The Data Centre Equation: Five Per Cent of a City’s Water, Six Times Its Power
San Gorgonio Pass Water Agency, a State Water Contractor serving 225 square miles of western Riverside County in California, commissioned a white paper examining what a large data centre would require from the communities it serves. Its central finding runs against the way the debate usually frames the problem. A hypothetical facility with a 100-megawatt IT load would consume approximately 419 acre-feet of water a year, roughly five per cent of the City of Banning’s annual water use, and would draw nearly six times as much electricity as the entire city consumed in 2024. The agency states plainly that no developer has brought a data centre project forward in the Pass area and that it advocates neither for nor against development. The research exists so that the questions surface before they become decisions.
The ratio is the finding
The two figures are more useful together than either alone, and the comparison is straightforward once both sit against the same baseline.
Water demand equals five per cent of one city’s annual use. Electricity demand equals six hundred per cent of the same city’s annual use. Set against each other, the facility is roughly 120 times more demanding of the local grid than of the local water supply, relative to that community. Whatever the debate about data centres and resources has been about, this study suggests it has been about the wrong resource.
The water figure also deserves a sense of scale. Four hundred and nineteen acre-feet is about 136 million gallons a year, or roughly 374,000 gallons a day for a facility drawing 100 megawatts. Per unit of compute that is a modest number, and it is the kind of efficiency that closed-loop and air-cooled designs have made routine in new builds. The electricity figure has no equivalent escape: a 100-megawatt facility draws 100 megawatts continuously, and 876 gigawatt-hours a year is not a figure a substation absorbs without having been designed for it.
Why water attracts the attention
The finding is counterintuitive enough to need explaining, because the public conversation has fixed on water as the emblem of data centre strain and this study says that emphasis is misplaced.
Three things explain the gap. Water is visible: a cooling plume, a retention pond and a drought headline all photograph well, while a substation does not. Water is local in an obvious way: a facility drawing from an aquifer affects the people above that aquifer and nobody else, which makes it a planning question a community can act on. And water is emotionally immediate in a way that megawatt-hours are not, particularly in a state that has spent a decade arguing about drought.
Electricity is harder to see and easier to defer. A grid connection is an engineering negotiation with a utility rather than a public argument with a neighbour, and the capacity either exists or requires construction. That makes the power question more consequential and less discussed, which is precisely the combination that produces surprises later.
The indirect water the study excludes
The release states one limitation in the water estimate that is worth carrying forward, because it cuts in an interesting direction. The 419 acre-feet figure covers direct consumption only, and excludes the water used in generating the electricity the facility draws and in manufacturing the chips it runs on.
That exclusion matters because the study’s own finding makes it large. If the facility consumes six times a city’s electricity, the water embedded in producing that electricity is not a rounding error. Thermal generation consumes water for cooling; the figure varies enormously by source and by plant, but it is not zero for any of them. A full accounting would raise the water number substantially, and the study is honest that it has not done that accounting.
This is the sort of caveat that strengthens a paper rather than weakening it. Stating what a figure excludes lets a reader see the boundary of the claim, and it points at the next piece of work: the indirect water attributable to a 100-megawatt load would be a genuinely useful follow-up.
What six times a city’s power actually means
The electricity finding deserves a paragraph on what it implies operationally, because a multiple of a city’s consumption is vivid and not immediately actionable.
A 100-megawatt IT load running continuously draws 876 gigawatt-hours a year. Distribution utilities plan substations and transmission connections years ahead, and a single customer requiring that much capacity is not an incremental addition to a load forecast. It is a project in its own right, with its own environmental review, its own land requirement for the substation, and its own queue position behind whatever else the utility has already committed to build.
That queue is where the practical constraint sits. A community that welcomes a data centre may still wait years for the interconnection, and a developer that needs power sooner may propose building generation on site, which brings its own land use, emissions and permitting questions. The study notes that the requirement could call for significant new electrical infrastructure or a separate power solution, and those are two very different futures for the neighbourhood.
The fiscal question is the one that decides
The third strand of the paper is the one a local official will read first, and the paper frames it as a question rather than an answer.
A data centre employs relatively few permanent workers and places limited additional demand on schools, parks and police. That combination is unusual: most large developments bring residents or employees who need services, and a facility that generates tax revenue without proportional service demand is fiscally attractive on its face. The release makes exactly that point about a well-structured project.
The condition attached to it is the whole argument. If infrastructure and utility costs shift to the public, the financial impact moves in the opposite direction. A facility that needs a new substation, a transmission upgrade and a water main extension, paid for by ratepayers rather than the developer, can cost a community more than it contributes. The difference between those two outcomes is a contract, not a technology.
That is why the paper’s three questions are the right three. What will a data centre require from the community, who will pay for those demands, and what will the community receive in return. None of them is answerable from a press release, and all of them are answerable before a council approves a project.
What this study is, and what it is not
The provenance deserves stating plainly because it bounds the conclusions. Jagger Mattox, an intern and UC Berkeley student, wrote the white paper at the agency’s request. It models a hypothetical facility rather than an actual proposal. It uses the City of Banning as a comparison because comparable data was readily available, and the release is explicit that the choice reflects data availability rather than any current or potential project.
Those are real limits. A single hypothetical facility at one size, compared against one city, without sensitivity analysis across different cooling designs, power sources or facility scales, is an illustration rather than a general finding. An intern’s work is not peer reviewed, and the release does not claim it is.
What the study does have is method transparency and institutional caution. The author names the industry data behind his estimates, states what the estimate excludes, explains why the comparison city was chosen, and the agency publishing it explicitly declines to take a position on development. For a document intended to start a local conversation rather than settle one, those are the right instincts, and they exceed what many consultant reports with a commercial interest manage.

What would make the analysis general
Three extensions would turn a local illustration into something other communities could use directly.
A sensitivity range across cooling technologies is the first. Water consumption per megawatt varies by an order of magnitude between evaporative and closed-loop designs, and a reader in a different climate needs to know which end of that range applies to them.
The indirect water figure is the second, for the reasons above. It would also settle whether the study’s headline comparison survives a full accounting, which is the most important open question it raises about itself.
And a worked fiscal example is the third. The paper states that cost shifting reverses the economics without modelling it. Two scenarios, one where the developer funds the interconnection and one where ratepayers do, would show a council exactly what is at stake in the negotiation.
What to watch
Whether the agency publishes a second edition with the indirect water and a sensitivity range, because that would make the analysis reusable by other water suppliers facing the same question. Whether other State Water Contractors commission similar work, since a shared methodology across agencies would be more useful than a single local study. And whether the ratio holds in jurisdictions where water is scarcer than in western Riverside County, because five per cent of a city’s supply in a wet year is a different proposition from five per cent in a basin already overdrafted, and the finding that power matters more may be the more portable half of the conclusion.
That portability is the real test of the work: a water supplier in a desert and one in a temperate valley will weight the two demands differently, and a methodology that lets each run the comparison against its own baseline would be more valuable than any single result.

