Samsung is reportedly about to get the Nvidia nod for its high bandwidth memory (HBM) needs, specifically HBM4.

According to Bloomberg, the Samsung Electronics division is in the final qualification phase with the U.S. chip giant, finally putting it on par with HBM4 rival SK Hynix.

While the report claimed Samsung is aiming for mass production in February in time for March’s Nvidia GTC event, Korean news outlets estimated its custom HBM4E design to be completed between May and June of this year, in time for Nvidia’s next-gen Rubin GPUs to debut in the second half of 2026.

Sources claimed Samsung’s HBM4 chips achieved data transfer speeds of 10.7 Gb/s, meeting Nvidia’s threshold of 10 Gb/s, with some reports putting it at 11.7 Gb/s. In comparison, SK Hynix and Micron Technology recorded speeds of about 8.3 Gb/s and around 8 Gb/s, respectively, for their HBM4 efforts.

The Korean giant reportedly sent initial samples to Nvidia in September, although various reports placed this during November of last year. The following month saw SK Hynix reportedly delay HBM4 chip production. Sources familiar with the matter said that while Hynix had planned to gradually increase HBM4 production in the first half of 2026, the increased chance of a delay in the Nvidia Rubin launch schedule made it re-focus on HBM3E production for existing Blackwell chips. Some of that delay was down to Nvidia allegedly changing its design specs to request speeds of over 11 Gb/s per pin.

According to Hankyung, Samsung has also passed AMD’s own HBM4 specifications.

Samsung success

Samsung’s success was put down to a novel way in working with dynamic random access memory (DRAM). Despite some early weak yields, the Korean firm persisted with a sixth-generation 10‑nanometer-class (1c) DRAM process. For HBM4E stacks, this 1c DRAM was combined with an in-house 4‑nanometer logic die.

This is in contrast to its rivals, who instead stack DRAM dies on a passive interposer, a thin silicon layer that routes signals between the memory and GPU. In place of the logic die, the GPU or accelerator handles the logic within the setup.

According to a Digitime report from November, Samsung expected to allocate up to 150,000 wafers per month of 1c DRAM capacity to HBM4 by the end of this year, including 80,000 wafers of new 1c capacity and converted portions of older DRAM lines.

Samsung will apparently supply half of the Small Outline Compression Attached Memory Module 2 (SOCAMM2) that Nvidia requires for Rubin mass production. If chipmakers like Nvidia and AMD become the focus of DRAM production in place of servers, then this may impact original equipment manufacturers like HPE, as predicted by financial analysts toward the end of last year.

The alleged HBM4 success will likely give Samsung another win on top of its gilded DRAM status, with the company poised to reclaim its status from Hynix as the world’s leading DRAM vendor in light of AI driving demand for memory in the hardware stack.

This hasn’t stopped Samsung Electronics chairman warning the company on complacency. According to new internal training material revealed this week, chair Lee Jae-yong is keen Samsung execs “die to live,” based on a Chinese proverb used across Asia translating as, “only if you are determined to die, will you survive.”

It seems Samsung may have enforced such absolute commitment to succeed with its HBM4 efforts.