Nvidia is considering a significant spec change to its upcoming Rubin Ultra GPU, potentially shipping the chip with far less high-bandwidth memory than originally targeted. The adjustment, reported by TrendForce in early August 2026, is a direct response to tightening supply of HBM from SK Hynix, Samsung, and Micron. The original vision for Rubin Ultra included configurations with up to 1 TB of HBM4E memory when unveiled at its GTC developer conference in 2025. The current Vera Rubin GPU ships with up to 288 GB of HBM4 in 12-Hi stacks, delivering roughly 22 TB/s of memory bandwidth. Nvidia is now testing at least three variants with reduced memory, including one configuration using 8-Hi HBM4 stacks carrying approximately 192 GB, a 33% reduction compared to Vera Rubin. TrendForce’s August 4 report noted Nvidia is evaluating four distinct HBM configurations, with the pivot driven by anticipated DRAM and HBM shortages expected to deepen into 2027. Yield and production challenges at all three major HBM suppliers have created a wafer allocation bottleneck. If Rubin Ultra ships with less memory per GPU, enterprises running very large AI models will likely need more GPUs, bringing additional costs for power, rack space, interconnect infrastructure, and licensing. A shift toward lower-tier HBM configurations affects demand for the most profitable memory products, influencing how SK Hynix, Samsung, and Micron prioritize production road