Tag
The article argues that AI model efficiency is improving so rapidly that the hardware needed for a fixed level of intelligence halves roughly every 3 months, making renting frontier intelligence or owning trailing-edge hardware more economical than buying new hardware.