SK hynix Sees AI Infrastructure Growing 20-Fold by 2030

Ahn Hyun, President and Chief Development Officer of SK hynix, delivers a keynote speech titled “Asking About the Direction of Memory Innovation in the AI Era” at the 28th Semiconductor Trade Exhibition (SEDEX 2026) held at COEX in Gangnam-gu, Seoul, on Oct. 14. (Yonhap) As the demand for artificial intelligence (AI) inference surges, competition in the semiconductor industry is expanding beyond the performance of individual memory toward optimizing the overall structure. SK hynix has put forward next-generation memory technologies, including the advancement of high bandwidth memory (HBM), three-dimensional (3D) stacked DRAM that vertically stacks semiconductor chips, and high bandwidth flash (HBF), which has a storage capacity 10 times larger than HBM. During his keynote speech at the 28th Semiconductor Exhibition (SEDEX) 2026 held at COEX in Gangnam-gu, Seoul on Oct. 14, Ahn Hyun, head of development at SK hynix, explained the rapid growth trend of the AI inference market and forecasted, “The performance scale of AI infrastructure will grow up to about 20 times by 2030 compared to this year.” Ahn diagnosed that the AI inference market is growing steeply around three axes: ‘task complexity,’ ‘context length,’ which is the amount of information processed at once, and ‘inference request volume.’ He explained that due to the explosion in demand for AI inference, the capacity of ‘KV cache,’ which stores and reuses previously processed information, is also significantly increasing, and as a result, bottlenecks are occurring in data movement between different memory hierarchies such as HBM, DRAM, and solid state drives (SSD), slowing down AI processing speeds and increasing power consumption. The next-generation technology that SK hynix presented as a solution is 3D stacked DRAM. This is a technology that reduces the physical distance between the computing unit and the data by stacking DRAM on top of logic semiconductors such as graphics processing units (GPU) and neural processing units (NPU). Another solution, HBF, is a next-generation memory that applies the through-silicon via (TSV) stacking and high-speed input/output technologies of HBM to NAND flash, establishing a new memory hierarchy between HBM and DRAM. SK hynix aims to secure the bandwidth of HBF at half the level of the same generation of HBM while increasing the capacity by about 10 times. Samsung Electronics also showcased next-generation memory and advanced packaging technologies that maximize data movement efficiency at SEDEX 2026. Samsung Electronics exhibited HBM4E (7th generation HBM), a high-performance memory for AI, and LPDDR-PIM, which integrates computing functions inside the memory. The company also unveiled mock-ups of ‘zHBM’ based on 3D memory technology and ‘zNAND-O,’ a 3D NAND solution optimized for edge AI. Ha Jung-woo, vice chairman of the National AI Strategy Committee, visited the Samsung Electronics booth and evaluated, “As the global number one company in the semiconductor industry, I was able to check the roadmap on how it can develop by connecting AI and semiconductors in the future.” Hosted by the Korea Semiconductor Industry Association, this exhibition features 300 semiconductor-related companies operating 888 booths.