SK Hynix just said the memory shortage will last until 2030. It also said DRAM and high-bandwidth memory are effectively sold out through 2029. It has signed long-term agreements with 14 major customers, including large cloud providers. The deals typically run five years, with some extensions under discussion. Pricing reflects market conditions within agreed ranges, and customer deposits help ensure purchase plans turn into actual orders. Those agreements already cover more than 50% of planned capacity for 2027 and 2028. When current negotiations are included, DRAM and high-bandwidth memory through 2029 are effectively sold out. The company expects demand to stay ahead of supply through 2030 because advanced process complexity and new fab construction take time. Near-term demand comes from enterprise and personal AI; longer-term demand comes from physical AI such as autonomous driving and robotics. HBM4 12-layer products are shipping on schedule. M15X capacity is set to reach maximum by year-end. HBM4 is expected to become the main high-bandwidth memory product by volume and revenue this quarter. Next year’s capital spending is expected to rise substantially, including the start of the first Yongin fab and more extreme-ultraviolet process capacity at existing plants. Spending levels and timing will be set against the multi-year demand visibility from customer talks. The comments effectively push back on peak-cycle worries tied to China’s CXMT. CXMT said on October 10 it would release a server DRAM module using a denser 4F² cell structure by year-end. SK Hynix shares fell more than 10% on October 13. Industry views still treat CXMT’s threat as unproven: the company has not disclosed capacity, speed, or customers, its process is judged two to three generations behind, it cannot use extreme-ultraviolet tools, and server products require customer qualification. Eugene Investment said CXMT’s influence is more likely to matter after 2029–2030. KB Securities projects 389 trillion won in operating profit for SK Hynix next year and notes that Samsung and SK Hynix now trade at roughly 3 times earnings, levels last seen in deep market stress. $SKHY

SK Hynix just pushed back hard on the idea that the memory cycle is peaking. It says high-bandwidth memory is effectively sold out through 2029 and shortages will last until 2030. The company posted record results: revenue of 104.62 trillion won, up 328% year-over-year and 32% from the prior quarter, and operating profit of 83.33 trillion won, up 632% year-over-year. Operating margin reached 79.7%. CEO Kwak Noh-jung said long-term supply agreements are turning customer trust into a different growth structure. It has signed long-term agreements with 14 customers. Those deals already cover more than half of planned capacity for 2027 and 2028. When current negotiations are included, DRAM and high-bandwidth memory volume through 2029 is effectively sold out. Customer requests for high-bandwidth memory are running at twice available supply. CFO Kim Woo-hyun said feasible demand growth next year looks like mid-20% for DRAM bits and high-teens for NAND, with supply lagging through 2030. On the concern that some buyers might switch from 12-layer to 8-layer stacks, the company said total capacity needed does not fall. Fewer layers per stack simply means more stacks, which raises demand for base dies and back-end processing. It also noted that major hyperscaler cloud revenue grew 30% to 80% year-over-year and that AI paid users keep rising. Global cloud capital spending is projected to rise more than 60% next year, and the AI data-center infrastructure market is expected to grow at a 40% annual rate through 2030. Investment is also spreading beyond a few big tech firms to enterprise, sovereign AI, and on-premise demand. The company announced an additional 60 trillion won in shareholder returns, split evenly between dividends and buybacks. Combined with the earlier 40 trillion won buyback and cancellation, the total for the year reaches 100 trillion won. Net cash stands at 98.7 trillion won. $SKHY If SK Hynix has already locked in more than half of 2027 and 2028 capacity and says DRAM and high-bandwidth memory are effectively sold out through 2029, does the next constraint stay on how fast it can add cleanroom space, or on whether those long-term agreements hold if any major customer delays its own build-out?