01 Where Ai Water Comes From

Complete transcript and visual descriptions

Slide 1: AI’s Water Footprint: What the Evidence Can—and Cannot—Tell Us
Visual description: Title slide identifying Tao Wen and the Syracuse University module.

Welcome to “AI’s Water Footprint: What the Evidence Can—and Cannot—Tell Us.” I am Tao Wen, Associate Professor in the Department of Earth and Environmental Sciences at Syracuse University. My work examines water quality, hydrogeochemistry, and coupled human-natural systems. I am not an author of the three papers in this module. My role is to help you read them through a water scientist's lens: Where is the water coming from? What happens to it? What evidence supports the claim? And what would we still need to know locally? You do not need experience in hydrology, computing, or statistics. By the end, you should be able to ask better questions about the water quantity and water quality associated with AI infrastructure.

Slide 2: A Central New York Community Has a Question
Visual description: Two competing claims about a fictional data-center proposal point to one question: what evidence would decide?

Imagine that a company proposes an AI data center in a fictional Central New York community. Supporters point to jobs and computing capacity. They say the facility will use water efficiently. Critics worry that it could strain local supplies, affect drinking water, or harm fish. Those concerns should not be settled by the most confident voice or the most dramatic number. They are questions about evidence and local context. A national estimate may help the community understand possible pathways, but it cannot substitute for a facility design, a water permit, a discharge sample, or knowledge of the receiving watershed. The community's guiding question is therefore not simply, “Is AI good or bad?” It is: “What can current peer-reviewed evidence tell us, and what must be measured or documented here before a decision?”

Slide 3: Water Can Be Used On Site and Through Electricity
Visual description: Direct water moves from a local source to data-center cooling. Indirect water is associated with electricity generation that supplies the facility through the grid.

Start with the computing. Servers use electricity and produce heat. That heat must be moved away from the equipment. Some facilities use cooling systems that withdraw water and lose some of it through evaporation. Other designs use less on-site water but may use more electricity. Climate, operating conditions, and cooling technology all matter. We call water used at the data-center site **direct water use**. In the diagram, follow the solid path from a water source to the cooling system. Water may enter the site, circulate, return to a treatment system or receiving water, or leave as water vapor. Now follow the second path. A data center also needs electricity. Producing that electricity can involve water use somewhere else on the power grid. In the system boundary used by our studies, that is **indirect water use**. Its amount depends partly on which electricity sources supply the facility and where generation occurs. An important point is that “off site” does not mean “unimportant.” Indirect water use may occur in another watershed, where water availability and ecological conditions are different. A complete estimate must state whether it counts only the facility, the electricity supply, or a wider supply chain.

Slide 4: Withdrawal and Consumption Ask a Different Question
Visual description: A facility withdraws 100 cubic meters, returns 70 to the same watershed, and consumes 30 through evaporation.

Direct versus indirect tells us **where** a water pathway occurs. Withdrawal versus consumption tells us **what happens to the water**. These are different distinctions. A **withdrawal** is water taken from a river, lake, aquifer, or public water system. Some withdrawn water may return. **Consumption** is the portion that is not returned to the same source at a useful time, often because it evaporates or is transferred elsewhere. Consider a simplified example. A facility withdraws 100 cubic meters, returns 70 cubic meters to the same watershed, and loses 30 cubic meters through evaporation. The withdrawal is 100. The consumption is 30. Saying only “the facility uses 100” hides an important difference. Return flow is not automatically harmless. Its timing, temperature, and chemistry may matter. Likewise, consumed water is not automatically unacceptable. Its significance depends on season, local supply, competing users, and the watershed's ability to support the demand.

Slide 5: Water Quantity Is Not Water Quality
Visual description: Quantity questions concern source, volume, timing, and return flow. Quality questions concern temperature, dissolved solids, treatment chemicals, permits, and receiving water.

**Water quantity** questions include: How much is withdrawn? How much is consumed? From which source? During which season? How much returns, and when? **Water quality** questions ask about the condition of water. Cooling can warm water. Evaporation can concentrate dissolved salts in circulating cooling water. Facilities may remove part of that circulating water as **blowdown** to control buildup. A local review might therefore examine discharge temperature, dissolved solids, treatment chemicals, permit limits, and conditions in the receiving water. These are possible pathways, not proof that every data center pollutes. A facility may manage, treat, reuse, or discharge water under different designs and regulations. To claim either “there is no water-quality risk” or “this facility will contaminate the stream,” we would need evidence about the actual system and site. Quantity and quality also interact. A warm discharge may have a different effect in a large, cool river than in a small stream during low flow. The same withdrawal may matter differently during a wet spring and a dry late summer.

Slide 6: Professor Wen's Water-Scientist Lens
Visual description: Five questions surround a local watershed: source and season; withdrawal or consumption; direct or indirect; return-flow condition; and receiving-water context.

Here is the water-scientist lens we will use throughout the module. First, ask about **source and season**. Is the source a river, aquifer, lake, reclaimed-water system, or municipal supply? When is demand highest, and who else relies on that water? Second, ask whether a number describes **withdrawal, consumption, or both**. Do not treat those terms as interchangeable. Third, inspect the **system boundary**. Does the estimate include on-site cooling, electricity generation, both, or something wider? Where does the indirect water use occur? Fourth, ask about **return flow and condition**. How much water returns? Where does it go? What are its temperature and chemistry, and what permit limits apply? Fifth, ask about the **receiving watershed**. What are the stream flow, existing stresses, habitats, and sensitive species? What monitoring data already exist? An annual national number cannot answer those questions by itself. It can help us identify what to investigate. Local evidence is what supports a local decision.

Slide 7: Checkpoint: Sort Four Examples
Visual description: Four examples to classify by direct, indirect, withdrawal, consumption, or water quality.

Pause now for a two-minute checkpoint. You will see four short examples. Sort each by the distinction it makes most visible: direct, indirect, withdrawal, consumption, or water quality. More than one concept can sometimes apply, but choose the label best supported by the wording. The feedback will explain the overlap. Remember: direct and indirect describe where the pathway occurs. Withdrawal and consumption describe what happens to water. Quantity and quality describe different dimensions of potential impact.

Sources

- T. Xiao et al. (2025), Environmental impact and net-zero pathways for sustainable artificial intelligence servers in the USA, Nature Sustainability, doi:10.1038/s41893-025-01681-y.
- N. Lei et al. (2025), The water use of data center workloads: A review and assessment of key determinants, Resources, Conservation & Recycling 219, 108310, doi:10.1016/j.resconrec.2025.108310.
- H. I. Jager & H. S. Yoon (2026), Data centers: an emerging threat to freshwater biodiversity in the United States, Water Biology and Security, 100585, doi:10.1016/j.watbs.2026.100585.