Oh no: NVIDIA's next-gen Vera Rubin AI systems to eat up MILLIONS of terabytes of SSDs

The only other word I heard more than "AI" at CES 2026 was "DRAM" and the on-going crisis, but now it could get worse with reports that NVIDIA's next-gen Vera Rubin AI systems will eat up MILLIONS of terabytes of SSD capacity in the years to come.

That's just Vera Rubin let alone Rubin Ultra, let alone NVIDIA's next-gen Feynman GPU architecture after that... but in a new X post by @Jukan, we're hearing from a Citi analysis of the subject.

Citi explained: "We estimate that approximately 1,152TB of additional SSD NAND will be required per Vera Rubin server system to support NVIDIA's ICMS operations. Accordingly, assuming Vera Rubin server shipments of 30,000 units in 2026 and 100,000 units in 2027, NAND demand driven by ICMS is projected to reach 34.6 million TB in 2026 and 115.2 million TB in 2027".

"This represents 2.8% of expected global NAND demand in 2026 and 9.3% in 2027, a meaningful scale that is likely to create significant upside potential for demand. As a result, with this structural demand increase factored in, the global NAND supply shortage is expected to intensify further".

NVIDIA could use up to 16TB of NAND per GPU per rack = 1152TB of SSDs in a single Vera Rubin NVL72 AI server configuration. Citi estimates that Vera Rubin shipments could scale to 100,000 units in 2027, meaning that NVIDIA alone would require 115.2 million TB, which represents around 9.3% of the total global NAND demand projected for the next few years.

I have heard many times from industry contacts that after the DRAM crisis will come the NAND crisis, and we'll see SSDs and M.2 SSDs skyrocket in price... sigh.