Samsung Unveils Three-Pillar AI Memory Strategy at OCP Global Summit 2026
Samsung Electronics detailed speed, scalability and persistence plans and showed CXL memory pooling with AMD and Nvidia Groq 3 wafers in San Jose.
Samsung Electronics announced a three-pillar AI memory strategy focused on speed, scalability and persistence at OCP Global Summit 2026 in San Jose, California, on December 12 local time, according to inews24.com. The company said the strategy covers faster data processing, expanded memory capacity and persistent memory capabilities for AI systems.
As part of the speed pillar, Samsung said it is developing HBM5 and custom HBM. The company displayed an HBM5 model alongside next-generation semiconductor wafers, zHBM and PIM technology at the event.
For scalability, Samsung and AMD jointly validated CXL-based memory expansion technology and demonstrated memory pooling. Samsung also showed its CMM-D CXL memory module, which dynamically allocates memory capacity based on usage. Samsung's tests showed that dynamic memory allocation cut time to first response by up to 38% and raised effective throughput by up to 3.3x versus fixed allocation.
Samsung Foundry exhibited wafers for Nvidia's Groq 3 LPU AI inference chip and the Groq 3 LPX system, along with a Samsung 2nm process wafer.
Earlier cooperation
In March, Samsung and AMD signed an MOU to cooperate on next-generation AI memory and computing technology, including HBM4 for AMD's MI455X AI accelerator and DDR5 for server CPUs. On December 7, AMD CEO Lisa Su visited Samsung's Seocho office and met vice chairman Jun Young-hyun for over three hours to discuss expanding semiconductor cooperation.
Quick answers
What is Samsung's three-pillar AI memory strategy?
Samsung Electronics said the strategy focuses on speed, scalability and persistence. It includes HBM5 and custom HBM development, CXL-based memory expansion with AMD, and the CMM-D CXL memory module.
What did Samsung and AMD demonstrate at OCP Global Summit 2026?
Samsung and AMD jointly validated CXL-based memory expansion technology and demonstrated memory pooling. Samsung's tests showed dynamic memory allocation cut time to first response by up to 38% and raised effective throughput by up to 3.3x versus fixed allocation.
What Nvidia products were shown at Samsung's exhibit?
Samsung Foundry exhibited wafers for Nvidia's Groq 3 LPU AI inference chip and the Groq 3 LPX system, along with a Samsung 2nm process wafer.