Samsung, SK hynix Vie for CXL AI Memory Lead

As artificial intelligence (AI) services surge, memory shortages and bottlenecks plague data centers. In response, South Korea's semiconductor leaders, Samsung Electronics and SK hynix, aim to claim dominance in the next-generation AI memory market with compute express link (CXL) at the OCP Global Summit 2026. They both champion 'CXL-based resource sharing,' a crucial technology to process the data explosion in AI inference while addressing the limited supply of high-bandwidth memory (HBM).

Samsung Electronics presents a 'detailed solution-centric strategy,' zeroing in on individual module advancements and data management software. Their focus lies on resolving the 'KV (Key Value) cache' shortage, which stores contextual information in lengthy AI conversations. Samsung proposes a solution that combines its CXL memory module 'CMM-D,' SSDs, and memory-proximate processing technology 'CXL-PNM.'

The plan aims to tackle structural bottlenecks in next-generation agent-type AI systems through dedicated management software, efficiently distributing and storing infrequently used data.

In contrast, SK hynix counters with a 'rack-level system optimization strategy,' addressing the entire data center, not just individual servers. They plan to launch a 'memory-centric AI rack system' that integrates next-generation HBM and CXL technologies. By pooling memory and sharing data at the rack level, the strategy intends to significantly enhance overall system performance with an optimized memory hierarchy tailored to specific AI computational data.

While commercialization may take time, CXL's adoption is accelerating as global tech giants like Meta and Marvell integrate it into actual data centers. As KV cache growth outpaces data compression technology, attention centers on the winning strategy: Samsung's software and module-based approach versus SK hynix's rack-level integrated system.