Analysis: NVIDIA may adjust Rubin Ultra HBM specifications to ease HBM supply pressure

Citrini analyst Jukan published a report stating that NVIDIA is discussing adjusting the HBM configuration of its next-generation Rubin Ultra GPU, possibly shifting the main version from HBM4E 12Hi 384GB to HBM4 8Hi 192GB, with an HBM4E 8Hi version to be launched later. The analysis notes that if NVIDIA’s CoWoS capacity allocation reaches 1.2 million units in 2027, its accelerator output is expected to hit around 9.9 million units, of which Rubin Ultra production will be roughly 1.6 million units. If the lower-spec HBM scheme is adopted, NVIDIA’s 2027 HBM demand is projected to drop from 24.1 billion Gb to 21.6 billion Gb, a decrease of about 10%, with overall HBM demand potentially falling by around 4%. Jukan argues this "spec downgrade" is not due to weak demand, but rather NVIDIA’s strategy to boost accelerator output amid limited HBM supply. As DRAM manufacturers’ HBM expansion pace cannot match TSMC’s advanced process and packaging capacity growth, HBM supply has become a bottleneck for AI chip production. Additionally, DRAM makers also struggle to significantly adjust their capacity allocations. Currently, smartphone and PC vendors in the consumer electronics sector are grappling with storage shortages, and further cuts to traditional DRAM supply could intensify pressure on the end market. The analysis concludes that by adjusting HBM configurations, NVIDIA is poised to unlock more AI accelerator capacity with limited storage resources while easing the HBM supply bottleneck.