Researchers Build Backdoor Method to Rank AI Model Energy Use as Big Three Stay Silent
relianceschool · reddit · 2026-09-29
AI's rise is driving a historic surge in electricity demand, but data center consumption ultimately comes down to millions of individual queries — and models vary wildly in efficiency, from gas guzzlers to Priuses.
None of the big three proprietary developers — Anthropic, OpenAI, and Google — disclose how much energy their models consume per query, leaving users with no way to compare.
The Sustainable AI Group, a research and advisory firm, developed a backdoor estimation method to fill the gap and on Tuesday released an interactive dashboard ranking AI models by energy intensity.
More from Infra
- NVIDIA launches Open Secure AI Alliance for open-source AI safety tools — perplexity_ai · 2026-09-29
- AMD's 256-core EPYC packs 16-channel DDR5-12800, hitting 91% of RTX 5090 memory bandwidth — Dany0 · 2026-09-29
- Dev Seeks Local LLM Stack to Reverse Old Game Client, Hits Safety Guardrails — danielfrances · 2026-09-29
- EU opens 12-week consultation on minimum performance standards for data centres — tekbog · 2026-09-29
- Will Nvidia Vera Rubin Actually Speed Up LLM Pre-Training? And Are 10T+ Models Next? — Witty_County5128 · 2026-09-29
- Merge Gateway adds self-hosted model support for unified enterprise AI governance — shensi · 2026-09-29