SK海力士：HBM4已准备好首次量产，预计功耗效率提升40%

Core Insights - SK Hynix has completed the development of the next-generation HBM4 memory, achieving conditions for large-scale production, reinforcing its technological leadership in AI memory [1][3] - The new product features 2048 I/O terminals, doubling the bandwidth and improving power efficiency by over 40%, with a running speed exceeding 10Gbps, surpassing the JEDEC standard of 8Gbps [1][2] - The company anticipates that the application of HBM4 will enhance AI service performance by up to 69%, addressing data bottlenecks and significantly reducing data center electricity costs [1][3] Technical Specifications - HBM4 achieves significant breakthroughs in technical specifications, with 2048 I/O terminals doubling the bandwidth and improving power efficiency by over 40% [2] - The running speed of HBM4 exceeds 10Gbps, well above the industry standard of 8Gbps [2] Production and Market Strategy - To ensure stable large-scale production, SK Hynix employs the market-validated Advanced MR-MUF process and 10nm technology, aimed at minimizing technical risks during production [3] - The completion of HBM4 development is seen as a milestone for the industry, with the company focusing on timely supply of products that meet customer demands in performance, power efficiency, and reliability [3] Industry Positioning - SK Hynix's AI Infrastructure President, Justin Kim, describes HBM4 as a symbolic turning point that will be central to overcoming technological challenges [4] - The company aims to become a full-stack AI memory supplier by providing high-quality and diverse performance memory products needed in the AI era, as demand for high-bandwidth memory continues to surge [4]