Rubin's reduction in allocation leads to a deep pullback in memory stocks, market opinion summary: may only be a transitional plan, positive for SSD and optical interconnect BlockBeats news, on June 5th, this morning last night, well-known investment research firm SemiAnalysis released an investment research report stating that Nvidia's next-generation AI server cluster Rubin NVL72 will make significant adjustments in memory configuration, with the single cabinet capacity dropping from the original plan of 55TB to 28TB. At the same time, most Rubin systems will use 96GB SOCAMM modules instead of the original plan of 192GB modules. The report caused a huge market shock, with Micron closing down 7.7% and South Korean SK Hynix opening down 8.32%. The market view is generally cautious and optimistic about this, believing that there is an overreaction in the market. US stock KOL Herman Jin stated that the fundamental reason for the reduction in memory allocation is insufficient supply, not a decrease in demand. The Nvidia switch solution may also have similar negative news. It emphasizes once again that the demand on the model side is the key indicator of whether the AI craze has ended. Another viewpoint suggests that the next step for the market should focus on whether the memory allocation reduction measures are just a transitional solution. In addition, when the system memory used to carry large context (KV Cache) on the CPU side shrinks, GPU computing power bottlenecks will inevitably shift to the SSD side and interconnect side. CSPs need to purchase more high-performance SSDs or adopt higher performance in cabinet connection solutions, which is beneficial for NAND enterprises such as KIOXIA, SanDisk, and optical link enterprises LITE, Marvell (MRVL), and Corning.