SK Hynix is slated to commence deliveries of its sixth-generation HBM4 modules to Nvidia during February 2026, marking the first customer shipments of its high-performance, low-power DRAM targeted at AI accelerators. Initial deliveries will support early production runs of the Vera Rubin GPUs ahead of broader product rollout.

HBM4 delivers significantly higher bandwidth and reduced power consumption compared with HBM3, leveraging advanced through-silicon vias and tighter stack integration. These enhancements are designed to meet the demanding data throughput and energy efficiency requirements of large-scale AI model training and inference.

The availability of HBM4 modules is critical for Nvidia's Vera Rubin GPU ramp, as high-bandwidth memory is a key cost and performance driver. Secured supply of next-generation memory could accelerate Nvidia's AI accelerator shipments and support margin improvements through optimized component sourcing.