On April 20, 2026, SK Hynix made a significant announcement regarding the commencement of mass production for its 192GB SOCAMM2 memory modules, which are tailored specifically for Nvidia’s Vera Rubin platform. Leveraging sixth-generation 10-nanometer LPDDR5X low-power DRAM technology, these modules incorporate a compression connector design. This innovative feature not only ensures a slim profile but also enhances scalability, making them an ideal choice for next-generation AI servers.

When compared to traditional RDIMMs, the SOCAMM2 modules offer a substantial performance boost, delivering more than twice the bandwidth. Additionally, they achieve a remarkable reduction in power consumption by over 75%. This efficiency not only alleviates storage bottlenecks commonly encountered in AI large model training and inference processes but also significantly enhances the overall system processing speed.