Information terrorism on AI and the water we actually waste

In a few areas, such as water consumption, the information terrorism on artificial intelligence risks distorting reality by preventing us from looking at the problem from the right perspective and blocking all reasoning in the middle.

The first half is now known to most. By 2027, analyses tell us, artificial intelligence could consume, between server cooling and power generation, between 0.3 and 0.6 billion cubic metres of water per year out of a gross water withdrawal of around 6.6 billion cubic metres, most of which will be returned to the system. These are real, documented figures that certainly deserve attention and concrete action: the problem exists and must be addressed. But there is also a second half of the argument. Every year in the world, 126 billion cubic metres of water are put into water networks and do not reach their destination (or are not billed). This is what experts call Non-Revenue Water: physical leaks in the pipelines, undetected faults, unauthorised connections, metering errors, unaccounted consumption. A haemorrhage whose cost can be estimated at almost fifty billion dollars a year. It is on this scale that AI consumption must be measured to grasp its true proportion. That half a billion cubic metres that AI will consume each year represents less than 0.5% of the water that networks already disperse today. In short: for every litre that artificial intelligence evaporates in its data centres, the world's water networks let more than two hundred evaporate.

In Italia the proportion is, if possible, even more pitiless. Aqueducts waste three and a half billion cubic metres of water every year: almost half of all that is put into the system (out of a European average of 25%). In other words, our country alone wastes almost six times the water consumption that AI will generate globally by 2027. It is in this passage that information terrorism, i.e. the systematic selection of data according to the narrative one wants to support, not only distorts the perception of the problem, but also prevents one from seeing the solution. Because the question that is never asked is the obvious one: given that no one argues that AI-generated waste should be minimised, the real issue is how much does AI consume or how much are we able to harness AI to help us save throughout the water cycle?

Some examples? In Sheffield, UK, using AI and IoT, visible leaks were reduced by almost half. In Monterrey, Mexico, machine learning has achieved water savings of between 17% and 35%. Europe's most advanced utilities have already brought losses down to below 15-20%; the world's best have dropped below 5%. Even in Italia, companies such as Gruppo CAP or MM have launched projects that exploit AI to reduce intervention times, monitor plants, and optimise incoming flows: all activities that, on the whole, reduce waste. In short: we are not talking about remote hypotheses, but about things that can already be done.

Let's try a deliberately cautious exercise: let's assume we apply digital technology, artificial intelligence and intelligent sensor technology to just 10 per cent of global water networks, achieving an average saving of 10 per cent: a fraction of what has already been demonstrated in the field. Even in this minimal scenario, around 1.3 billion cubic metres of water would be recovered each year: more than twice the water consumption of AI itself. The balance, even under the most conservative assumptions, is unequivocal.