Nvidia's AI Server Cluster Rubin NVL72 Faces Memory Configuration Changes

Nvidia's upcoming AI server cluster, Rubin NVL72, has undergone significant memory configuration changes, according to a recent report by SemiAnalysis. The report highlights that due to supply chain constraints, the memory capacity per cabinet has been reduced from the initially planned 55TB to 28TB, a decrease of approximately 50%. The original 192GB high-capacity SOCAMM memory modules have been replaced with a reduced 96GB version. According to Odaily, this development has impacted memory-related stocks, including Micron and SK Hynix, which have experienced downward pressure. SemiAnalysis founder and CEO Dylan Patel commented on the situation, noting that while people often share their findings, they tend to overlook much of the content in the original report. Patel emphasized that their initial report did not use subscription-driven headlines.