Odaily Planet Daily reports that Citrini analyst jukan said in a post on X that, according to TrendForce’s latest storage industry research, the DRAM supply shortage is expected to continue until 2027, and the HBM4e certification timeline remains uncertain. Since the third quarter of 2026, NVIDIA has been re-evaluating the HBM configuration for Rubin Ultra.

The original plan was to use a 12-layer stacked HBM4e configuration, but NVIDIA is currently parallel evaluating multiple designs, including an 8-layer stacked HBM4e configuration, a 12-layer stacked HBM4 configuration, and an 8-layer stacked HBM4 configuration. The final specifications have not yet been confirmed. In addition to NVIDIA, some CSPs are reportedly also considering reducing the HBM capacity of their next-generation in-house ASICs.

NVIDIA will use 12-layer stacked HBM4e as the baseline design for Rubin Ultra during the first half of 2025 through 2026. But starting in the early third quarter of 2026, NVIDIA began reviewing lower-spec alternative options. TrendForce says this change is driven by two key supply-side constraints: first, the expected overall DRAM shortage in 2027 will limit the wafer capacity that storage manufacturers can allocate to HBM production; second, uncertainty remains regarding the HBM4e 12-layer stacked certification timeline and the rate at which mass-production yields can improve.

TrendForce states that in the Rubin Ultra generation, NVIDIA’s top priority is to increase I/O speed, while expanding GPU shipments is a secondary priority. If NVIDIA ultimately decides to downgrade the HBM specifications, it is expected to do so by reducing the number of stacked DRAM layers. Whether HBM4e can complete certification on schedule and enter mass production will determine whether Rubin Ultra’s I/O speed can rise from the previous-generation Rubin’s 8 to 11.7Gbps to 14 to 16Gbps, or whether it can be maintained at around 11 to 12Gbps through optimizations to the HBM4 design. Within the same product generation, the number of stacked DRAM layers determines the trade-off between the HBM capacity per GPU and the number of GPUs that can be shipped.

TrendForce also believes that the final configuration will depend on the wafer-allocation decisions made by storage manufacturers. On the supply-demand side, the HBM bit shipments in 2027 are expected to grow year over year by 50% to 60%, but they are still expected to fall short of demand growth. With supply constraints continuing, HBM suppliers are expected to retain pricing power throughout all of 2027. The industry widely expects HBM prices to rise significantly. AI chip makers will face dual pressure—limited HBM supply and higher procurement costs—further strengthening the motivation to adopt configurations with lower HBM capacity.