0.24Wh
Electricity use of a text query to Gemini
Median · text query · Google measurement, August 2025
A median text query to the Gemini app consumes 0.24 Wh of electricity, according to the measurement published by Google in August 2025 on its own data centres.
!A median over text queries: long answers and other models differ from it.
Definition and scope
Median electricity consumed by a text query to the Gemini app, measured by Google on its own infrastructure.
✓What is included
- ✓AI chips in service
- ✓The processors and memory of the host servers
- ✓Machines kept in reserve
- ✓Cooling and equipment of the centre
✕What is excluded
- ✕Training the model
- ✕Hardware manufacturing
- ✕The user's device and the network
How the value is obtained
Measurement by Google in production. In the narrower and older approach (active AI accelerators only, in the sample of the 10% most efficient data centres), Google obtains 0.10 Wh: the broader scope more than doubles the value. This is not the share of the chips in the 0.24 Wh: in the full calculation, the chips account for 0.14 Wh.
Limits and uncertainties
- Measurement published by the provider itself, not audited by a third party.
- Valid for Gemini; it cannot be transposed as it stands to other assistants.
The order of magnitude
All figures0.24 Wh on an electricity scale, with 6 other reference figures, brought to the same unit. Each tick multiplies by 10.
In context
See the chart and its dataElectricity use of a question, an image, a video
Server electricity per task, in watt-hours, counted in written questions
Measurements and estimates, 2024 and 2025
An image costs about 10 times a question; a 5-second video about 300 times, with the most downloaded open generator.
AI Impactby Projet Celsius
Electricity
