[반도체 온에어] SK하이닉스, '소캠2' 양산…'베라 루빈'에 최적화

SK Hynix's mass production of SOCAMM2 signals a critical shift in the AI infrastructure market from training-heavy setups to power-efficient inference environments. By optimizing this high-capacity, low-power module for NVIDIA Vera Rubin, the company is directly addressing memory bottlenecks in large language models. This move diversifies their AI memory portfolio beyond HBM to capture the growing demand for sustainable AI server performance.