Samsung HBM3E Passes NVIDIA, HBM4 Race Begins

According to Korean media reports, Samsung Electronics' 12-layer fifth-generation HBM3E has reportedly passed NVIDIA's rigorous quality tests and is expected to begin shipping soon. This marks a key milestone, positioning Samsung as a critical contributor to NVIDIA's most advanced AI accelerators.

Industry insiders revealed that it took 18 months after development for Samsung to meet NVIDIA's demanding performance standards, following several unsuccessful attempts. The HBM3E has also cleared validation on AMD's MI350 and is compatible with NVIDIA's new B300 Blackwell Ultra GPU.

While NVIDIA has already secured significant HBM3E volumes from SK hynix and Micron, Samsung's participation carries symbolic weight. For Samsung, the win is less about immediate revenue and more about regaining technological credibility. With this breakthrough, the company is confident it can stay on track for HBM4—expected to power NVIDIA's upcoming Vera Rubin architecture.

HBM4 will push memory bandwidth even higher, with NVIDIA requiring speeds beyond 10Gbps per pin compared to the current 8Gbps standard. Samsung has already demonstrated 11Gbps performance using 1c DRAM and 4nm logic from its foundry, surpassing SK hynix's 10Gbps results, while Micron still trails behind.

Reports suggest Samsung is the first to hit the 11Gbps milestone, and the HBM4 ecosystem could extend to major players like AMD, Broadcom, and Google. Samsung plans to ship large volumes of HBM4 samples to NVIDIA this month in hopes of securing early certification, with mass production expected in the first half of 2026.

If validated ahead of rivals, Samsung stands a strong chance of reclaiming share in the next wave of high-bandwidth memory for AI.