---

SK hynix has started mass production of a 192GB SOCAMM2 memory module, targeting next-generation AI servers and setting the stage for higher-performance, lower-power infrastructure.

The new module is built on the company’s 1cnm (sixth-generation 10nm-class) LPDDR5X DRAM process and is designed specifically for AI workloads, including large language model (LLM) training and inference. The development highlights how memory innovation is becoming a critical lever in overcoming AI system bottlenecks, particularly in power efficiency and bandwidth scaling.

AI-focused memory architecture

SOCAMM2 (Small Outline Compression Attached Memory Module 2) represents a shift from traditional server memory approaches by adapting low-power mobile DRAM for data center use. The module combines a compact form factor with scalability, while its compression connector improves signal integrity and simplifies replacement.

According to SK hynix, the 192GB SOCAMM2 delivers more than twice the bandwidth of conventional RDIMM modules while improving power efficiency by over 75%. This makes it particularly suitable for high-performance AI operations, where memory throughput and energy consumption are key constraints.

The product has been optimized for the NVIDIA Vera Rubin platform, underscoring the growing alignment between memory vendors and AI accelerator ecosystems. As AI models grow in size — often reaching hundreds of billions of parameters — memory performance is increasingly central to overall system efficiency.

Tackling AI bottlenecks

SK hynix expects SOCAMM2 to address critical bottlenecks in AI infrastructure, especially during LLM training and inference. By improving both bandwidth and efficiency, the module aims to accelerate processing speeds across entire systems.

The company also pointed to a broader shift in the AI market, where the focus is moving from inference toward training workloads. This transition is driving demand for memory solutions that can handle intensive data processing while maintaining low power consumption — an area where LPDDR-based modules could offer a competitive edge.

To support global cloud service providers, SK hynix said it has already stabilized its mass production system and expanded its supply portfolio for AI-focused memory.

“By supplying the 192GB SOCAMM2, SK hynix has established a new standard for AI memory performance,” said Justin Kim, President & Head of AI Infra (CMO, Chief Marketing Officer) at SK hynix. “We will solidify our position as the most trusted AI memory solution provider, through close collaboration with our global AI customers.”

With hyperscalers and AI chip vendors racing to scale infrastructure, innovations like SOCAMM2 underline how memory is becoming a strategic differentiator in the AI era.

---

---