Artificial Intelligence (AI) has recently emerged as a focal point in global sustainability discussions. The narrative frequently leans towards disaster, with headlines claiming AI will drain national power grids or exacerbate water scarcity in regions hosting data centers. But how much energy does an AI query actually use? Is the typical prompt as environmentally damaging as people think?
Google’s newly released figures illustrate a more measured picture. One that does not dismiss the scale of AI’s resource demands, but does help put single-query emissions into perspective.
Understanding AI’s Energy Footprint
At first glance, estimating AI’s environmental impact can feel intimidating. The models are huge, the data centers that power them are energy-intensive, and the supporting infrastructure is complex. This makes AI an easy target for extreme claims, which are often echoed widely even when the subtleties are missing.
However, new data provides a more accurate picture of how much electricity, carbon, and water a single AI prompt requires. And the results may surprise you.
Google’s Newest Findings
Google recently published updated figures on the energy use of its Gemini model. While these figures have not been independently verified – something Google acknowledges – they do offer a useful snapshot of the typical energy profile of a text-based AI query.
According to the report, a single Gemini prompt uses around 0.24Wh of energy, which translates to roughly 0.03 grams of CO₂ per query. Alone, this number is almost negligible, particularly when compared to familiar everyday activities.
Putting AI into Perspective
According to the Carbon Disclosure Project, a latte from Starbucks produces around 0.844kg of CO₂. If one AI prompt generates roughly 0.03g of CO₂, then a single Starbucks latte has the same carbon footprint as approximately 28,000 AI text queries.
Similarly, a typical flush uses about 5 liters of water, while an AI query uses 0.26mil. This means that one toilet flush consumes roughly the same amount of water as 19,000 AI queries.
Where AI’s True Footprint Lives
Although individual AI queries have tiny footprints, the infrastructure powering them does not. Data centers consume huge amounts of electricity, require significant water for cooling, and are expanding rapidly as AI adoption continues to climb. So, while each prompt is low impact, the collective system powering AI carries a substantial, and growing, environmental footprint.
Why This Conversation Matters
The real story of AI’s environmental impact is one of scale and balance. A single prompt is remarkably low impact, far lower than public headlines suggest. The true environmental load lies in the large, continuously expanding data centers required to power AI. To manage this responsibly, organisations need clear, transparent data and a realistic understanding of where AI fits within their broader sustainability picture.
As AI usage increases, so does the need for accurate carbon accounting, credible methodologies, and informed decision-making. At Enistic, we help organisations understand their digital emissions, integrate AI usage into existing compliance frameworks, and plan sustainability strategies grounded in real data, rather than speculation.
If you would like support evaluating your digital footprint, or building a smarter path to Net Zero, get in touch today.