The Thirsty Cathedral

The building has no windows and no sign. It sits at the end of a road that was widened two years ago at public expense, behind a berm planted with fast-growing evergreens, and from the nearest residential street you cannot see it at all. You can, however, hear it: a low, continuous, mechanical sigh that residents describe as being like a distant motorway that never has less traffic.

What you are hearing is cooling.

The facility occupies roughly forty thousand square metres and, according to its grid interconnection agreement, has a contracted capacity of ninety megawatts — comparable to the average draw of a city of eighty thousand people. Almost none of that electricity does anything a member of the public would recognise as work. It runs processors, then removes the heat those processors produce — because a modern server rack concentrates more thermal energy into a cubic metre than any structure built before this century had to dissipate.

This is the part of the industry that does not appear in the marketing.

The numbers they publish, and the ones they do not

The sector's preferred metric is power usage effectiveness, or PUE: total facility energy divided by the energy delivered to the computing equipment itself. A PUE of 2.0 means half your power is going to cooling and overheads. The best modern facilities report figures between 1.1 and 1.2, and the improvement over two decades has been genuine and substantial.

It is also, examined closely, a measure with a convenient blind spot. PUE is a ratio, not a quantity. A facility that doubles in size while holding its PUE constant has doubled its consumption and improved nothing, yet its headline efficiency number is unchanged. Industry communications lean heavily on the ratio and rather less on the total, and when I asked three operators for absolute annual consumption at named sites, two declined on grounds of commercial confidentiality and the third did not reply.

The figure that is genuinely hard to obtain is water.

Cooling a data centre by evaporation is thermodynamically efficient and cheap, which is why it is common. Warm water is passed over a large surface, some of it evaporates, and the heat departs with the vapour. The water that evaporates is consumed — not returned to the river or the aquifer, but transferred to the atmosphere. A large evaporatively cooled facility can consume in the region of a million to two million litres a day during hot weather, which is precisely when the surrounding area is least able to spare it.

Operators have begun publishing water usage effectiveness alongside PUE, and several have committed to becoming "water positive" by a stated date. Read the definitions carefully. In most such schemes, replenishment need not occur in the watershed where the consumption took place. A project restoring wetland several hundred kilometres away may be counted against water evaporated here. Whether that constitutes a meaningful commitment depends entirely on what one thinks water scarcity is a problem about.

What the planning file shows

Planning applications are public, and they are frequently more informative than press releases, because the applicant is addressing regulators rather than customers.

The file for this site runs to eleven hundred pages. It contains the noise assessment, which concludes that the increase over existing background levels at the nearest dwellings will be "not significant" — a technical term with a specific threshold, carrying no implication that the noise will be unnoticeable. It contains the backup generator specification: thirty-two diesel units, permitted to run for testing, and the source of the monthly rumble residents' groups have complained about since commissioning.

And it contains the economic case, which is the document that actually secured approval. Construction: around fourteen hundred jobs, temporary. Permanent operational staff: fifty-two. Annual local tax contribution: a figure large enough that the council's own submission describes it as transformative for the authority's finances.

That last point deserves to be stated without cynicism, because it is the crux of the matter. These facilities pay substantial local taxes and employ very few people. For a local authority with an ageing population and a deteriorating revenue base, a windowless building that generates seven-figure annual receipts and requires almost no municipal services is, on any conventional assessment, an excellent deal. The objection is rarely to the arithmetic. It is to what the arithmetic leaves out.

Where the constraint actually binds

Speak to grid engineers rather than campaigners and the emphasis shifts.

The binding constraint on this industry is no longer land, capital or chips. It is interconnection — the physical and administrative process of connecting a very large load to a transmission network that was designed, over decades, for a different distribution of demand. Queues for connection in the most sought-after regions now extend for years. One jurisdiction has effectively paused new connections in its capital region after the sector's share of national electricity consumption passed twenty per cent. Several utilities have begun requiring firm load commitments and substantial upfront payment for network reinforcement, which quietly transfers cost from ratepayers to developers, and which is probably the single most consequential policy development of the past three years.

An industry executive, speaking on condition that his employer not be identified, put the position more candidly than any of the official responses I received.

"Everyone talks about efficiency. Efficiency is not the issue. We are the most efficient we have ever been per unit of computation, and we are consuming more than we have ever consumed, because demand has grown faster than efficiency by a wide margin. That's not a scandal, it's arithmetic. What's a scandal is that we let local authorities approve these on the basis of a tax line without anyone at national level modelling what happens when forty of them land in the same grid region."

The honest ledger

None of this amounts to a case against data centres, and any report that pretends otherwise is not worth reading. The computation happening inside that building underpins hospital records, payment systems, logistics, scientific research and the ordinary functioning of nearly every institution the complainants themselves rely on. Nobody proposing that it stop has explained what should stop.

The case is narrower and harder to dismiss: an industry with unusual water requirements, unusual grid impact and unusually concentrated siting has been permitted to grow at unusual speed under a disclosure regime designed for warehouses. The remedy is not prohibition. It is publication — absolute consumption, watershed-level water use, regional grid impact, in standard units, on a schedule, audited.

The plant hums on. It will hum tonight, and in August, when the reservoir behind the evergreens is lower than anyone would like.