WS Investor
04 Mar 2026, 17:39
Micron Technology, Inc. (NASDAQ: MU) announced it has begun shipping customer samples of the industry’s first **256GB LPDRAM SOCAMM2 memory module**, designed for next-generation AI data center infrastructure.
The new module is powered by the industry’s first **monolithic 32Gb LPDDR5X die** and delivers major improvements in energy efficiency and computing performance for AI and high-performance computing workloads. According to Micron, the module uses **one-third of the power and one-third of the physical space** compared with traditional DDR5 RDIMM server memory.
The 256GB SOCAMM2 also increases memory capacity by **33% compared with previous 192GB modules**, enabling up to **2TB of LPDRAM per 8-channel server CPU**. This higher capacity supports large AI models with expanded context windows and complex inference workloads.
In AI applications, the module can deliver **2.3× faster “time to first token”** for long-context large language model inference, while offering **three times better performance per watt** in CPU-based high-performance computing workloads.
Micron said it is collaborating with **NVIDIA** to co-design advanced memory architectures for AI infrastructure, as demand for higher capacity, bandwidth efficiency, and lower power consumption continues to rise in modern data centers.
The new module is powered by the industry’s first **monolithic 32Gb LPDDR5X die** and delivers major improvements in energy efficiency and computing performance for AI and high-performance computing workloads. According to Micron, the module uses **one-third of the power and one-third of the physical space** compared with traditional DDR5 RDIMM server memory.
The 256GB SOCAMM2 also increases memory capacity by **33% compared with previous 192GB modules**, enabling up to **2TB of LPDRAM per 8-channel server CPU**. This higher capacity supports large AI models with expanded context windows and complex inference workloads.
In AI applications, the module can deliver **2.3× faster “time to first token”** for long-context large language model inference, while offering **three times better performance per watt** in CPU-based high-performance computing workloads.
Micron said it is collaborating with **NVIDIA** to co-design advanced memory architectures for AI infrastructure, as demand for higher capacity, bandwidth efficiency, and lower power consumption continues to rise in modern data centers.