Artificial intelligence is an invaluable resource for the development of mankind on the planet. Computational capacity can improve life chances and create new and better hopes for mankind but none of this is free and perhaps many aspects continue to be underestimated.
The purpose of this article is to suggest a reflection on the usefulness of technologies such as blockchain or artificial intelligence, in relation to their consumption in a world with fewer and fewer primary resources available. The aim is not to declare their uselessness, on the contrary their strategic importance to the point of reflecting on their actual level of deployment.
Artificial intelligence and water consumption
In 2022, Google, Microsoft and Meta consumed more than 2 billion cubic metres of fresh water: 2 billion cubic metres of fresh water is an impressive amount and, according to Dr. Veronica Balocco in the Corriere delle Comunicazioni, an estimated 4.2 and 6.6 billion cubic metres of fresh water will be consumed in 2027 (for more information click here).
What fresh water is for
Fresh water is used to cool the processors in charge of carrying out the calculations necessary for artificial intelligence algorithms: the busier the processor, the higher the temperature released and the greater the demand for water. However, let’s take a practical example: CINECA’s LEONARDO computer receives incoming water at 37°C, which is run to cool the systems involved in the calculations and at its outlet reaches a maximum of 47°C.
The case of LEONARDO is interesting because from the presentation made by CINECA we learn that the refrigeration system is a closed circuit in which water is never wasted, and in fact, being at 37°C, it is easy to see that there is constant recirculation. It is an intelligent idea that, as CINECA President Dr. Francesco Ubertini claims, requires however a certain ‘ethics in the use’ of the system: it must be used for important things.
What is 2 billion cubic metres of water
The case of Roma Capitale
Answering this question is not easy but it is possible by doing a few simple mathematical operations.
Data
Per capita water consumption: 155 cubic metres (source: ANSA)
Population of Rome: 2,800,000 inhabitants (source: Municipality of Rome for year 2023)
Calculations
Total consumption = Per capita consumption x Population
Total consumption = 155m3/year x 2,800,000 inhabitants
Total consumption = 434,000,000m3/year
It should be noted, however, that this value (434,000,000) only concerns non-housing use and does not take into account industrial, commercial, tourist uses. Therefore, it is quite plausible to round this figure up to about 550,000,000m3.
Estimated total annual consumption = 550,000,000m3
Estimated total daily consumption = 1,506,849m3
--------------
1,506,849m3 = 1,506,849,000 litres
In Rome, in essence, an estimated 1,506,849,000 litres of water per day will be consumed in 2023. A very high value that does not hold a candle to the figures declared by London (a city far more populated than Rome).
The case of London
According to the official website of the City of London, daily water consumption is 2,600,000 litres or 2,600,000m3. Also keep in mind the difference between the two cities:
| Rome | London | |
|---|---|---|
| Population | 2.848.084 | 8.982.000 |
The case of New York
New York is one of the largest cities in the world and, according to the city’s official portal(NYC.GOV), water consumption is said to be around 1,000,000 gallons. It should therefore be noted that:
1 gallon = 3.78541 litres
1,000,000 gallons/day x 3.78541 = 3,785,410,000 litres/day
Reconsider what Dr. Balocco reported:in 2025, the annual consumption of fresh water is estimated at 4.2 and 6.6 billion cubic metres to run A.I. In essence, one year of cooling A.I. costs as little as one day’s consumption of New York and Rome combined.
| Cities | Litres/day |
|---|---|
| New York | 3.785.410.000 |
| London | 2.600.000.000 |
| Rome | 1.506.849.000 |
| Milan | 630.000.000 |
Does it seem little? Does it sound sustainable? Bear in mind that a low-income country, according to information repeatedly published by the WHO and UNICEF, lives on 20 litres of water per day.
Block-Chain and resource consumption
The other technology that has been the subject of much discussion in terms of energy consumption is the blockchain, and questions have been raised for years on the possibility of optimising such consumption. Below are some very simple calculations to estimate the energy consumption required for mining Bitcoin and Ethereum.
Bitcoin
Suppose we want to use a single high-performance ASIC miner based on the AntminerS19 Pro system. The average hourly consumption is 3,250 watts (or 3.25 Kw/h).
Daily consumption: 3.25 kW/h x 24 hours/day = 78 kWh/day
A single system will therefore consume 78 kW in one working day. Imagine using 1,000 Antminer S19 systems and creating what is called a mining-farm, i.e. a multiple bitcoin production system, the result would be as follows.
1,000 systems x 78kWh/day = 78,000kWh/day (i.e. 78 MWh/day)
Ethereum
For Ethereum, imagine using an NVIDIA RTX 3080 GPU with a power consumption of about 300 watts (or 0.3 kW). The result would be:
Daily consumption = 0.3 kW×24 hours/day = 7.2 kWh/day
1,000 GPU×7.2 kWh/day = 7,200 kWh/day (i.e. 7.2 MWh/day)
Summing up consumption
Summarising the consumption estimate and trying to put it in an everyday context, we can state the following in the table.
| Type | Energy required |
|---|---|
| Bitcoin Farm from 1,000 AntMiner systems | 78 MWh/day |
| Ethereum Farm 1,000 GPU NVRTX3080 | 7.2 MWh/day |
These are very significant consumption figures which, clearly, should be closely monitored. Considering the average energy consumption per capita and population, it is possible to estimate that:
19,178 MWh/day consumed for Milan (pop. 1,400,000)
11,918 MWh/day consumed for Turin (pop. 870,000)
5,014 MWh/day consumed for Florence (pop. 366,000)
Finally, to understand the energy consumption produced by Bitcoin Farms even better, one can consider the famous statistic that the entire Bitcoin network consumes 120 Terawatt hours (TWh) annually and Norway, again annually, consumes 124 Terawatt hours (Source: Forbes). This statistic may help to better understand.
| Consumption in TWh/year | Subject |
|---|---|
| 124 | Norway |
| 120 | Bitcoin Farm |
| 120 | Holland |
| 83 | Belgium |
| 83 | Finland |
| 50 | Portugal |
| 34 | Denmark |
It is important to understand that energy production is not always based on renewable and ‘clean’ resources but, very often, produces the pollution and consumption necessary to generate the required energy.
Weighing, yes, but how?
Let us focus on artificial intelligence: the use of coolants and desalinators are certainly solutions but the more important question is what do we use artificial intelligence for? If one certainly wanted to be controversial, the use of AI for constructing memes with Gerry Scotti’s face, certainly, could be avoided while the use of such technology to optimise production cycles, medical research, energy consumption systems, is welcome. It is therefore a question of weighing its use carefully in order to avoid burdening the ecosystem with unnecessary causes, thereby fuelling its deterioration, but weighing is difficult when the technology is massified and can potentially have millions of applications (from the most useful to the most useless).
The energy consumption generated by artificial intelligence could be counteracted by resource optimisation and, probably, by higher access costs, which companies, however, will want to avoid in order to prevent the barrier to entry to the technology and the reduction of the user base. The question therefore arises: is the free use model of artificial intelligence, as we are currently experiencing it, really sustainable?
It is worth asking these questions: the sustainability of these technologies cannot be at the expense of life on the planet, despite their convenience and efficiency.
