AI Reasoning Models Consume Up To 100x More Energy
A study released Dec. 5, 2025 by the AI Energy Score project (Hugging Face's Sasha Luccioni and Salesforce's Boris Gamazaychikov) found AI reasoning models used about 100 times more power on average to answer 1,000 text prompts than non-reasoning variants across 40 open models, including OpenAI, Google, Microsoft and DeepSeek. The tests showed some models rose from roughly 50 watt-hours to 308,186 watt-hours with reasoning enabled, underscoring potential power-grid strain and rising inference costs.
Key Points
- 1Measured 100x average power increase for reasoning-enabled models across 1,000 prompts.
- 2Highlights significant inference-phase energy demands that could strain grids and raise electricity costs.
- 3Advises practitioners to choose lighter models or disable reasoning for routine, low-complexity queries.
Scoring Rationale
Strong empirical measurements across 40 models demonstrate large energy gaps; limited by single-project, non-peer-reviewed release.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
