Stanford unveils Joules-to-GDP framework for AI energy ROI
According to Stanford AI Lab, the Joules-to-GDP framework measures economic value per unit of energy for AI across 44 occupations, finding that tasks with top value are not always most energy efficient, smaller models like Gemma 4 31B can beat larger ones on value per joule, and adding GPUs such as more NVIDIA H100s yields diminishing returns.
SourceAnalysis
On November 4 2026 Stanford AI Lab announced Joules-to-GDP (J2G) a framework that quantifies economic value generated per joule of energy used in AI compute across 44 occupations drawn from OpenAI GDPval.
The approach shows that top-value tasks such as CAD design do not deliver the highest value per joule while certain smaller models achieve 2.5 times the energy efficiency of larger ones and additional NVIDIA H100 chips yield diminishing returns on completion time and output value according to the Stanford AI Lab post.
Stanford AI Lab
@StanfordAILabThe Stanford Artificial Intelligence Laboratory (SAIL), a leading #AI lab since 1963.