Samsung Showcases Next-Gen Memory and Storage at OCP Global Summit, Demos CXL Solution with AMD
On October 12, Samsung Electronics showcased its next-generation memory and storage technologies at the 2026 OCP Global Summit, covering memory expansion, compute speed, and data storage. The company also demonstrated a CXL-based memory expansion solution jointly with AMD.
As AI models grow larger and AI agents take on more complex tasks, the market demands for greater memory capacity, faster data processing, and more efficient storage continue to rise. One of the highlights was the CXL-based memory expansion solution demonstrated with AMD. CXL technology allows processors and other computing components to share memory resources, enabling AI systems to utilize additional memory capacity without relying entirely on memory installed inside a single server. The solution uses Samsung's CMM-D memory pooling technology, which allows multiple systems to share memory more flexibly and helps handle increasing workloads during AI inference.
In the compute speed area, Samsung showcased customized high-bandwidth memory (HBM), HBM5, and compute-in-memory technology, which performs certain calculations inside the memory chip to reduce data transfer and improve energy efficiency. The company also exhibited 2nm gate-all-around wafers, Groq language processing unit wafers, and NVIDIA's Groq 3 LPX system, highlighting its capabilities in memory and advanced chip manufacturing.
For data storage and management, Samsung introduced enterprise solid-state drives and technologies designed to optimize information storage and retrieval in AI inference, including key-value cache optimization technology that helps AI models reuse previously processed information for more efficient response generation.
Samsung also emphasized its commitment to working with customers to optimize the entire AI system, rather than focusing solely on individual semiconductor products. Its Samsung MAX Lab allows customers to test memory and storage technologies against real-world AI workloads before commercialization.
