DRAM crunch pushes AI system designers toward lower-memory architectures

The article says DRAM has become a constrained resource in the AI stack as manufacturers prioritize DDR5 and HBM for data centers. It describes supply as tight and says prices have risen to as much as three to four times the levels some teams paid a year earlier. The pressure is strongest for high-capacity modules tied to cloud infrastructure demand, while lower-capacity memory in the 1GB to 2GB range is described as comparatively stable. The market signal is clear: DRAM cost and availability are no longer just procurement issues; they are influencing system architecture. AI workloads with large external-memory footprints face more cost volatility, while edge designs and smaller models may become more attractive when they reduce dependency on external DRAM.