Micron memory for AI (obviously) Having exited the consumer market, Micron is currently concentrating on improving the appeal of its AI product lineup. In

米Micron Technologyは3月3日に容量256GBのLPDRAM SOCAMM2を発表した。

Micron and General Motors announced a strategic customer agreement covering long-term supply of LPDRAM, NOR and UFS NAND for future vehicle platforms. The company statement links the deal to Micron’s $2 billion

## Micron 256GB SOCAMM2 Memory Samples for AI Servers Micron developed the first 256GB SOCAMM2 memory samples which they sent to AI server markets. Micron Technology revealed its 256GB SOCAMM2 memory module through customer sample shipments on March 3 2026. The release introduces the maximum available capacity for LPDRAM modules. The hardware operates through the industry’s first monolithic 32Gb LPDDR5X design which engineers created to handle the enormous memory requirements of AI training and inference and high performance computing (HPC). The 256GB SOCAMM2 memory module delivers solutions to the existing "memory bottleneck" problem which occurs in contemporary AI workloads that require high bandwidth and low latency due to their extensive model parameters and large context windows. Micron achieved higher density through the usage of 32Gb monolithic die technology without losing the power saving benefits that LPDRAM technology delivers. The memory has been optimized by Micron to work with new AI infrastructure systems through their partnership with NVIDIA. Ian Finder from NVIDIA Data Center CPU Product management said that SOCAMM2 architecture provides essential capacity and bandwidth requirements for upcoming AI processors. The SOCAMM2 form factor provides multiple benefits when compared to existing RDIMM (Registered DIMM) systems used in servers. * Power and Space Efficiency: The SOCAMM2 modules consume only one third of the power required by equivalent RDIMMs while occupying only one third of the physical footprint. This allows for significantly higher rack density in data centers. * The 256GB SOCAMM2 system reduces "time to first token" for real time Large Language Model (LLM) inference with long context windows by more than 2.3 times when compared to current solutions. * Performance Per Watt: The modules offer triple the power efficiency of typical memory products in CPU applications which function independently. * The 8 channel CPU now enables support for 2TB of LPDRAM which allows users to run advanced inference tasks together with persistent key value (KV) caching. The SOCAMM2 system uses a modular structure that enables simple server maintenance while maintaining essential performance requirements for liquid cooled systems. Data center operators can use this solution which permits memory capacity growth to match new AI demands. Micron helps establish the JEDEC standard for low power modular memory through its participation in the specification process. Micron provides data center solutions which include 8GB components and the new 256GB SOCAMM2 modules to system designers who require AI and general purpose computing components.

Micron's new 256GB LPDRAM module cuts power use and size, boosting AI memory efficiency and speed. A step forward in data center tech. ⚡💾

Micronが、サーバ／データセンター用の「LPDRAM SOCAMM2」において256GBモデルをサンプル出荷を開始した。省電力性と高速性を両立したLPDDR5Xメモリをより大容量に使えるようにすることで、メモリ回りのTCOを改善する狙いがある。

Micron memory for AI (obviously) Having exited the consumer market, Micron is currently concentrating on improving the appeal of its AI product lineup. In

Micron presenta la primera memoria LPDRAM SOCAMM2 de 256 GB del mundo diseñada para ser usada en centros de datos de IA.

JEDEC Solid State Technology Association, the global leader in standards development for the microelectronics industry, today announced it is nearing completion of JESD328: LPDDR5/5X Small Outline Compression Attached Memory Module (SOCAMM2) Common Standard, an upcoming standard for low-profile LPDRAM modules developed specifically for data center AI applications. When published, JESD328 is designed to provide a memory platform that delivers modular, low-power, high-bandwidth memory for AI CPU servers and accelerated computing platforms. JESD328 will define a compact, serviceable LPDDR5X-based module, Small Outline CAMM2 (SOCAMM2). JESD328 SOCAMM2 is set to introduce an optimized, compact form factor for servers with a mechanical outline designed for high-density data center chassis and board layouts, with scalability to provide a path to support the large module capacities required by AI training and inference servers. Read full story

JEDEC Solid State Technology Association, the global leader in standards development for the microelectronics industry, today announced it is nearing completion of JESD328: LPDDR5/5X Small Outline Compression Attached Memory Module (SOCAMM2) Common Standard, an upcoming standard for low-profile LPDRAM modules developed specifically for data center AI applications. When published, JESD328 is designed to provide a memory platform that delivers modular, low-power, high-bandwidth memory for AI CPU servers and accelerated computing platforms. JESD328 will define a compact, serviceable LPDDR5X-based module, Small Outline CAMM2 (SOCAMM2). JESD328 SOCAMM2 is set to introduce an optimized, compact form factor for servers with a mechanical outline designed for high-density data center chassis and board layouts, with scalability to provide a path to support the large module capacities required by AI training and inference servers. Read full story

JEDEC Solid State Technology Association, the global leader in standards development for the microelectronics industry, today announced it is nearing completion of JESD328: LPDDR5/5X Small Outline Compression Attached Memory Module (SOCAMM2) Common Standard, an upcoming standard for low-profile LPDRAM modules developed specifically for data center AI applications. When published, JESD328 is designed to provide a memory platform that delivers modular, low-power, high-bandwidth memory for AI CPU servers and accelerated computing platforms. JESD328 will define a compact, serviceable LPDDR5X-based module, Small Outline CAMM2 (SOCAMM2). JESD328 SOCAMM2 is set to introduce an optimized, compact form factor for servers with a mechanical outline designed for high-density data center chassis and board layouts, with scalability to provide a path to support the large module capacities required by AI training and inference servers. Read full story

JEDEC Solid State Technology Association, the global leader in standards development for the microelectronics industry, today announced it is nearing completion of JESD328: LPDDR5/5X Small Outline Compression Attached Memory Module (SOCAMM2) Common Standard, an upcoming standard for low-profile LPDRAM modules developed specifically for data center AI applications. When published, JESD328 is designed to provide a memory platform that delivers modular, low-power, high-bandwidth memory for AI CPU servers and accelerated computing platforms. JESD328 will define a compact, serviceable LPDDR5X-based module, Small Outline CAMM2 (SOCAMM2). JESD328 SOCAMM2 is set to introduce an optimized, compact form factor for servers with a mechanical outline designed for high-density data center chassis and board layouts, with scalability to provide a path to support the large module capacities required by AI training and inference servers. Read full story

JEDEC Solid State Technology Association, the global leader in standards development for the microelectronics industry, today announced it is nearing completion of JESD328: LPDDR5/5X Small Outline Compression Attached Memory Module (SOCAMM2) Common Standard, an upcoming standard for low-profile LPDRAM modules developed specifically for data center AI applications. When published, JESD328 is designed to provide a memory platform that delivers modular, low-power, high-bandwidth memory for AI CPU servers and accelerated computing platforms. JESD328 will define a compact, serviceable LPDDR5X-based module, Small Outline CAMM2 (SOCAMM2). JESD328 SOCAMM2 is set to introduce an optimized, compact form factor for servers with a mechanical outline designed for high-density data center chassis and board layouts, with scalability to provide a path to support the large module capacities required by AI training and inference servers. Read full story