Samsung will begin mass shipments of HBM4 memory in February, beating competitors to the punch

Samsung Electronics will be the first company to supply HBM4 chips for commercial production

When Who First Commercial Supply of HBM4 Memory Mid-February Nvidia (Vera Rubin platform)
* Supplier – Samsung Electronics, the first manufacturer that will begin shipping HBM4 chips.

* Customer – Nvidia. The memory will be used in the next‑generation computing platform Vera Rubin, designed for AI supercomputers.

* Public unveiling – Accelerators based on HBM4 are planned to be showcased at the GTC 2026 conference (March).

Technical specifications
Parameter Value Transfer speed up to 11.7 Gb/s (37 % higher than the JEDEC standard of 8 Gb/s) Throughput per stack 3 TB/s Capacity with a 12‑layer configuration 36 GB Potential capacity with a 16‑layer configuration up to 48 GB
* The chip was designed to exceed industry JEDEC standards. It uses a sixth‑generation DRAM process (10 nm, 1c) and Samsung’s own 4‑nanometer wide logic crystal.

Manufacturing
* Production site – Pyeongtaek Campus Line 4 plant.

* After modernization, the output of HBM4 wafers will be about 200,000 units per month, covering roughly 25 % of all Samsung DRAM manufacturing capacity.

Market outlook
Company Market share estimate for HBM4 SK hynix ~70 % Samsung ~30 % Micron almost no position
According to SemiAnalysis analytics, the HBM4 market will be largely divided between SK hynix and Samsung; Micron lags behind.

Strategic significance
In previous generations of HBM, Samsung lagged behind competitor SK hynix. With the launch of HBM4, the gap may not only close but surpass the competitor, given the growing demand from data centers for AI processing.