DigiTimes reports that CXL is attracting more attention as systems need additional memory capacity beyond what can be attached directly to processors and accelerators. The interconnect can provide a path to expand and pool memory, particularly for capacity-heavy workloads. The report also says engineers from OpenAI and Intel see clear limits to using CXL as a replacement for high-bandwidth memory. HBM remains critical for workloads whose performance is constrained by bandwidth rather than capacity. The distinction matters for the memory market: CXL could increase demand for expandable DRAM tiers while HBM continues to serve the fastest accelerator-attached workloads.
AI Infrastructure · Sep 19, 2026
CXL Can Expand System Memory, but HBM Remains Essential for Bandwidth-Intensive AI
Growing memory needs are increasing interest in CXL for adding capacity beyond local memory, according to DigiTimes. Engineers at OpenAI and Intel emphasize that CXL does not replace HBM where AI workloads depend on maximum bandwidth.
Price impact: 2Direction: upSource: DigiTimes Daily
OpenAIIntelCXLDRAMHBMAI infrastructure
Original sourceBack to news archive