Vi e w p o i n t | 11 Jan 2026 16:22:46 ET │ 11 pages Global Semiconductors NAND Supply Shortage Expected to Deepen Further as Nvidia Adopts Inference Context Memory Storage CITI'S TAKE Nvidia recently announced it will adopt an Inference Context Memory Storage (ICMS) for Vera Rubin platform, with the aim of resolving memory bottlenecks in large-scale inference computations. The new architecture is expected to adopt 16TB TLC SSDs and offload Key-Value (KV) Cache to more scalable storages to enhance Rubin's ...