Google steps forward from the data fog and says: “This is how much energy our AI consumes.” Transparency points? Absolutely. But wait a minute – are the numbers too good to be true? Here you get both applause and suspicion in an AI-scented package.
Table of content
Google reveals the energy efficiency of Gemini
According to Google, a simple text query to their Gemini AI only consumes 0.24 Wh of electricity – roughly equivalent to watching TV for nine seconds – uses up 5 drops of water and emits 0.03 grams of CO₂. It almost sounds like environmentally friendly magic. The company also claims that consumption has drastically dropped within just one year, both in electricity and emissions. They say the calculations cover everything from active chips to passive servers and the entire datacenter’s overhead. Yet at the same time, Google plans to expand new nuclear power for AI. Perhaps because AI searches use 10 times more power than a typical search query, and the AI hype means more and more companies need extreme computing power. Who owns that power? Major players like Google, of course.
A breath of fresh transparency
The fact that Google actually shares such detailed figures is a bit like seeing a unicorn in a server room – it doesn’t happen often. The tech giants are rarely eager to show what happens behind the scenes, especially when it comes to environmental impact. So, hats off to Google for that. But numbers are numbers and there is criticism of these.
But what is missing from the numbers?
Critics have already started waving warning flags. The figures only apply to the typical text query, not the more advanced questions involving image, sound, or video. And more importantly – they do not include all the water used indirectly, for example to cool power plants that supply electricity to the data center. And the CO₂ figures are based on “market data,” not actual emissions.
Moreover – while each individual prompt has become more efficient, the total energy consumption has increased. Massively. This is the classic Jevons paradox: the more efficient something becomes, the more it is used. And Google is no exception.
How does this compare with the rest of the AI world
Generative AI is energy-hungry, especially during the training phase. Some models have required massive amounts of electricity and water – we’re talking pool size. And although Google’s figures concern usage, not training, it gives a hint of how resource-demanding the entire ecosystem is. Looking only at “per prompt” consumption is a bit like counting calories in a Big Mac without counting fries and soda. For example, Deepseek is a really thirsty monster, ChatGPT5 is said to consume 0.40 Wh of electricity.
Conclusion applauded with a big pinch of salt
Google deserves credit for releasing some kind of numbers – that is more than most tech giants dare to do. But don’t take it as absolute truth. The figures are narrow, exclude large parts of AI’s real impact and present a somewhat polished picture. More independent review and a touch of reality are needed before we write a green card for AI.
Sources
https://www.theverge.com/report/763080/google-ai-gemini-water-energy-emissions-study
