Samsung Electronics has achieved a significant breakthrough in the high-bandwidth memory market, with its next-generation HBM4 memory outperforming competitors in speed and power efficiency tests for Nvidia’s upcoming Vera Rubin AI accelerator, according to Maeil Business Newspaper. During a visit last week, Nvidia representatives confirmed that Samsung delivered “the best results in the memory industry,” prompting a request for supply volumes that far exceeded Samsung’s internal projections. This success marks a dramatic reversal from the HBM3E generation, where Samsung lagged behind rivals by nearly a year.
The turnaround is attributed to a high-risk technical strategy in which Samsung skipped the D1b DRAM process entirely to advance directly to the 10-nanometer D1c process. By combining this new DRAM with logic dies produced using a 4-nanometer foundry process, Samsung became the first manufacturer to achieve data transfer speeds exceeding 11 gigabits per second. While SK hynix maintains a roughly three-month lead—having already started supplying paid samples—Samsung has successfully narrowed the gap from the previous generation.
Market data reflects this recovery, with Samsung reclaiming the number two spot in the HBM market during the third quarter of 2025 with a 22% share, overtaking Micron. Contracts are expected to be formalized in the first quarter of 2026, with full-scale deliveries scheduled for the second quarter to meet Nvidia’s timeline for the Vera Rubin platform launch in Q3 2026.
