[News] NVIDIA Rumored to Ditch SOCAMM1 for Next-Gen SOCAMM2, Opening Door for Samsung, SK hynix

While Micron reportedly passed NVIDIA’s verification and became the first of the big three memory makers to ship SOCAMM (Small Outline Compression Attached Memory Module), the U.S. AI chip giant may revise its adoption plan for this new low-power DRAM (LPDDR) server memory module—a move that could benefit Samsung and SK hynix, according to etnews.

The report indicates that NVIDIA, in a surprising move, has scrapped SOCAMM1 in favor of SOCAMM2 and has begun testing with the three major memory makers. According to etnews, taking the quality evaluation period into account, SOCAMM2 mass production is expected to begin in early 2026.

Industry sources cited by etnews say NVIDIA initially planned to launch SOCAMM1 this year, but technical issues forced the project to be halted twice, blocking large-scale orders. While Micron led in SOCAMM1, Samsung and SK hynix now may have a chance to catch up.

Rumors of hurdles in NVIDIA’s SOCAMM rollout have been swirling for some time. ZDNet reported in May that the company postponed the launch of SOCAMM alongside the GB300, and now the module is expected to make its debut with NVIDIA’s Rubin, slated for 2026.

Samsung Flags Progress on SOCAMM2

As The Korea Herald reported, SOCAMM vertically stacks advanced LPDDR5X chips to enhance data processing performance while significantly reducing energy consumption. In March, Micron announced it is the world’s first and only memory company shipping both HBM3E and SOCAMM products for AI servers in the data center.

Etnews notes that SOCAMM2 keeps the same 694 I/O pins as SOCAMM1 but increases data transfer speeds to up to 9600MT/s, surpassing SOCAMM1’s 8533MT/s. NVIDIA is reportedly exploring next-generation LPDDR6 for the module as well.

Notably, citing Samsung’s 2025 Sustainability Report, Hansbiz reveals that even before SOCAMM’s first generation hits full-scale production, Samsung is already developing SOCAMM2.

As per Samsung, the company has developed SOCAMM2, a server module optimized for AI computing based on LPDDR, and provides the best high-performance, low power solutions for the AI era through SSD products based on PCIe 5th generation with a 5nm controller.

Read more

• [News] SOCAMM Ignites a New Memory War After HBM — Samsung and SK hynix Join the Race
• [News] NVIDIA Reportedly to Procure Up to 800,000 SOCAMM Modules in 2025 for AI Product

(Photo credit: Micron)