Nvidia Excludes Micron from HBM4 Supply Chain for Vera Rubin AI Chip
Nvidia has finalized Samsung and SK Hynix as the sole suppliers of HBM4 memory for its next-generation Vera Rubin AI accelerator, sidelining Micron Technology. The decision sent Micron's stock tumbling 6.74%, while Samsung and SK Hynix shares dropped 7.81% and 9.52%, respectively. Nvidia itself saw a 3.01% decline.
Samsung cleared Nvidia's HBM4 quality tests at 10 Gbps and 11 Gbps, while SK Hynix continues work on meeting the 11 Gbps threshold. SK Hynix is projected to supply over 50% of Nvidia's HBM4 in 2026, with Samsung capturing 28% of the volume. Production begins in March, with Vera Rubin slated for launch in late 2026.
Though excluded from the flagship Vera Rubin product, Micron will still supply HBM4 for the mid-tier Rubin CPX inference accelerator. The NVL72 rack configuration, pairing 72 Rubin GPUs with 36 Vera CPUs, promises 10x better performance per watt than the current Blackwell architecture.
