BlockBeats News, August 7th, Nvidia is evaluating the adjustment of the next-generation AI GPU Rubin Ultra's HBM configuration, planning to launch a lower-memory version than originally planned to alleviate the mass production pressure caused by the high-end HBM shortage. Insiders revealed that Nvidia has tested at least three different versions with varying memory capacities in the past few weeks, some adopting lower memory specifications. This means that even Nvidia, which dominates the GPU market, must make compromises between product specifications and supply capabilities. Rubin Ultra is positioned higher than the upcoming Rubin series, originally designed to be equipped with higher capacity and bandwidth HBM to enhance large model training and inference performance.

If the final configuration adopts lower memory, customers running large language models and other large AI workloads will need to deploy more GPUs, and system costs and cluster complexity will consequently increase. For cloud providers such as Microsoft, Meta, Amazon, and Google, which continue to expand their AI capital expenditure, data center construction costs may further rise. Nvidia's move also reflects that, despite the company having the strongest bargaining power in the industry chain, HBM supply constraints remain a core bottleneck for AI computing expansion. Previously, SK Hynix and Micron's high-end HBM production capacity has been consistently tight, and Nvidia's proactive reduction in configuration further confirms that the HBM supply-demand contradiction is difficult to alleviate in the short term.