Samsung and AMD unveil CXL memory expansion system at OCP 2026
Samsung Electronics and Advanced Micro Devices used OCP Global Summit 2026 to show a CXL-based memory expansion design built around CMM-D memory pooling, a setup aimed at easing the capacity limits that appear as AI inference workloads grow. The companies presented the system as a way for multiple machines to share a common memory pool instead of relying only on memory installed on each server motherboard. Samsung tied the pitch to a broader AI memory strategy that also included HBM5, custom HBM, zHBM, processing-in-memory, and newer SSD technologies such as QLC NAND and Flexible Data Placement. The company also highlighted KV cache optimization, Direct Storage, and its Samsung MAX Lab program, which lets customers test technologies under real AI workloads before commercialization. Beyond memory, Samsung displayed Groq LPU wafers and an Nvidia Groq 3 LPX system, with reports saying the chips are being mass-produced on Samsung Foundry’s 4 nm process, and it also showed a wafer made with its 2 nm GAA process. Even so, the OCP presentation did not include performance figures, customer names, or a commercialization timeline.








