NAND Memory Crunch Looms with Nvidia’s Ambitious Tech Expansion

As we already understand, the situation in the memory market won’t improve until 2027-2028, yet new data suggests that things could soon get significantly worse in the SSD market. City analysts note that Nvidia’s upcoming Vera Rubin accelerators will require a vast amount of memory. Our estimates suggest that to support operations on each Nvidia Vera Rubin server, approximately 1152 TB of SSD NAND memory will be needed. Accordingly, assuming Vera Rubin server deliveries reach 30,000 units in 2026 and 100,000 units in 2027, the projected demand for NAND memory driven solely by these servers will hit 34.6 million TB in 2026 and 115.2 million TB in 2027. This constitutes 2.8% of the expected global demand for NAND memory in 2026 and 9.3% in 2027, likely creating substantial potential for demand growth.

As a result, considering this structural demand increase, further strengthening of the global NAND memory supply shortage is anticipated.

Nvidia’s Latest Announcements at CES 2026

At CES 2026, Nvidia revealed that the Bluefield-4 DPUs will be connected to a new data storage solution called Inference Memory Context Storage (ICMS). This should significantly accelerate data processing, but, as City forecasts, it will notably increase NAND memory demand. According to Citi data, through the new solution, Nvidia might outfit each GPU in a rack with about 16TB of NAND memory, totaling 1152 TB in an NVL72 configuration. It’s worth recalling that SSDs have already become noticeably more expensive, but in light of RAM, the growth is very modest and often does not reach even 100%