The Vera Rubin platform from NVIDIA is expected to debut in the second half of 2026, featuring GIDS (GPU Independent Direct Storage) which allows the GPU to manage memory and storage autonomously. This architecture enables the use of HBF (High Bandwidth Flash), a high-capacity memory tier that offers up to 16 times the density of HBM at a lower cost per bit. The integration of HBF, HBM4, and NVLink 6 aims to resolve memory bottlenecks for AI scaling while significantly reducing power consumption in server racks.