The Water Dissensus – A Water Alternatives Forum
Improving “water efficiency” of data centres is a red herring
Improving "water efficiency" of data centres is a red herring
By Sameer Shah
with contributions from Nicole Wilson & Cassandra Workman
Billions of dollars are being invested to build new data centres in the USA as well as other countries. They are facing growing opposition from local communities, concerned, among other things, by their potential impacts on local water supplies. How should we understand and evaluate the complex issues, water amongst them, that data centres generate?
Last week, I began lecturing to a wonderful group of nearly 50 students in my course, "Society & Sustainable Environments." This is an introduction to sustainability thinking. Much like prior years, we spend the first few weeks differentiating between reductionism and holism—and the appropriateness of when to use these different scientific and philosophical approaches.
The difference between the two is central to our recent Water Alternatives article, "Water Efficiency Gains Won't Resolve Datacentres' Contested Future". In it, we argue that the development of data centres that are more water efficient may not quell the growing public opposition to their construction.
On what basis do we make this argument and, more importantly, why does it even matter?
Data centres are home to the IT hardware that Artificial Intelligence (AI) and modern computational applications require. Many critiques have been launched at the uptick in data centre construction. A major one is their ecological "footprint." This includes their water footprint—captured both in the amount of water consumed (i.e., not returned to its original source) for "direct" purposes, i.e., infrastructure cooling, and "indirect" purposes, including the water consumed in generating the electricity required for powering said data centres.
Scientists that estimate the volumes of water consumed across the AI supply chain and for operating data centres are making important public contributions. These studies are uncovering the scale of resource consumption, potential trade-offs between power and water use related to data centres' cooling practices, and possible means of reducing water use, including through more efficient cooling technologies. We refer to such contributions as "assessment-driven" research.
The problem with assessment-driven research is not what it is—but what it encourages, absent of wider, critical contributions that interrogate how, and for whom water should be used. Assessments can inadvertently create an information ecosystem in which water overuse is framed as a core resource management problem to be fixed with more efficient water-use technology. Fixes, such as "closed loop" cooling systems, are now part and parcel of passionate public defences of water use from corporate actors, such as Microsoft and OpenAI. For example, in June at Microsoft Build 2026, Microsoft's CEO went as far as to indicate that "the daily water usage [for state-of-the-art data centres] over the course of an entire year is roughly equivalent to what a single restaurant would use."
Putting aside how such claims can mislead the public about the water consumed across operational and supply chains, they can be a starting point for understanding why a reductionist focus on water consumed at a single site can mean little in the context of growing public outcries.
Reductionism, as a philosophy, nudges us to focus on the individual elements that comprise our systems: when we understand how those individual parts work, we can understand how the whole works. Deriving broader meaning about data centres, such as how they may be received by the public, by looking inwardly at their characteristics, such as the volumes of water used and the kinds of cooling technologies employed, can yield erroneous conclusions. One example of such a conclusion is, if we can reduce water use, we might expect a corresponding decrease in opposition.
But should we expect this? We suggest, on the contrary, that water efficiency gains could act as a license for expediting buildout and diminishing the concerns local communities have about data centres and AI. Water represents far more than a stock to be managed. It often serves as an entry point into communities, for better or worse. Debates over whether water should be used for certain ends often stands in for the larger kinds of change that local communities seek to prevent in the first place. Even more, we now see opposition uniting beyond local borders. The public is voicing immense concern about the potential wealth inequalities that AI can further, public health implications, such as social isolation and cognitive dependence, and increasingly, cybersecurity and cyberterrorism risks. If communities—and the public at large—are concerned about a broader set of cultural, economic, and relational changes vis-à-vis data centre development, why should we expect "water efficient" installations to reduce opposition, even marginally?
In returning to where we started, I want to communicate the same message I do to my students. Data centres, and the interests they advance, are taking hold in complicated settings that involve people, nature, cultures, histories, and economic interests. Instead of assuming an AI future is inevitable, many communities are doing the exact opposite—mobilizing, organizing, and challenging the very infrastructures that AI depends on. This should signal to us that a more holistic orientation is needed, one that considers the many different perspectives and interests around our collective future. Paying attention to these kinds of connections and relationships is what holism implies. Instead of looking inward at specific qualities of data centres, holism encourages us to look outwards, examining the relationships and risks that manifest when water is used as an input to further data centres.
Existing assessment-focused research, while certainly well-intentioned, can easily slip into naturalizing efficiency measures as the most adequate counter to the concerns around the ecological footprint of AI and data centres. This can narrow the public conversations that are so desperately needed in a moment of enormous and rapid technological transformation. In sum, reducing the story to one of closed-loops cooling systems will not resolve data centres contested futures.
Comments 5
In Nigeria and by extension the Global South, both data centres and water access are very limited and the conversation on how advancements in technology that favours half the global population affects social structures that not only benefits the other half but the entire whole. The framework of holism is fundamental and big tech companies need to be pragmatic about what they deem necessary to global knowledge economy and carry along ideas and projects that are tangible and lifelong. In this regard, water supply. If the analogy of every corporate organisation setting up a building, facility or industry that accompanies the construction of a water supply project, that will go along way in the global front in sustainable water provision. That's holism; and beyond wishful thinking of what big tech can do with thier corporate social responsibility funds, Global South can make laws to enforce water supply to communities where buildings are situated.
Amidst the fanfare of the digital revolution, the development of Artificial Intelligence (AI) is frequently championed as the primary driver of future civilization. However, behind this narrative of progress lies an ironic reality: the main obstacle in AI development today is not algorithmic limitations or computational power, but rather the total lack of exemplary leadership from AI developers and tech corporations in building an environmentally friendly and sustainable industrial ecology.
To this day, the public has yet to see a concrete example of a tech giant capable of engineering an AI operational ecosystem without sacrificing environmental sustainability, let alone avoiding exploitation in the name of efficiency.
The Dominance of the "Benefits Over Impact" Paradigm
Data and facts surrounding AI production remain heavily dominated by a one-sided development paradigm. The industry prioritizes economic utility and the acceleration of model capabilities over seriously mitigating the ecological hazards it creates.
1) Water and Energy Crisis: Data centers training large-scale AI models require billions of liters of fresh water for cooling systems, alongside massive electricity consumption that remains reliant on fossil fuels.
2) Carbon Footprint: The training process of a single large-parameter AI model can generate a carbon footprint equivalent to the emissions of hundreds of transcontinental flights, without adequate ecological compensation mechanisms.
Public Perception at the Grassroots: A New Exploitative Industry
This disparity between progress narratives and ground realities breeds understandable skepticism among grassroots communities. For lower-income populations directly experiencing environmental degradation—such as clean water scarcity and pollution—the arrival of the AI industry appears no different from traditional extractive industries.
The general public compares—and even considers worse—the AI industry to its exploitative predecessors, both in terms of:
1. Natural Resource Exploitation: Massive extraction of groundwater and energy consumption conducted without transparent environmental accountability.
2. Human Resource Exploitation: Reliance on low-wage labor in developing countries for data labeling and content moderation, which is psychologically demanding, without providing adequate social welfare.
Without genuine breakthroughs in ethical and sustainable industrial ecology, AI risks losing its social legitimacy and becoming merely a new form of capital concentration that damages the environment.
The concern that water-efficiency gains may not resolve the contested future of data centres points to a broader governance problem. In my view, efficiency matters, but it is not the same as sustainability.
A data centre may reduce litres of water used per unit of computing while total water demand continues to rise because of expansion in capacity, infrastructure and service demand. In water-stressed or highly variable environments, particularly semi-arid regions, the more important question is whether that demand is compatible with the hydrological limits and governance capacity of the basin in which the facility operates.
This is where the debate needs to move beyond technical performance indicators. A facility can be highly efficient at the engineering level and still generate significant water-governance friction if policy objectives, institutional capacity, operational practice and the interests of existing users are poorly aligned. The key questions are therefore not only how efficiently water is used, but also where it comes from, when it is abstracted, how cumulative demand is assessed, how drought restrictions are applied, how ecological requirements are protected, and who bears the consequences when water becomes scarce.
Looking at it from this perspective, efficiancy should be treated as one component of a wider socio-technical governance assessment, not as a proxy for sustainability. The issue is less about rejecting efficiency and more about recognising its limits. It cannot by itself answer questions of allocation, legitimacy, institutional accountability or distributional impact. The policy challenge, therefore, is to ensure that technical efficiency is evaluated together with absolute demand, basin conditions, competing uses and the capacity of institutions to manage scarcity transparently and adaptively. Otherwise, efficiency risks becoming a way of narrowing a political and hydrological problem into a technical metric.
We welcome this important debate about the water impacts of data centres. We share the authors' concern that indirect water use is often overlooked and that technical efficiency alone cannot address the wider environmental and social consequences of expanding data centre infrastructure.
At the same time, we would caution against treating water efficiency itself as a misleading direction. Improving efficiency can reduce water consumption and remains an important objective. The problem arises when efficiency improvements are presented as sufficient justification for development, without examining the full water footprint or its consequences for local communities.
Efficiency is necessary, but it is not sufficient.
In our view, responsible planning must begin well before a data centre is built. The first questions should concern whether a proposed location is appropriate at all. How much water is available? What are the existing and future needs of the local population, agriculture, and ecosystems? What will be the impact of electricity generation and its associated indirect water consumption? And could another location offer a more responsible solution?
The potential use of waste heat should also be considered from the outset. Wherever technically and economically feasible, recovered heat could supply district heating systems or nearby residential and public buildings. Such opportunities depend on local infrastructure and demand and should influence site selection rather than be explored only after construction.
Equally essential is transparent and comprehensive disclosure of both direct and indirect water use. Communities and public authorities need reliable, understandable information to assess the full consequences of a proposed development.
But transparency alone is not participation.
People affected by these decisions should have a meaningful voice before approvals are granted, including opportunities to question proposals, consider alternatives, and influence the outcome. Particular attention must be given to protecting access to safe drinking water and safeguarding ecosystems, especially in regions facing water scarcity.
Ultimately, the debate should not be limited to how efficiently a data centre uses water. We must also ask whether it belongs in a particular location, who benefits from it, who bears the risks, and who has the power to decide.
Technical innovation, responsible spatial planning, and democratic water governance must go hand in hand.
The most important decisions about a data centre's water footprint are often made before the first server is installed.
An important distinction indeed. water efficiency is not synonymous with water sustainability. From a WASH and environmental engineering perspective, reducing water use per unit of computation means little if total demand continues to rise at the expense of community water security and ecological integrity. The real measure of progress is not merely how efficiently data centres use water, but whether their consumption remains within sustainable catchment limits while safeguarding equitable access to this essential resource. Engineering can improve efficiency; responsible governance must ensure sustainability.