NuAxiSemiconductor trade intelligence

[News] Samsung Showcases CXL Memory Expansion With AMD at OCP 2026, Expands AI Memory Push Beyond HBM

What happened

According to ZDNet, Samsung Electronics teamed up with AMD to demonstrate a Compute Express Link (CXL)-based memory expansion solution at the OCP Global Summit 2026; featuring Samsung's CMM-D-based memory pooling technology, the solution enables multiple systems to share memory resources more efficiently and flexibly expand capacity. As noted by NewsPim, Samsung sees CXL-based memory expansion as a way to overcome the capacity limitations of conventional systems, particularly as AI agents drive higher memory requirements for inference. The demonstration follows a memorandum of understanding signed by Samsung and AMD in March to expand cooperation on next-generation AI memory and computing technologies; per The Korea Times, during her October 7 visit to South Korea, AMD CEO Lisa Su also discussed potential collaboration with Samsung in memory and foundry technologies (a discussion, not a confirmed collaboration). At the event, Samsung outlined plans to expand its AI memory and storage portfolio, highlighting next-generation technologies including HBM5, custom HBM, zHBM, and processing-in-memory (PIM), alongside CMM-D and CXL-based solutions, while advancing next-generation SSD technologies including quad-level cell (QLC) NAND and flexible data placement (FDP), and showcasing KV cache optimization and Direct Storage technologies. Through its Samsung MAX Lab, customers can evaluate emerging technologies including PIM, CMM-D, and next-generation SSDs under real-world AI workloads before commercialization. Samsung also displayed a Groq LPU wafer and NVIDIA's Groq 3 LPX system, with the chips mass-produced using Samsung Foundry's 4nm process, and a wafer manufactured using its 2nm gate-all-around (GAA) process.