NVIDIA's next-gen Vera Rubin AI systems are projected to consume millions of terabytes in SSD capacity in the upcoming years, potentially triggering a NAND supply shock.
NVIDIA's New Storage Solution Could Alone Take Up a Massive Portion of the Global NAND Output
One of the biggest bottlenecks in agentic AI environments is that query processing generates a huge temporary memory log for context building called the KV Cache, and currently, the data is stored in HBM modules. However, considering how rapidly data requirements within AI clusters are scaling up, HBM cannot hold the capacity onboard, which is why at CES 2026, NVIDIA announced that the Bluefield-4 DPUs will be connected to a new storage solution, called the Inference Memory Context Storage (ICMS). While this will significantly boost data processing, it could trigger a situation similar to DRAM shortages.
Based on an analysis by Citi, it is claimed that with one Vera Rubin system, NVIDIA could equip roughly 16 TB of NAND per GPU in a rack, which accounts for 1,152 TB in a single NVL72 configuration. And, based on Citi's estimates, Vera Rubin shipments could scale up to 100,000 units in 2027, which means that, from NVIDIA alone, the demand for NAND storage could increase to 115.2 million TB, representing 9.3% of the total global NAND demand projected for the upcoming years. Vera Rubin, equipped with ICMS, could alone create a supply shock that the NAND industry has not yet factored in.
Given that NVIDIA has projected agentic AI to be the next primary focus of the applications layer, the need for a sufficient KV Cache pool is significant for upcoming server racks, which is why the demand for ICMS is expected to increase. It is also important to factor in that the NAND industry is already experiencing shortages due to the ongoing data center buildout, the inference craze, and the fact that NVIDIA now aims to acquire a significant portion of the total global NAND output.
It would not be incorrect to say that the NAND industry could face a similar situation to what we are witnessing with DRAM, given that AI manufacturers aren't looking to halt their pursuit of superior compute capabilities. And, well, for the average consumer, getting access to general-purpose SSDs and storage devices could become another nightmare.
News Source: Jukan
About the author: Muhammad Zuhair is a hardware and technology reporter for Wccftech, specializing in the semiconductor industry and the complex interplay between technology, manufacturing, and geopolitics. His coverage focuses on the corporate strategies and technological roadmaps of industry giants like TSMC, NVIDIA, Samsung, and Intel. Zuhair's expertise lies in deconstructing complex topics such as fabrication nodes (e.g., 2nm process), the economic impact of policies like the CHIPS Act, and the strategic development of AI infrastructure from NVIDIA, AMD and Intel.
Follow Wccftech on Google to get more of our news coverage in your feeds.
