Core Viewpoint - The competition in the low-power server memory sector is intensifying as Samsung Electronics and SK Hynix collaborate with major U.S. chip companies Nvidia and Intel, driven by the surge in memory demand due to the rise of artificial intelligence [1][2]. Group 1: SK Hynix Developments - SK Hynix has launched a new high-capacity, high-performance DRAM product, specifically a 256GB DDR5 RDIMM based on 32Gb fifth-generation 10nm DRAM, which has been certified for compatibility and performance by Intel [1][2]. - The new RDIMM is designed to maintain system stability even with multiple DRAM chips installed and is primarily targeted at enterprise data centers [1]. - SK Hynix claims that servers using this new RDIMM can achieve a 16% improvement in AI inference performance while reducing power consumption by 18% [2]. Group 2: Samsung Electronics Developments - Samsung has developed a new memory product optimized for AI data centers, called the Small Compression Additional Memory Module (Socamm2), which has been provided as samples to Nvidia [2][3]. - Socamm2 is a server memory module made of stacked low-power double data rate DRAM, offering more than double the bandwidth and a 55% reduction in power consumption compared to conventional RDIMMs [3]. - Samsung is collaborating with global partners to establish international standards for Socamm2, which is not yet commercialized due to unresolved standardization and platform compatibility issues [3]. Group 3: Market Dynamics - Despite Intel lagging behind Nvidia in the GPU sector, it still holds over 70% of the data center CPU market share, which is crucial for the adoption of new memory products [2]. - The competition among major memory chip manufacturers, including Micron, Samsung, and SK Hynix, is intensifying as they all aim to supply Socamm2 to Nvidia for its next-generation AI platform, Vera Rubin, expected to launch in the second half of next year [3].