Water impact
When discussing the environmental impact of artificial intelligence, the debate almost always focuses on the energy consumed by data centers and CO₂ emissions.
Much less attention is given to its water footprint: the amount of water required to cool the processors that power increasingly advanced models.
Over the past twelve months, this concern has intensified, driven by the race among major tech companies to develop tools based on generative artificial intelligence. These systems, built on large language models, are capable of analyzing and producing vast amounts of textual, numerical, and multimodal content.
However, their operation requires extremely high computing power, which translates into the use of increasingly large data centers. To keep these infrastructures running, a constant cooling system is needed, often relying on water to absorb the heat generated by servers. During this process, part of the water is lost through evaporation, while another portion can be recovered and reused in subsequent cycles.
Growth figures
Some data help illustrate the scale of the phenomenon.
According to widely cited estimates, training the GPT-3 model in Microsoft’s state-of-the-art data centers in the United States may involve the direct evaporation of around 700,000 liters of clean freshwater, used to prevent overheating of computing systems.
At a global scale, researchers at the University of California suggest that the impact will become even more significant: by 2027, global demand for artificial intelligence could require between 4.2 and 6.6 billion cubic meters of water withdrawals.
A matter of scale
These figures are striking, but on their own they tell only part of the story. A useful comparison helps put things into perspective: in Italy, around 12.3 billion cubic meters of water per year are estimated to be used solely for agriculture.
In the United States, agriculture and industry together account for roughly 90% of total freshwater consumption, while artificial intelligence remains, both globally and nationally, a marginal fraction compared to these dominant sectors. At a global level, therefore, the water consumption of data centers is not currently the main driver of pressure on water resources.
The real risk of data centers: hyper-local concentration
However, as noted by J.P. Morgan experts, the issue becomes critical due to geographical location.
Water is a highly local resource, and new industrial demand from AI is concentrating in areas already classified as water-stressed.
In the United States, for example, four (Northern Virginia, Dallas, Chicago, and Silicon Valley) of the five largest data center markets are located in regions with high or medium-high water risk. Their location is mainly driven by proximity to end users, telecommunications networks, and state-level incentives.
In such contexts, even a “marginal fraction” can become the factor that strains local water basins, directly competing with the needs of residents and farmers.
Case studies and local community impact
The impact of large technology hubs becomes particularly evident in real-world cases such as Newton County, Georgia. Since 2018, alongside the construction of Meta’s data center, local water pressure has dropped dramatically to a mere “trickle,” damaging private wells and making household appliances unusable.
This crisis has led to a projected 33% increase in municipal water rates over the next two years and places the county at risk of a total water deficit by 2030.
In Indiana, Amazon’s massive AI-focused complex, spanning over 4 km² of cornfields, has raised similar concerns about groundwater management. During early construction phases alone, the company was authorized to pump 8 million liters of water per hour from the ground for 730 days.
This process, known as “dewatering,” has caused several private wells used by local farmers to run dry, significantly disrupting a long-established agricultural community.
These examples show that the issue does not lie in data centers themselves, but in their pressure on fragile water basins. Many of these facilities are located in areas where water sources depend primarily on rainfall recharge.
The infrastructure gap
The state of water infrastructure is now one of the least visible yet most critical challenges for economic and industrial stability, in a context shaped by rising resource demand linked also to artificial intelligence development.
In the United States, much of the existing water network is between 50 and 100 years old and is approaching or has already exceeded its useful life.
This structural obsolescence translates into growing fragility precisely as demand pressures increase.
The issue also has a financial dimension. The estimated funding required to maintain current service levels is around $90–100 billion per year.
To make the system truly resilient, including climate adaptation and new industrial demand, estimates rise to about $150 billion annually, with a cumulative need exceeding one trillion dollars over several decades.
In this scenario, public intervention alone appears insufficient to close the gap. Private sector involvement and public–private partnerships therefore become central, often seen as more efficient mechanisms for delivering and managing complex infrastructure.
Investing in upgrading local water networks and resource management technologies is no longer just an environmental responsibility, but a strategic necessity. Companies able to ensure operational continuity in water-scarce contexts and contribute to territorial resilience will also be better positioned competitively, supported by a growing “social license” to operate.
Towards a less “thirsty” AI
The future of AI sustainability depends on technological innovation and social responsibility. Companies are beginning to explore closed-loop liquid cooling systems or air-based alternatives to reduce direct dependence on freshwater withdrawals.
Artificial intelligence has both the opportunity and the responsibility to drive change by adopting a rigorous and transparent framework for measuring its water footprint, balancing technological demand with local community needs.
Only by integrating water management into overall corporate strategy will it be possible to ensure that digital development does not come at the expense of a vital and irreplaceable resource.