Micron has highlighted that its low-power DRAM (LPDRAM), originally developed for smartphones and other battery-powered devices, uses roughly a third of the power that standard DDR5 memory consumes when deployed in AI data centre workloads. The company is positioning this efficiency as a selling point as data centres scramble to curb the enormous energy demands created by the AI boom, with memory increasingly recognised as a significant contributor to overall power draw in AI server racks.

However, the article's author questions whether this development is as positive as it sounds, suggesting that efficiency gains of this kind often do not reduce overall energy consumption but instead enable operators to pack in more computing hardware, a pattern sometimes described as Jevons paradox. In other words, more efficient memory could simply accelerate the scale and pace of AI infrastructure build-outs rather than easing the industry's total power appetite.