A deal between Samsung and NVIDIA was recently announced, namely the supply of 50% of SOCAMM2 memory modules in 2026.
According to information from the South Korean publication Hankyung, Samsung and NVIDIA have signed a contract for the supply of 50% of all SOCAMM2 memory [Samsung’s proprietary next-generation high-bandwidth memory] to be produced in 2026!
This figure is, one could say, terrifying. To buy up 50% means that NVIDIA has an enormous outlet for this memory, and in huge quantities.
Against the backdrop of ever-increasing DRAM prices, the cessation of consumer RAM production, and a greater shift towards server memory, it is clear that this memory is destined for AI.
Including for NVIDIA’s superchip which they plan to release, Vera Rubin. It consists of two Rubin processors and a central Vera chip, which support SOCAMM2 DDR5 modules.
These processors will replace the LPDDR5X ECC memory used in current models.
This was created to meet the need for higher memory density, increased clock speeds, and to enable further improvements.
According to estimates by many professionals, the DRAM shortage could last until 2027


