Securing NVIDIA orders, as per reports, Samsung readies the mass production of HBM4 chips, challenging SK hynix in the AI memory race.
Samsung is reportedly close to becoming the first supplier to mass-produce and ship HBM4 after the Lunar New Year. According to a Bloomberg report, its initial share is expected in the mid-20% range.
The chips are set for use in NVIDIA GPUs powering generative AI systems, aligning with the launch of its next-generation Vera Rubin accelerator.
Meanwhile, NVIDIA is preparing to unveil Vera Rubin, at GTC 2026 in mid-March. The launch has intensified competition among memory suppliers racing to secure HBM4 orders for high-end AI systems. Industry sources say NVIDIA has provisionally allocated supply volumes for late 2025.
At the same time, SK hynix is expected to receive the largest portion of early HBM4 demand, in the mid-50% range. Samsung is likely to follow in the mid-20% band. Micron is estimated to hold about 20%. The allocations reflect capacity planning and qualification timelines, as HBM production requires long lead times.
Samsung is said to have cleared NVIDIA’s quality tests ahead of schedule. Shipments are expected to begin from the third week of this month. The company is positioning its HBM4 as a performance leader. It combines sixth-generation 1c DRAM with a 4nm logic base die.
This approach delivers data rates of up to 11.7Gbps. The single-stack bandwidth reaches up to 3TB/s. That is around 2.4 times higher than the prior generation. Capacity is 36GB with 12-high stacks. It could reach 48GB with 16-high stacking.
According to a TrendForce report, Costs remain higher due to advanced nodes and packaging complexity. Output of 1c DRAM is about 70,000 wafers per month, near 10% of Samsung’s DRAM capacity. Expansion at Pyeongtaek Line 4 is underway. Monthly output is expected to reach approximately 190,000 wafers within a year.
After the reports, on Monday, February 9, 2026, Samsung’s shares jumped 5.4% to 168,700 won.
