NVIDIA to Test Lower HBM Configurations for Rubin Ultra GPU Amid Supply Constraints

Huoxing Finance reports that on August 7, NVIDIA is evaluating adjustments to the HBM configuration of its next-generation AI GPU, Rubin Ultra, planning to launch a version with lower memory capacity than originally intended to alleviate production pressures caused by shortages of high-end HBM. Sources familiar with the matter revealed that over the past several weeks, NVIDIA has tested at least three different memory capacity variants, some using lower-spec memory. This indicates that even NVIDIA, which dominates the GPU market, must compromise between product specifications and supply capabilities. Rubin Ultra is positioned above the upcoming Rubin series and was originally planned to feature higher-capacity and higher-bandwidth HBM to enhance performance in large model training and inference. If the lower-memory configuration is ultimately adopted, customers running large AI workloads such as large language models will need to deploy more GPUs, increasing system costs and cluster complexity. For cloud providers like Microsoft, Meta, Amazon, and Google, which continue to expand their AI capital expenditures, data center construction costs may rise further. NVIDIA’s move also underscores that, despite its strongest bargaining power in the supply chain, HBM supply constraints remain a core bottleneck in AI computing expansion. Previously, high-end HBM capacity from SK Hynix and Micron has remained tight; NVIDIA’s proactive reduction in configuration further confirms that the HBM supply-demand imbalance is unlikely to ease in the short term.