DRAM contract prices have jumped 63% in the current quarter, hitting 15-year highs as demand from on-device AI accelerators surges. AMD's Gorgon Halo and RTX Spark chips, designed for local agentic computing, require up to 192GB of DRAM — making bill-of-materials costs prohibitive for price-sensitive edge deployments.
The shortage inverts the economics of edge AI at consumer price points. OEMs and system integrators now face a dilemma: absorb margin pressure, raise end-user pricing, or defer local-compute feature launches until memory markets normalize. This benefits centralized cloud-inference models in the near term.