Nvidia has finalized the supplier lineup for its upcoming Vera Rubin AI accelerators, with Samsung Electronics and SK Hynix securing exclusive rights to supply the critical HBM4 memory. According to reports, Micron Technology, a long-time supplier for previous HBM generations, was unexpectedly left off the initial vendor list for the flagship product.

Industry sources indicate the two Korean giants met Nvidia's stringent performance and yield requirements. Nvidia reportedly set two quality benchmarks for HBM4: 10Gbps and 11Gbps speeds. Samsung has largely passed validation and delivered initial samples last month, while SK Hynix is optimizing its process to meet the higher standard.

Vera Rubin, set to debut at GTC 2026 on March 16th, aims to deliver 5x the performance of current systems. Each accelerator will feature 16 HBM4 stacks totaling 576GB, surpassing AMD's next-gen MI450 which offers 432GB. HBM4's high speed and capacity are critical to achieving this leap.

While Micron may still supply HBM4 for potential mid-tier Rubin variants, its absence from the top-tier lineup marks a significant shift. Mass production for HBM4 could begin as early as this month.

ICgoodFind : The HBM4 battle is decided. Samsung and SK Hynix dominate Nvidia's next-gen supply chain, leaving Micron on the sidelines for now.