Samsung Electronics Co. is set to become Nvidia Corp.’s largest supplier of second-generation SOCAMM memory modules in 2026, capturing roughly half of the chip designer’s projected requirements for the specialized AI data center component.
The South Korean technology giant will deliver approximately 10 billion gigabytes of SOCAMM2 modules out of Nvidia’s total 20 billion gigabyte demand, according to people familiar with the matter. The contract represents a significant shift in competitive positioning, with Samsung displacing Micron Technology Inc., which had been first to market with the original SOCAMM standard earlier this year.
SOCAMM, or Small Outline Compression Attached Memory Module, functions as auxiliary memory positioned near CPUs in AI servers to coordinate processing tasks for graphics chips. The second-generation version offers roughly 50% greater capacity and 20% lower power consumption than its predecessor, which Nvidia reportedly shelved due to technical difficulties.
Samsung’s turnaround stems from stabilized production yields on its fifth-generation 10-nanometer DRAM technology, the primary component in SOCAMM2 modules. The required monthly output of 30,000 to 40,000 wafers represents just 5% of the company’s total DRAM manufacturing capacity.
SK Hynix Inc. is expected to fulfill most remaining orders, with Micron taking a reduced share compared to its earlier market position.
