Analysis: NVIDIA May Adjust Rubin Ultra HBM Specifications to Alleviate HBM Supply Pressure

August 4th, Citrini analyst Jukan posted that NVIDIA is discussing adjusting the HBM configuration scheme of the next-generation Rubin Ultra GPU, possibly changing the main version from HBM4E 12Hi 384GB to HBM4 8Hi 192GB, and later releasing the HBM4E 8Hi version.

The analysis stated that if NVIDIA's 2027 CoWoS capacity allocation reaches 1.2 million wafers, it is expected that its accelerator output will be around 9.9 million units, with approximately 1.6 million units being Rubin Ultra. If a lower-spec HBM scheme is adopted, NVIDIA's 2027 HBM demand is expected to decrease from 24.1 billion Gb to 21.6 billion Gb, a decrease of about 10%, and the overall HBM demand may fall by around 4%.

Jukan believes that this "spec downgrade" is not due to weakened demand but rather NVIDIA's strategy to increase accelerator output in a limited HBM supply situation. As the expansion speed of HBM by DRAM manufacturers cannot match TSMC's advanced processes and packaging capacity expansion, HBM supply has become a limiting factor for AI chip output.

In addition, DRAM manufacturers also find it difficult to significantly adjust capacity allocation. Currently, consumer electronics companies in the smartphone and PC sectors are facing storage shortages, and further reducing traditional DRAM supply may exacerbate market pressure at the end.

The analysis believes that NVIDIA, by adjusting the HBM configuration, is expected to unleash more AI accelerator capacity under limited storage resources and alleviate the HBM supply bottleneck.