As artificial intelligence processes billions of queries, the world’s most fundamental resource is emerging as an increasingly urgent corporate sustainability concern. But the debate goes beyond how many litres a chatbot consumes, raising bigger questions about local water security, corporate accountability and the ecosystems that sustain freshwater supplies.

How much water does it take to ask artificial intelligence a question? The answer depends on whom you ask, what technology is being used and, crucially, what is being measured.

The precise figures vary by model, data centre, climate, cooling technology and methodology. Google estimates that a median Gemini Apps text prompt consumes approximately 0.26 millilitres, or about five drops, of water. OpenAI chief executive Sam Altman has cited roughly 0.32 millilitres for an average ChatGPT query. Independent researchers, however, have arrived at considerably larger estimates. A study led by researchers at the University of California, Riverside estimated that generating 20 to 50 responses could require approximately half a litre of water.

The figures are not directly comparable. They reflect different AI systems, time periods and accounting methods, including whether water consumed in generating electricity is counted. The Riverside estimate was based on earlier AI infrastructure, while Altman’s figure has not been accompanied by a publicly verified methodology.

Nevertheless, the disparity highlights a growing challenge for the technology industry. As AI embeds itself deeper into daily life and data centres multiply, water consumption is moving from a niche sustainability metric to a mainstream business, political and infrastructure issue.

Local opposition has appeared in communities where data-centre development intersects with concerns about electricity, land and water. A Gallup survey found that 71% of American adults opposed constructing AI data centres in their local area, compared with 53% who opposed building a nuclear power plant nearby. Although water is only one of several concerns driving opposition, the findings suggest that public resistance to AI infrastructure is already substantial, even as the industry works to explain its environmental footprint and economic benefits.

Freshwater scarcity, after all, is a resource constraint that needs little explanation. A human being can generally survive only around three days without water, although this varies with environmental conditions and individual health. Data centres, meanwhile, have developed their own dependence on reliable supplies of it.

Where the Water Actually Goes

Data centres have both direct and indirect water footprints. Directly, water may be used in cooling systems that remove heat from servers and other equipment. In facilities that rely on evaporative cooling, some of that water is lost to the atmosphere. Indirectly, water can also be consumed in generating the electricity that powers these facilities, depending on the energy mix and the technologies used by power plants. The distinction between water withdrawal and consumption matters. Withdrawal measures the amount taken from a source, while consumption generally refers to the portion that is not returned for immediate use. A facility may withdraw significant quantities without consuming all of them, although both can affect local water availability.

According to the Environmental and Energy Study Institute, a large data centre can consume up to five million gallons of water a day, comparable to the needs of a town of 10,000 to 50,000 people. This represents an upper-end estimate rather than the typical consumption of every facility. The United Nations University Institute for Water, Environment and Health projects that, by 2030, the water footprint associated with data centres could be equivalent to the basic annual domestic water needs of 1.3 billion people living in Sub-Saharan Africa.

That comparison illustrates the potential scale of demand rather than suggesting that data centres will physically draw water from the region. The more immediate concern is geographical. As developers expand into locations offering relatively inexpensive electricity, land and infrastructure, water availability becomes another factor in determining where facilities can operate sustainably. In water-stressed regions, reliance on local freshwater supplies can bring data centres into competition with households, agriculture and other industries. The consequences depend not simply on how much water a facility consumes, but on where that water comes from, the condition of local resources and the availability of alternatives.

Efficiency is Real, Scale is the Problem

The industry has begun responding. Operators are exploring direct-to-chip and other forms of liquid cooling that can reduce or, in certain configurations, eliminate the need for evaporative cooling. Companies are also adopting metrics such as Water Usage Effectiveness to measure and manage consumption, while some are pursuing water-positive commitments.

NVIDIA says its latest AI infrastructure can use closed-loop liquid cooling with dry coolers, potentially eliminating evaporative cooling water consumption in suitable climates. Such systems can reduce direct water demand considerably, although their performance depends on local temperatures, infrastructure and operating conditions. The World Economic Forum has similarly highlighted innovative circular water-management strategies, including water reuse, optimisation and replenishment initiatives. It identifies opportunities for water savings of up to 75% through such approaches, although actual reductions depend on the technologies and facilities involved.

The difficulty is implementation. New facilities can be designed around more water-efficient systems, while existing data centres may be considerably harder or more expensive to retrofit. Economics, local climate, available infrastructure, regulatory requirements and the availability of alternative cooling technologies all affect what operators can adopt. There is also a broader question of scale. Even if individual facilities become more water-efficient, rapid growth in AI infrastructure could offset some of those gains. Reducing consumption per query or per unit of computing power does not necessarily mean reducing the industry’s overall water footprint.

Water as the New Carbon

For investors and corporate communications leaders, this debate should feel familiar. Over the past two decades, carbon emissions have moved from a largely technical environmental metric to a board-level concern, influencing investment decisions, regulation, corporate reporting and supply-chain strategy.

Water is following a similar trajectory, but with an important distinction. While greenhouse gas emissions contribute to a global climate problem, the immediate consequences of water consumption are often intensely local. A litre consumed in a water-abundant region does not carry the same implications as a litre consumed in a community already experiencing drought. That makes water an increasingly important consideration in infrastructure planning, risk management and corporate accountability. Companies must look beyond aggregate consumption figures to understand the condition of the water systems on which their operations depend.

In this context, water protection initiatives feel more urgent than ever, a theme also reflected in conversations at Impact Week in Singapore, where a range of environmental issues and initiatives were discussed. Among them was National Geographic’s Blue Boundaries programme, which approaches water protection through the interconnected ecosystems that sustain freshwater supplies, biodiversity and coastal resilience. Although Blue Boundaries is not an AI-focused initiative, its approach offers a useful perspective on an increasingly urgent business challenge: understanding water as a system rather than simply a resource to be measured.

The Case for Thinking Beyond Litres

Launched in partnership with the Chubb Charitable Foundation, National Geographic’s Blue Boundaries is a seven-year, multimillion-dollar commitment, the largest single grant in the Society’s 137-year history. It aims to protect interconnected freshwater wetlands, coastal ecosystems and reefs that sustain biodiversity, livelihoods and water security.

Its first cohort of Explorers, announced in September 2026, will work across three regions. In the Lower Mekong Basin, researchers are protecting giant freshwater stingrays and tracking migratory megafish as indicators of river health. Along the Mississippi River, scientists are studying water-cleaning microbes, agricultural runoff and the carbon balance of ancient cypress swamps. Across southern Mexico, Belize and Guatemala, projects will examine manatee movements, freshwater aquifer health and the role of Indigenous knowledge in wetland conservation.

The significance of Blue Boundaries, viewed alongside the AI water debate, is its systems-level approach. Rather than defining success simply by litres saved, it combines scientific research, conservation and community engagement to protect the ecosystems on which freshwater supplies depend.

For the AI industry, the lesson is equally relevant. Its water footprint cannot be reduced to consumption figures alone. Where water comes from, how it is used and what happens when demand scales are just as important. As AI infrastructure expands, water security must become a consideration in corporate strategy, not simply another sustainability metric.