SK Hynix is expanding long-term supply agreements (LTA) with global big tech companies to secure a foothold in the long-term demand for AI memory semiconductors (hereinafter "memory"). The strategy aims to enhance predictability in production and profitability by securing stable demand visibility beyond short-term performance. With the full-scale shipment of HBM4 (6th-generation high-bandwidth memory) and next-generation DRAM expected in the second half, record-breaking performance is anticipated to continue.
On the 29th, SK Hynix announced that its revenue and operating profit for the second quarter totaled 79.3187 trillion won and 60.5426 trillion won, respectively. The quarterly operating profit marked a record high for five consecutive quarters. Factors such as the sale of investment assets in Kioxia (formerly Toshiba Memory), valuation gains, and favorable exchange rate effects contributed to net profit reaching 93.9226 trillion won.
While revenue and operating profit did not meet the market's elevated expectations (consensus estimate), they remained within a reasonable range of recent industry forecasts. Park Jun-deok, head of DRAM marketing, stated, "The timing for expanding shipments of certain high-value-added DRAM products has been pushed to the second half, and portfolio composition affected overall ASP (average selling price). However, we are confident that performance improvements will strengthen further in the second half."
SK Hynix expects full-scale shipments of HBM4 sales, 10-nanometer 6th-generation (1c) DRAM, and SOCAMM2 in the second half. It is anticipated that the high-value-added 321-layer NAND flash (hereinafter "NAND") will account for half of domestic production capacity by year-end. The company also indicated that negotiations for HBM supply next year are progressing smoothly.
Expanding LTA is a core strategy to prepare for long-term growth in the AI memory market. Song Hyun-jong, chairman of SK Hynix, stated during the earnings announcement, "We have concluded LTA (long-term supply agreement) negotiations with over 10 customers, including key clients so far, and are continuing discussions with major industry customers."
The contracting parties reportedly include major global big tech companies such as NVIDIA and Microsoft. Recently, SK Group signed a letter of intent (LOI) with NVIDIA to build AI infrastructure worth more than $500 billion. Park explained, "We have secured a stable profitability foundation centered on HBM, and long-term cooperation with key AI customers including NVIDIA supports this."
SK Hynix's LTA, typically based on a five-year term, goes beyond merely extending the contract period. While contract terms vary depending on the customer and product, prices are not fixed but designed to reflect market fluctuations. This means that if memory prices rise in the future, they can be incorporated into the contract price.
Along with long-term purchase commitments, conditions such as deposits were included to support contract fulfillment. This strengthens customers' actual purchase obligations and enhances visibility of medium-to-long-term demand. Park noted, "This simultaneously improves downside stability for performance while allowing us to leverage additional demand and growth opportunities when the market is favorable."
The market views LTA expansion as one factor limiting the extent of ASP increases. This is because the sharp recent rise in DRAM prices is not immediately reflected in long-term contract volumes. However, analysts assess that the positive effect of securing long-term demand and profitability outweighs the short-term sacrifice of some price gains.
SK Hynix plans to expand production capacity based on the demand visibility secured through LTA. It aims to develop Icheon, Gyeonggi, and Yongin as key production hubs for next-generation DRAM and AI memory, while focusing on strengthening NAND and advanced packaging capabilities at its Cheongju facility in North Chungcheong Province. Equipment investment this year is expected to reach the late 40 trillion won range.
SK Hynix forecasts that memory demand will continue to rise across the entire AI ecosystem. As AI services evolve into agent forms that perform complex tasks on behalf of users, demand is expanding not only for HBM but also for server-grade DRAM and enterprise SSDs.
Song (President) stated, "DRAM and NAND demand this year are expected to increase by the mid-20% and late-10% ranges, respectively, under constrained supply conditions. If supply constraints ease and latent demand is met, market growth could expand further."
SK Hynix also anticipates that major cloud service providers' (CSPs) AI infrastructure investments will remain robust in the medium to long term. This is because AI service competitiveness is directly tied to the core businesses of big tech companies, including search, advertising, cloud, and software. The emergence of high-efficiency AI models is also expected to lower usage costs and entry barriers for AI services, thereby increasing overall usage volume and memory demand.
Additionally, shareholder return initiatives are being reviewed from multiple angles. SK Hynix stated that due to regulatory and procedural constraints related to the ADR (American Depositary Receipt) offering, it is difficult to disclose new important information such as specific shareholder return plans at this time.
Song (President) added, "With significantly enhanced cash generation capabilities, we will be able to meaningfully expand shareholder returns while achieving investment goals for future growth and maintaining financial soundness. We will communicate with the market once specific plans are finalized."
in INDUSTRY
