AI-Driven HBM Demand Reverses 20 Years of Memory Price Declines
A study by computer science professor Daniel Lemire highlights that surging demand for High Bandwidth Memory (HBM) has pushed general memory prices per GB back to 2007-2008 nominal levels. This rapid price increase has effectively reversed nearly two decades of consistent cost declines in just a few months. The surge in memory prices impacts the broader hardware ecosystem, raising costs for consumer PCs, enterprise servers, and data centers. As AI infrastructure consumes massive amounts of wafer capacity, standard DRAM production is squeezed, forcing industries to either optimize memory usage or scale up manufacturing. Currently, DDR5 memory costs between $11.41 and $13.28 per GB, matching nominal DDR2 prices from 2008. This price pressure is exacerbated by a 3-to-1 wafer conversion ratio, meaning producing one HBM wafer consumes the capacity of three standard DDR5 wafers.
## BACKGROUND
High Bandwidth Memory (HBM) is a 3D-stacked DRAM technology designed to provide massive throughput for AI accelerators and high-performance computing. Because HBM requires complex manufacturing processes like through-silicon vias (TSVs), major manufacturers like Micron, Samsung, and SK Hynix are shifting production capacity away from standard DDR5 to meet AI demands.