SK Hynix Predicts AI Infrastructure Performance to Grow 20x by 2030
At SEDEX 2026, SK Hynix Development Head Ahn Hyun predicted that AI infrastructure performance must scale by approximately 20 times by 2030. This surge is driven by rapid expansions in AI inference demands, context lengths, and task complexities. As AI inference workloads explode, data movement between storage hierarchy layers has become a major bottleneck restricting performance and increasing energy consumption. Addressing this requires holistic architectural solutions rather than just upgrading individual memory chips. SK Hynix is advancing next-generation memory technologies including 3D stacked DRAM, which places memory directly on logic chips for lower power and higher bandwidth, and High Bandwidth Flash (HBF), which brings silicon-through-via stacking to NAND flash.
## BACKGROUND
KV Cache (Key-Value Cache) is a crucial optimization technique in large language models that stores intermediate attention mechanism calculations to speed up text generation, though its growing footprint strains memory systems. High Bandwidth Flash (HBF) is an emerging memory tier positioned between HBM and SSDs to provide high-capacity, high-bandwidth storage for AI workloads.