Reports indicate that Samsung's HBM4 has passed final quality tests for NVIDIA's upcoming "Vera Rubin" AI platform. This means mass production could begin next month. The new memory modules may be showcased at the GTC 2026 AI conference in March. Industry insiders suggest that Samsung's HBM4 will be integrated into the next-generation "Rubin" AI accelerators.

This development is important because NVIDIA's current "Blackwell" AI GPUs use HBM3E. Previously, Samsung lagged behind rivals in the HBM3 and HBM3E era. The new HBM4 is expected to operate at 11.7 Gb/s, exceeding the 10 Gb/s JEDEC standard. Full-scale supply is projected to commence by June, aligning with the "Vera Rubin" manufacturing schedule.

NVIDIA announced an upgrade to the Vera Rubin memory bandwidth at CES 2026. This upgrade could be enabled by Samsung’s HBM4 modules. The high-bandwidth memory market is crucial for the performance of advanced AI accelerators. Quanta Computer anticipates the first shipments of server hardware to hyperscaler customers starting late summer.

High Bandwidth Memory is vital for AI applications, and Samsung's success here could shift the balance of power. The competition between Samsung, NVIDIA, and other players will be intense. As AI continues to evolve, the demand for cutting-edge memory solutions will only increase, making this a space to watch.