Stanford team proposes Joules-to-GDP: measuring AI's economic value per joule of energy
niloofar_mire · x · 2026-11-06
A Stanford team (co-led with John Hennessy and members of the Hazy Research group) introduces Joules-to-GDP (J2G), a framework measuring how efficiently AI converts energy into economic value across tasks, models, and hardware.
- Analyzed tasks across 44 occupations in finance, engineering, law, and admin work using OpenAI's GDPval
- Key finding: highest-value tasks aren't the most energy-efficient — CAD design earns 10% more per task than building websites, but websites earn 27% more value per joule
- Core insight: picking the right model on the right accelerator can yield several times more value per joule for the same task
The authors argue this metric will matter increasingly as AI serving scales across the economy.
More from Infra
- Nvidia chips pile up in Hong Kong as China withholds import approval — pstAsiatech · 2026-11-06
- Chart: UK IT investment at 0.66% of GDP vs US 2.21%, missing the AI capex boom — KeeganMcB · 2026-11-06
- Kimi K3.1 Pro gets Day-0 TPU support: serving a 2.8T-param MoE with llm-d on TPUs and GPUs — vllm_project · 2026-11-06
- Level1Techs: CUDA porting lags while AMD's R9700 turns out surprisingly fast — Thrumpwart · 2026-11-06
- Satirical pitch: cool data centers with fluorinert and market them as 'water-free' — typedfemale · 2026-11-06
- US stocks lost $3.8T in a day; chipmakers took $1.15T of the hit — CharlieMuirhead · 2026-11-06