Source: The Register
Google’s technique for reducing AI model memory footprint is being misread as a demand killer for RAM manufacturers like Micron and SK Hynix, when the real issue is that current memory prices are simply unsustainable for the broader market—vendors are already struggling to quote accurately. The efficiency gains matter far less than the economics: if AI training and inference remain cost-prohibitive at current DRAM pricing, adoption stalls regardless of algorithmic improvements, which means the memory industry’s problem is overcapacity and margin compression, not technological disruption. This is a pricing story masquerading as an innovation story.