wafers per month by the end of 2026, reflecting the competitive nature of the next-generation memory market [10][13]. - The ongoing expansion by both SK Hynix and Samsung highlights the strong confidence in the demand driven by AI applications, as the supply of DRAM is currently unable to meet the rapid growth in demand [11][13].