One H200-hour of energy yields $424 of net value via GLM 5.2, with 13x spread across jobs
rohanpaul_ai · x · 2026-11-05
A "Joules-to-GDP" study shared by Rohan Paul quantifies AI's economic return per unit of energy: one NVIDIA H200 at 700W for an hour (2.52 MJ, equivalent to 23 hours of laptop use) is enough for GLM 5.2 to produce about $424 of net value and save 9.63 hours of expert work.
Using two metrics — NVJ (net dollar value per joule) and TSJ (expert time saved per joule) — the study covers 220 professional tasks, 13 occupation categories, 5 models, and 4 accelerators:
- Returns vary up to 13x across occupations: Computer & Math leads at 524 USD/MJ vs. Office & Admin at 40 USD/MJ; for time saved, Architecture & Engineering tops at 7.89 h/MJ vs. Legal at 1.50 h/MJ (5.26x spread)
- The model producing the most total value isn't the most energy-efficient: GLM 5.2 leads in total output but sits off the efficient frontier, while GPT-OSS and Gemma models are closer to it.
Related event: Stanford's Joules-to-GDP Framework Quantifies AI Energy-Economic Efficiency(5 posts)→
More from AGI Musings
- AI2 exec Bindu Reddy: AI's reputation is unfairly bad and the industry should celebrate it — bindureddy · 2026-11-05
- Many grant solicitations now flagged as 100% AI-generated by detector Pangram — analisereal · 2026-11-05
- Why AI research keeps looking under the lamppost: stale problem frames inherited from Kant — fkasummer · 2026-11-05
- Michael Nielsen on why scientists can't tell an anomaly from a revolution — michael_nielsen · 2026-11-05
- Anthropic's S-1 Paints an Autonomous Labor Future—and an Oversight Problem — asusarla · 2026-11-05
- New paper uses 288M job postings to measure where AI raises the value of work it can't do — robseamans · 2026-11-05