Rumor
NVIDIA reportedly switches Rubin CPX memory specification from GDDR7 to HBM
Wednesday, February 18, 2026 at 03:19 AM
Industry channel checks indicate that NVIDIA has modified the memory specifications for its Rubin CPX GPU, switching from GDDR7 to HBM. While Rubin CPX was originally intended to handle less memory-intensive AI prefill workloads using GDDR7, real-world performance requirements for memory bandwidth and capacity have reportedly necessitated the upgrade to HBM. This shift is expected to further tighten the global HBM supply if finalized.
Context
Nvidia is reportedly pivoting its upcoming Rubin CPX GPU memory specification from GDDR7 to HBM. Originally, the Rubin CPX was designed to handle "prefill" operations—a specific segment of AI inference workloads—using standard graphics memory. However, recent channel checks indicate that real-world performance requirements have forced a shift to high-bandwidth memory. This transition suggests that even secondary AI tasks now demand the extreme throughput that only HBM can provide, rendering GDDR7 insufficient for Nvidia's next-generation architecture.
This unexpected spec change carries significant implications for the semiconductor supply chain, particularly for leading suppliers like Micron. Because most industry analysts have not yet factored Rubin CPX demand into their HBM models, this shift could trigger a massive spike in latent demand. With HBM supply already projected to be tight through 2025, this move is expected to intensify global shortages. For investors, it reinforces the necessity of high-bandwidth memory across the entire AI hardware stack and suggests significant revenue tailwinds for memory producers.
Related Companies
Nvidia
NVDA