[Analysis: NVIDIA May Adjust Rubin Ultra HBM Specifications to Ease HBM Supply Pressure] BlockBeats News, August 4th, Citrini analyst Jukan published an article stating that NVIDIA is discussing adjustments to the HBM configuration plan for the next-generation Rubin Ultra GPU. The main version may be adjusted from HBM4E 12Hi 384GB to HBM4 8Hi 192GB, with an HBM4E 8Hi version to be introduced later. The analysis indicates that if NVIDIA's CoWoS production capacity allocation reaches 1.2 million units in 2027, its accelerator output is expected to be approximately 9.9 million units, with Rubin Ultra accounting for about 1.6 million units. If a lower-spec HBM configuration is adopted, NVIDIA's HBM demand in 2027 is projected to decrease from 24.1 billion Gb to 21.6 billion Gb, a reduction of about 10%. Overall HBM demand may decline by approximately 4%. Jukan believes this "specification downgrade" is not due to weakened demand but rather a strategy by NVIDIA to increase accelerator production under limited HBM supply conditions. As DRAM manufacturers' HBM production expansion cannot keep pace with TSMC's advanced process and packaging capacity growth, HBM supply has become a bottleneck for AI chip production. Additionally, DRAM manufacturers face challenges in significantly reallocating production capacity. Currently, consumer electronics sectors such as smartphones and PCs are already affected by memory shortages, and further reductions in traditional DRAM supply could exacerbate pressure on the end market. The analysis suggests that by adjusting HBM configurations, NVIDIA is likely to release more AI accelerator capacity under constrained memory resources while alleviating the HBM supply bottleneck. [Original Link]