Joules-to-GDP: Measuring the economic value of AI per unit of energy

Researchers from Stanford, UC Berkeley, and UT Austin released "Joules-to-GDP," a paper examining the economic output value and expert time saved by open-weight AI models per unit of energy consumed. The study found that the model producing the most net value per task is not the most energy-efficient, and that hardware choices, additional GPUs, and redundant retries significantly impact energy returns. Open article →