SEOUL, South Korea / RankWire.AI / – Samsung Electronics has introduced new approaches in memory and storage technology aimed at enhancing the capabilities of artificial intelligence systems. The company shared these plans during the 2026 OCP Global Summit in San Jose, California, held from October 12 to 15, with a focus on boosting performance through high-performance memory, expanding memory via Compute Express Link, and next-generation solid-state drives. Additionally, Samsung announced a joint CXL demonstration in collaboration with AMD. These initiatives are designed to address existing capacity limitations that hinder current AI server performance.
Samsung articulated its strategy around three primary objectives: increasing speed, enabling memory scalability, and ensuring data persistence. To improve speed, the company showcased customized high-bandwidth memory in conjunction with HBM5, the eighth generation of this technology. For scalability, it promoted CXL as a method to extend memory capacity beyond what is installed in a single server. Regarding data persistence, Samsung introduced advanced SSD-based storage and management solutions, along with storage software designed to optimize data processing efficiency within AI systems.
According to Samsung, the rising deployment of AI agents necessitates increased memory during inference, as larger models generate more data that must remain accessible to processors. Conventional systems restrict the amount of memory each processor can access, but Samsung presented CXL technology as a viable solution to surpass these limitations. The company also highlighted key-value cache optimization, which allows AI models to reuse previously processed data instead of performing redundant computations.
Partnership with AMD Expands to CXL Memory Expansion
In a collaborative effort, Samsung and AMD will demonstrate a CXL-based solution that increases memory capacity in AI systems, enabling processors to share memory resources instead of relying solely on local server memory. Samsung stated that engineers have validated this solution in real system environments. This demonstration is an extension of the memorandum of understanding signed between the two companies in March 2026, which focuses on next-generation AI memory and computing. On October 7, AMD CEO Lisa Su visited Korea and met with Jeon Young-hyun, Vice Chairman of Samsung’s Device Solutions division.
Another aspect of Samsung’s showcase was memory pooling, exemplified by the demonstration of CMM-D, a CXL-based DRAM module that connects multiple memory units into a single shared pool, allowing systems to draw additional capacity as needed. The company also exhibited processing-in-memory technology, which performs certain calculations within memory chips to reduce data movement and energy consumption. Their exhibits included 2-nanometer gate-all-around wafers, Groq LPU wafers, and Nvidia’s Groq 3 LPX system.
Early Co-Design Efforts Inform Samsung’s AI Memory Development Roadmap
Samsung’s Corporate Vice President and Head of the Application Engineering Team, Shim Ho-jun, outlined the company’s plans during an executive session, emphasizing that the increasing size of AI models and heightened inference demands make expanding memory capacity and bandwidth critical for system performance. Shim also highlighted that Samsung is intensifying co-design and technical validation collaborations with global partners from the earliest stages of development, aiming to more closely align its memory and storage products with customer system designs.
Samsung MAX Lab offers a platform for clients to evaluate memory and storage solutions under real-world AI workloads, providing tailored co-design services for specific system configurations. The lab enables testing of pre-commercial technologies, including PIM and next-generation SSDs, before their market release. As a Diamond Sponsor, Samsung participated in the summit, following its appearances in 2026 at Nvidia GTC, COMPUTEX, and the Future of Memory and Storage conference, as part of its expanding presence in AI memory innovations.