Micron launches 192 GB socamm2 for low-power AI data centers

Micron Technology announced customer sampling of a 192 GB socamm2, a small outline compression attached memory module designed to broaden adoption of low-power DRAM in Artificial Intelligence data centers. The socamm2 builds on Micron’s earlier low-power DRAM socamm and delivers 50 percent more capacity in the same compact footprint. Micron positions the added capacity as a direct enabler for performance improvements in real-time inference workloads.

The company says the additional capacity can cut time to first token, or TTFT, by more than 80 percent for real-time inference. The 192 GB socamm2 uses Micron’s 1-gamma DRAM process technology and delivers greater than 20 percent improvement in power efficiency, enabling tighter power design optimization across large data center clusters. Those efficiency gains become more significant at scale, where full-rack installations can include more than 40 terabytes of CPU-attached low-power DRAM main memory.

Micron highlights the socamm2’s modular design as improving serviceability and providing a pathway for future capacity expansion in data center deployments. The announcement focuses on customer sampling as the next step toward broader market availability and frames the product as a response to the industry shift toward more energy-efficient infrastructure to support growth in Artificial Intelligence workloads.