Follow the Water Past the Fence

The word water changes the atmosphere of the meeting.

People can be uncertain about megawatts. They know the experience of opening a tap. They know a garden turning dry, a notice asking for conservation, a pipe being repaired in the street. A claim about water reaches ordinary life quickly.

That closeness makes care essential. A frightening number can run ahead of what it measures. A soothing phrase can hide the question residents meant to ask. Both can leave a community less able to judge a real proposal.

Imagine the developer says its cooling system is highly efficient. A resident asks whether the town can support it during a dry summer. The answer is repeated: highly efficient.

It may be true. It is still not the answer.

Efficiency compares something with something else. Availability concerns a source, other demands and a time. To connect the two, we need to follow the water through the design and into the local system that supplies it.

We should begin with the route, not the adjective.

Three words that should travel together

Withdrawal means water taken from a source. Consumption concerns the portion no longer available to the immediate water environment, including evaporation. Return flow is water that goes back after use.¹

Those distinctions prevent several misleading shortcuts. Water entering a facility is not necessarily all lost to the local system. Water returned somewhere is not necessarily restored at the same place, time or condition. A purchase from a water utility is not the same record as a direct withdrawal from a river or well.

The ordinary phrase water use can cover different parts of this account. Before comparing two claims, ask what each includes.

Suppose one facility reports the water delivered through its meter. Another reports an estimate of the portion evaporated. A third describes a wider account that includes electricity production elsewhere. Putting the three figures side by side without labels could reward the least complete report rather than the best design.

This is an accounting problem with physical consequences. The boundary determines which effects enter the discussion and which remain out of sight.

A public record should make the boundary easy to find.

Heat has to go somewhere

Computing equipment produces heat during operation. Cooling must move that heat away from the parts that need protection, and eventually reject it to the surroundings or put some of it to another use.

There are several places in that chain where water may or may not be involved. A liquid near a chip tells us something about how heat is carried from the equipment. It does not, by itself, tell us how heat leaves the entire facility. The final design and operating conditions matter.

A 2025 review and modeling study of data-center workloads examined differences in hardware, utilization, cooling, climate and electricity supply. Its message for comparison is that site efficiency measures alone do not reveal the resources needed for a given amount of computing work. The paper modeled benchmark workloads; its results are not universal footprints for present-day chat requests.²

For a planning decision, the next step is to ask for the actual configuration. Where does heat move? Under what weather conditions does the system use water? What changes during unusually hot periods? What happens if the preferred source is unavailable? How much electricity does an alternative cooling mode require?

These are questions for qualified designers and local resource managers to answer. The resident’s role is not to improvise a cooling system from a leaflet. It is to insist that the relevant answers exist and can be checked.

A good proposal should welcome that distinction. Clear questions let competent engineering speak more effectively than vague reassurance.

Two places in one account

Scroll sideways to see all columns. Keyboard: focus the table and use the left and right arrow keys.

Text equivalent: Follow the water account

Part of the figure	Meaning
Book	The Cloud Has a Power Bill
Title	Follow the water account
Starting point	Ask what is inside the reported number.
Account	Data-centre cooling
Location and time	Which source? Which season?
Input	Water source
Path	Withdrawal
Operation	Use
Output path	Consumption
Output path	Return flow
Account	Electricity generation
Location and time	Which source? Which season?
Input	Water source
Path	Withdrawal
Operation	Use
Output path	Consumption
Output path	Return flow
Withdrawal	Water taken from a source.
Consumption	Water unavailable for immediate reuse.
Return flow	Water returned after use.
Boundary	Beyond operation
Equipment manufacture	A further account with its own places and times.
Reminder	Where water is used, check each account separately. Return timing and quality can matter.
Status	Conceptual paths only. Equal arrows do not show volumes or a complete water balance.
How to read the diagram	Each panel reads from source to use, then branches into two separate paths. Neither output follows the other. The accounts need not share a watershed. Connections do not quantify a balance.

Figure 3. Follow the water account. An original conceptual map separates cooling at the data centre from water used in electricity generation. Equipment manufacture belongs to a further account. Equal-weight arrows show relationships, not volumes or a complete water balance. This explanatory aid is not measured site data or a validated environmental assessment.

The diagram asks you to keep the locations separate. A choice at the data center can affect a different account upstream. That other account does not disappear because it is farther from the meeting room.

Berkeley Lab’s 2024 U.S. report estimated 2023 direct data-center water consumption at 66 billion liters and electricity-related indirect consumption at nearly 800 billion. The latter was modeled using regional electricity mixes, not a trace of each facility’s procurement. These are dated national estimates for all data centers, not global AI totals or present per-query measurements.³

Those figures cannot decide a particular town’s water question. They illustrate why the scope of an account matters. The local proposal still needs its own source, design and operating plan.

Nor should the broader account become an excuse to ignore the near one. Telling a neighbor that most estimated water impact occurs elsewhere does not settle whether a local source can support additional demand. A large distant total and a smaller local pressure can both matter for different reasons.

Responsible comparison keeps the places visible rather than using one to erase the other.

The season belongs in the number

Imagine two possible sites for the same computing service. Both report the same annual water quantity. At one site, the need is spread through conditions the supplier can manage. At the other, much of the need occurs during the period when the source is most strained.

The annual totals match. The planning questions differ.

This imagined comparison is enough to show why a yearly figure cannot carry the entire judgment. We need information at the interval relevant to the risk. A monthly pattern may reveal what an annual sum hides. A drought plan may need triggers that operate faster than a monthly report. The appropriate detail should follow the decision, not the convenience of the publicity material.

The source matters too. A river, reservoir, aquifer and municipal network are different systems. The term water available should say available from where, under which conditions and after which other needs are protected.

A supplier’s ability to deliver today is relevant evidence. It should be considered alongside credible future conditions and obligations to other users. A permit limit tells us what is legally allowed under specified terms; it does not automatically tell us what will be used or whether every future condition has been resolved.

We should avoid demanding an impossible promise that no shortage can ever happen. We should demand a plan for the conditions that deserve attention and a clear rule for responding when they occur.

That is how uncertainty becomes a responsibility rather than a loophole.

A stressed source has a history

A new project can arrive in a system that already has problems. That history must enter the decision without being rewritten to make the new project responsible for everything.

Virginia’s environmental agency announced a groundwater study in July 2026 describing existing pressure in eastern Virginia, including domestic withdrawals and historic industrial demands. The notice warned of expected declines in the Potomac aquifer. It does not establish that data centers caused those conditions or that every proposed facility lacks a workable supply.⁴

The general principle is straightforward. A pre-existing problem can constrain new use even when the new user did not create it. Responsibility for the history and responsibility for a proposed addition are different questions.

Imagine a community has already promised more service than its dry-season plan can comfortably support. An applicant says it is being blamed for old decisions. That complaint may be fair. Yet the community still cannot solve the problem by ignoring the next commitment. It needs to examine the addition honestly and address the older weakness as well.

Conversely, opponents should not attach every local water problem to the most visible new facility. Doing so can direct attention away from other causes and needed remedies. A precise account is more likely to protect the source than a convenient accusation.

We need enough honesty to name the whole problem and enough discipline to assign particular obligations.

Reporting is a beginning

A Virginia statutory change effective January 1, 2027 requires specified water suppliers to report potable and reclaimed deliveries separately, including a defined category of air-permitted data centers. Monthly totals are required, with quarterly totals allowed when monthly data are unavailable. At this book’s cutoff, this future reporting provision was not in force. It is not a new consumption cap or universal real-time disclosure rule.⁵

Better information can improve decisions. It does not make the decision for us.

A report may show that a promise has been met, that a forecast was too low or that the available categories cannot answer the question. Officials still need authority and a process for responding. A supplier still needs an operating plan. The public still needs a way to understand the record.

The distinction between reporting and permission should stay clear. Recording a quantity does not necessarily authorize it. Having permission does not remove a reporting duty. Being below one threshold does not establish that every relevant condition has been satisfied.

For residents, this can be frustrating. They may be sent from the operator to the water supplier, from the supplier to a regulator and from the regulator to a planning office. Each office can truthfully say that a different part of the issue belongs elsewhere. The result can still be a public gap.

A good local agreement should identify who coordinates the answers. One readable record can link the separate decisions without pretending that one office has powers it lacks.

The citizen should not have to become the missing connection between public institutions.

Reclaimed water still needs a plan

Suppose a developer offers to use treated wastewater instead of drinking-quality water. That could be a valuable option. It deserves serious evaluation rather than an automatic objection or an automatic pass.

What treatment and delivery are required? Who pays for the extra infrastructure? What happens when supply or quality changes? Does the proposed use displace another planned use? Where would the water otherwise go? How will the system operate during maintenance or drought?

These are design questions, not reasons to reject reuse. A promising alternative becomes more credible when its dependencies are visible.

The same applies to heat reuse. Sending useful heat to another customer may be attractive where the temperatures, distance, timing and demand fit. The agreement should identify an actual user and a workable connection. A possible future customer should not be counted as a benefit already delivered.

A town might require a feasibility assessment without assuming the result. Sometimes the practical answer will be favorable. Sometimes the additional works will not make sense. An honest no to a particular arrangement is better than carrying a decorative promise through every public presentation.

We should encourage resourceful designs while refusing imaginary savings.

Decide the drought rule before the drought

The hardest meeting is the one held after the source is already strained and everybody has been promised continuity.

By then, changing operations may be expensive. Residents may feel they are being asked to conserve while a powerful user receives an exception. The operator may say that interruption would break commitments made under the original approval. Officials may discover that their authority is less clear than their earlier language suggested.

A better bargain settles the main questions earlier. Which conditions trigger a response? What must the facility do? What alternatives are technically available and lawfully permitted? Who verifies the change? Which essential uses receive protection? How is the public informed?

The response should be specific enough to test. Reduce use is less useful than a defined action under a defined condition. A plan should also account for side effects. Switching cooling modes can affect electricity demand. Using a different source can create another pressure. Moving work elsewhere can shift impacts rather than remove them.

That does not make adaptation pointless. It makes the design more demanding and more interesting. The aim is to reduce harm across the relevant system, with the best available evidence, rather than improve one isolated number.

A responsible operator benefits from knowing the terms before investing. A community benefits from knowing that difficult conditions will not require inventing authority in a hurry.

Preparation is a form of fairness.

The choice has to remain real

Information and mitigation can solve many concerns. They cannot guarantee that every site is appropriate for every design.

A public process should retain the ability to require a different configuration, a different scale, a later stage or another location when the evidence warrants it. In some cases, the particular proposal should not proceed. That conclusion can coexist with strong support for useful computing elsewhere or under better terms.

If the only possible result of review is approval, residents are being offered a performance of participation. If the only acceptable result is rejection, the review cannot recognize a genuinely improved proposal. Both positions close the question before the evidence arrives.

A real process can change its answer.

It should also explain why the chosen option is better than feasible alternatives. If a lower-water design has a different electricity cost, show the trade-off. If a different site reduces one pressure but creates another, show that too. There may be no perfect choice. A defensible choice makes its reasons available and keeps its promises subject to review.

The resident at the meeting returns to her original question. Can the town support this facility during a dry summer?

Now the answer can contain a source, a season, a design, an operating rule and a responsible institution. It may include an unresolved condition. It may require a change. It will certainly be more useful than highly efficient.

Water is local enough to demand that detail and connected enough to demand that we look beyond the fence.

We owe it both kinds of attention.