Data centers are increasingly adopting LPDDR5X memory, previously used in mobile devices, to manage power consumption and cooling constraints in AI-heavy environments. The new SOCAMM2 format enables this memory to be used in servers as removable modules, offering a modular alternative to traditional soldered implementations.
Micron began shipping 256 GB SOCAMM2 samples in March 2026. Compared to standard RDIMM, these modules consume up to 61% less power under load and up to 80% less in standby mode. They also occupy approximately one-third of the space of traditional modules and support speeds up to 9600 MT/s.
The Micron SOCAMM server lineup includes modules ranging from 48 GB to 256 GB. NVIDIA has also announced support for SOCAMM LPDDR5X memory for its Vera processor, targeting up to 1.5 TB of capacity per processor and peak bandwidth of 1.2 TB/s.
Performance testing shows that LPDDR5X-8533 provides a 13% increase in actual bandwidth compared to 6400 MT/s configurations. In AI data preparation tasks, performance improved by 11%, while processing time decreased by 10%. In analytics, increasing capacity from 512 GB to 1 TB accelerated specific workloads by 2–3 times.
