South Korean chipmaker SK Hynix has started mass production of its 192GB SOCAMM2 memory module, a next-generation unit built on sixth-generation low-power DRAM technology and designed specifically for Nvidia's upcoming Vera Rubin AI platform. The module delivers more than double the bandwidth and over 75 percent better power efficiency compared with conventional server memory, making it well-suited for training and running large AI models. SK Hynix says the product sets a new performance standard for AI memory and is intended to ease the memory bottlenecks that currently limit large-scale artificial intelligence infrastructure.