How much does a Gemini prompt consume?
A median text prompt in the Gemini apps uses 0.24 Wh of electricity, 0.26 mL of water (about five drops) and emits 0.03 g of CO₂e, according to figures Google published on August 21, 2025. Google's own equivalence: less than nine seconds of TV per prompt. It is one of the sector's rare official disclosures, detailed in a technical paper.
Figures published on August 21, 2025, per median text prompt (full-stack inference) · Sources: Google Cloud : impact environnemental de l'inférence IA · Google : « Measuring the environmental impact of delivering AI at Google Scale » (arXiv 2508.15734, août 2025)
How does Google measure the 0.24 Wh of a Gemini prompt?
It's a full-stack measurement on production systems: it includes the active AI chips, the host CPU and memory, idle standby machines, and the data center's energy overhead (PUE). The figure is the median Gemini Apps text prompt as of May 2025. Google notes that a narrower method counting only active chips would give 0.10 Wh: the chosen boundary changes the result by more than double.
Why only 0.26 mL of water?
That figure covers the water consumed on site to cool the servers, not the water needed to generate the electricity upstream. Researchers have criticised the boundary for that reason; counting the upstream, estimates climb toward 1 to 2 mL per prompt, and Mistral's full LCA counts 45 mL per response. None of these numbers is wrong: they don't count the same thing.
Is 0.03 g of CO₂e per prompt credible?
The figure is market-based: it factors in Google's clean-electricity purchases, which give its fleet a carbon intensity far below the average grid. On an average world grid (~480 g/kWh), 0.24 Wh would rather correspond to ~0.1 g of CO₂e in operation. The published figure is consistent, but it reflects Google's power mix, not the grid where you live.
Does Gemini consume less than ChatGPT or Le Chat?
On electricity, published values are close: 0.24 Wh for Gemini (Google), ~0.30 Wh estimated for ChatGPT (Epoch AI), 0.34 Wh reported by OpenAI. Google also highlights a 33x drop in energy per median prompt over one year, driven by software and hardware optimisation. Against Le Chat, a direct comparison is impossible: Mistral publishes a full life-cycle assessment, not an inference figure.
A Gemini prompt in everyday actions
The figures published for Gemini, converted with the site's reference actions (every action value is sourced, see the Sources page):
- 1 hour of LED light = 42 Gemini prompts
- 1 hour of TV = 292 Gemini prompts
- 1 glass of water = 962 Gemini prompts
- One 5-minute shower = 192,308 Gemini prompts
- 1 email with attachment = 633 Gemini prompts
- 1 km by car = 5,667 Gemini prompts