Google Gemini AI: energy, CO₂ and water impact per query
Recently, Google published a technical report detailing —for the first time with concrete figures— the energy consumption, carbon dioxide (CO₂) emissions, and water usage associated with each query to its Gemini AI. According to the analysis, an average text request to Gemini requires 0.24 watt-hours (Wh) of electricity, 0.03 grams of CO₂ equivalent, and consumes about 0.26 milliliters of water, equivalent to approximately “5 drops” per query.
Google compares this consumption to watching television for just a few seconds, to provide context for the individual-scale impact.
These data represent a step forward in transparency and provide a realistic basis to discuss the ecological footprint of AI —essential for an informed debate on digital sustainability.
Does this mean AI no longer pollutes much? It depends on usage
Individual impact: low but cumulative
A single query has a small impact —0.24 Wh and 0.03 g CO₂—, which at an individual level is negligible: equivalent to a few seconds of TV, almost anecdotal electricity use. For an occasional user, the annual footprint of moderate AI usage will be small.
However, multiplied by hundreds of millions of daily requests —as occurs with popular AI services— the cumulative effect can be significant. The real issue arises when considering the total number of global interactions: millions of queries per day, over years.
The other side: global scale and data center context
Query efficiency has improved significantly —according to Google, the footprint per prompt has been reduced up to 33 times in energy consumption and 44 times in carbon emissions over the past year.
However, total energy demand to power these services continues to grow. AI data centers require constant energy, cooling, and maintenance resources.
Moreover, training AI models (development phase before models are used) can require vast amounts of energy and water, and generate a considerable carbon footprint —a cost difficult to assign to each individual query, but real and significant.
Thus, even if each query “seems harmless,” the aggregated environmental impact of AI —training + infrastructure + usage— is a serious challenge if usage continues to expand unchecked.
Why this matters for an ecology, responsible consumption, and sustainable living website
1. Awareness of digital footprint
We live in an era where many daily activities are digital as well as physical. Every message, search, and AI interaction —though seemingly intangible— has an environmental cost. Sharing data like Google’s helps make this “invisible” footprint visible to readers.
2. Transparency and responsibility of big tech
That a company like Google publishes concrete data on energy, emissions, and water per query sets a precedent: technology must be accountable not only for functionality but also sustainability. This level of transparency should be expected from all AI service providers.
3. Incentive for responsible and conscious AI use
For those promoting responsible consumption, these figures encourage reflection: do we really need mass generation of texts, images, or instant prompts? Can energy and resource savings guide decisions on when and how to use AI?
4. Debate on digital infrastructure sustainability
The challenge is not just “making each query consume less.” It’s about redesigning, deploying, and regulating digital infrastructure. Efficient data centers, renewable energy, sustainable hardware, and transparency —everything counts.
Challenges, warnings, and unknowns
Query figures don’t reflect everything
Google’s data refers only to “inference” —the process of responding to a trained query. It does not include the environmental cost of model training, ongoing server maintenance, manufacturing, extreme cooling, or data center construction.
Thus, a complete view of AI’s impact must integrate all these phases.
Rebound effect risk
When a technology becomes more efficient, usage tends to increase. If an AI query costs “less” —energy, water, emissions— it may encourage proliferation and increase global consumption, nullifying efficiency gains.
Dependence on energy source
The real footprint heavily depends on how electricity for data centers is generated. Fossil fuels create a higher environmental impact than renewable energy. Therefore, energy type matters as much as usage efficiency.
What can be done — recommendations for users, companies, and society
For users
- Be aware of the environmental impact of queries: avoid unnecessary or trivial AI use.
- Prioritize quality over quantity: use AI only when it provides real value.
- Share and spread information about AI’s ecological footprint: awareness helps change collective habits.
For companies and technology developers
- Promote energy efficiency: optimize software, use efficient hardware, reduce idle consumption, optimize cooling.
- Transparency at all levels: publish data on consumption, emissions, and water for usage, training, and maintenance.
- Invest in renewable energy in data centers: ensure all digital infrastructure runs on clean energy.
For governments and regulators
- Set environmental reporting and control standards for data centers: energy, water, emissions.
- Encourage sustainable infrastructure and transition to clean energy.
- Promote research and adoption of more efficient AI models (lower consumption, smaller footprint).
