Stanford's Joules-to-GDP measures the economic value AI produces per joule of energy
rohanpaul_ai · x · 2026-11-06
Stanford, UC Berkeley and UT Austin researchers introduce Joules-to-GDP (J2G), a framework measuring net dollar value per joule of AI energy use, tested on 220 GDPval tasks across 44 occupations with 5 open models. Key findings: CAD design earns 10% more per task than web building but 27% less per joule; GLM 5.2 delivers the top net value ($72.27/task) while Gemma 4 31B hits 56% of it with 23% of the energy (2.5x NVJ); scaling one H100 to eight cuts task time only 22% while using 3.2x the energy; and against experts who themselves use AI, GLM 5.2's NVJ drops 70%. Code and dataset are open-sourced.
Related event: Stanford's Joules-to-GDP Framework Measures AI's Economic Return per Joule(10 posts)→
More from Infra
- Marin's open 535B-A23B MoE finishes main pretraining on 18T tokens, checkpoints soon — dlwh · 2026-11-06
- Buoyant Open-Sources Kinference, an Inference Control Plane for vLLM on Kubernetes — vllm_project · 2026-11-06
- vLLM ranks #2 most-engaged open-source repo in GitHub Octoverse, #8 by contributor growth — vllm_project · 2026-11-06
- Why AWS and cloud providers have clear incentive to build their own AI chips — DavidLinthicum · 2026-11-06
- Dev runs local AI on 128GB Framework Desktop, weighs buying a third as 192GB nears — gnukeith · 2026-11-06
- Dev reverse-engineers Wispr Flow, fine-tunes LFM3-1.7B for free local voice dictation — teortaxesTex · 2026-11-06