source: simon willison: the memory shortage is causing a repricing of consumer electronics
level: business
the global memory chip industry is dominated by just three manufacturers, and their production capacity is fixed by the number of wafers they can process. this capacity is split among ddr for desktops and servers, lpddr for phones and low-power devices, and hbm for gpus. until recently, hbm used only 2% of wafer allocation, but the ai boom is pushing that to 20% by the end of 2026. critically, a gigabyte of hbm consumes over three times the wafer capacity of a gigabyte of ddr or lpddr.
memory makers have learned from past industry consolidation to avoid over-provisioning fabrication capacity, so they prioritize high-margin hbm production. this shift directly reduces the supply of consumer device ram. the impact is already visible in the sub-$100 smartphone market, which is vital in regions like africa and south asia. as hbm demand grows, cheaper devices will face component shortages and price hikes.
the trend means that ai data center expansion is indirectly raising costs for everyday electronics. with wafer capacity constrained, manufacturers cannot quickly ramp up total output, so the reallocation toward hbm leaves less room for other memory types. this supply squeeze is expected to persist for several years, reshaping pricing across laptops, phones, and other consumer goods that rely on affordable memory.
why it matters: ai infrastructure growth is directly raising hardware costs for consumers and data science practitioners, affecting device affordability and edge deployment budgets.
source: simon willison: the memory shortage is causing a repricing of consumer electronics