SK hynix, Solidigm and others pitch AI memory beyond HBM at OCP Global Summit - Crypto Briefing

The HBM bandwidth ceiling is now the primary bottleneck for AI compute, prompting SK hynix and Solidigm to pivot toward CXL-based memory pooling and near-memory computing architectures. This strategic shift fundamentally restructures the storage value chain, compelling upstream DRAM manufacturers to prioritize system-level interconnect optimization over mere layer stacking, while intensifying the scramble for advanced packaging capacity. From a geopolitical standpoint, reliance on specific foundry nodes for these novel architectures forces hyperscalers to recalibrate supply chain redundancy costs to mitigate export control risks. As Micron and Samsung prepare counter-strategies, the competitive battleground will shift from single-die performance to cluster-wide energy efficiency. Within the next 18 months, high-bandwidth, poolable memory architectures are poised to displace traditional HBM as the data center standard. Storage vendors failing to transition from component suppliers to system architects face imminent marginalization in the AI infrastructure market.