CFM has updated the domain of its english website from en.chinaflashmarket.com to www.memorymarket.com, Please be informed.

Operating Profit Quadruples, Investment Scale Significantly Increases; SK Hynix Expects This Price-Rise Cycle to Be Longer Than in the Past

By: CFM 5 days ago

SK Hynix has released its financial report for the first quarter of fiscal year 2026, ending March 31, 2026: Revenue was 52.5763 trillion KRW up 96% quarter-on-quarter and 405% year-on-year; net profit was 40.3459 trillion KRW, up 165% quarter-on-quarter and 398% year-on-year. Operating margin for the quarter was 72%, with a net margin of 77%.

Driven by expanding AI infrastructure investment, market demand remained exceptionally strong. SK Hynix's revenue exceeded the 50 trillion KRW mark for the first time, with operating profit reaching 37.6 trillion KRW and an operating margin of 72%, setting a new all-time high since the company's founding.

Regarding the recent weakness in memory spot prices, SK Hynix noted that this is a temporary situation caused by inventory releases in some channels following a sharp price surge, and is not a signal of a cycle peak. Due to persistent supply shortages, this price-rise cycle could be longer than in the past.

Business Segment Details:

NAND accounted for 21% of revenue, approximately 11 trillion KRW. NAND bit shipments decreased 10% quarter-on-quarter, while ASP grew about 75% quarter-on-quarter. In the NAND flash business, SK Hynix plans to expand supply of its world-highest-stack 321-layer TLC and QLC products, as well as eSSDs, to strengthen competitiveness in the AI data center and AI PC memory markets.

DRAM accounted for 78% of revenue, approximately 41 trillion KRW. DRAM bit shipments were flat quarter-on-quarter, while ASP grew about 65% quarter-on-quarter. Regarding HBM, SK Hynix will further strengthen its comprehensive execution capabilities integrating performance, yield, quality, and supply stability. HBM4 mass production ramp-up is planned according to customer-agreed schedules. In DRAM, SK Hynix will fully launch the industry's first 6th-generation 10nm-class (1c) process-based LPDDR6 DRAM, as well as 192GB SOCAMM2 based on the same process, which began mass production this month.

Inventory: SK Hynix's inventory value for the quarter was 15.974 trillion KRW, up 12% quarter-on-quarter.

Regarding the impact of geopolitical conflicts on production capacity, SK Hynix stated that the company has diversified its raw material suppliers and secured sufficient inventory, so the actual impact is very limited.

Capital Expenditure

SK Hynix emphasized that in an environment where customer demand continues to exceed supply capacity, the ability to supply structural demand growth driven by the AI era has become a core competitive advantage. The company will focus on capacity ramp-up at Cheongju M15X and the Yongin cluster, advancing infrastructure construction and the introduction of core equipment such as EUV. This year's investment scale will increase significantly year-on-year.

Market Applications

Servers: AI is moving from training to inference and the era of Agentic AI. To efficiently process and store the continuously generated data from AI agents, and to meet growing memory requirements in overall server systems, demand for various memory products (including both DRAM and NAND) continues to rise.

PC/Smartphones: Under cost pressure from rising memory prices, market demand has shown signs of softening, reflected in batch order adjustments and product mix changes. However, strong server memory demand continues to offset this weakness and drive overall market growth.

Operational Updates by Product

HBM: For HBM4, SK Hynix is collaborating closely with customers in the early product development stage to ensure performance meets customer requirements, with production ramp-up planned according to agreed schedules. For HBM4E, the core chip will utilize 1c-nano process technology, with yield and mass production capabilities having reached a mature stage. The company expects to provide HBM4E samples in the second half of this year, targeting mass production in 2027.

DRAM: Development of the industry's first 1cnm LPDDR6 has been completed, with full-scale mass production and supply planned for the second half of the year, initially targeting the next-generation flagship models of a major smartphone customer. Additionally, 1cnm 192GB SOCAMM2 began mass production this month and is optimized for NVIDIA's Vera Rubin platform.

NAND: In client SSDs (cSSD), SK Hynix has begun supplying PQC21 products based on CTF process 321-layer QLC. In enterprise SSDs (eSSD), the company is flexibly addressing overall AI demand by building a full product lineup covering high-performance TLC and high-capacity QLC. To address the explosive growth of intermediate data throughput driven by AI model development, the company will transition over half of its domestic NAND flash production capacity to 321-layer technology by the end of this year.

Market Outlook

SK Hynix expects that in the second quarter, DRAM bit shipments will increase 7%-9% quarter-on-quarter, and NAND bit shipments (including Solidigm) will increase 16%-17% quarter-on-quarter.

SK Hynix stated that the current strong pricing is driven by structural market changes, not just temporary supply-demand mismatches, making this cycle different from past cycles. The diffusion of memory-efficient technologies will improve the economics of AI services, thereby driving overall service scale expansion, which will further boost memory demand. In an environment of persistent supply-demand imbalances, customers prioritize "securing supply" over "lower prices," and SK Hynix expects a favorable pricing environment in both the DRAM and NAND flash markets.

SK Hynix clarified that concerns about memory-efficient technologies (such as Google's TurboQuant) suppressing demand are unnecessary. These technologies do not reduce memory usage; rather, they improve the efficiency of processing long contexts and parallel inference with the same amount of memory. Ultimately, the proliferation of AI services will instead drive higher memory demand.