AI, Data Centres and Water

AI data centres are sometimes described as though every new server must be connected to a permanently running tap. If that were true, concern about rapid AI growth would be entirely understandable.

To evaluate the claim, it is useful to look at the leading AI provider: OpenAI. Its Stargate programme is building new infrastructure specifically to train and run advanced AI models. What does that additional capacity actually mean for water use?

A closer look at OpenAI’s Abilene campus

OpenAI’s flagship Stargate campus in Abilene, Texas, provides an unusually clear example. Unlike older facilities that may use evaporative cooling towers, Abilene uses a closed-loop, non-evaporative liquid cooling system.

The principle is straightforward. Water moves through sealed pipes, carries heat away from the computing equipment, passes through air-cooled chillers and circulates back around. The water is reused rather than continually evaporated and replaced. It is closer to an enormous sealed central-heating system than to an enormous tap left running.

The actual water figures

Each of the campus’s eight planned buildings needs about one million gallons for its initial fill—roughly two Olympic-sized swimming pools. That sounds substantial until the crucial distinction is made: it is a one-off fill, not annual consumption.

Crusoe, the company constructing the campus, expects routine cooling-system maintenance to require approximately 50,000 gallons per building each year. Across all eight buildings, that is around 400,000 gallons annually—roughly the water use of four average US households.

OpenAI provides another useful comparison: the water used at the Abilene site in a year is expected to be around half of what the City of Abilene uses in a single day.

The wider Stargate approach

OpenAI says closed-loop or similarly low-water cooling designs are being used across its Stargate AI campuses. These sites are being developed specifically because demand for AI is increasing, making them a useful measure of the additional infrastructure created by AI—not historic water use across every kind of data centre.

This does not make the facilities literally water-free. Water is needed for initial filling, maintenance and ordinary building use. But it shows that very large additions to AI computing capacity do not automatically require very large ongoing withdrawals from public water supplies.

What this means for therapists

The popular story often jumps from “AI requires data centres” to “AI use will create water shortages”. OpenAI’s new infrastructure shows that the second statement does not follow from the first. The decisive factor is how a facility is engineered.

For therapists, the conclusion should be reassuring. Using an AI tool to reduce administration, improve clinical documentation and preserve more attention for clients is not equivalent to choosing an environmentally reckless infrastructure model. The leading AI provider is demonstrating that advanced computing and extremely low ongoing cooling-water use can coexist.

Scrutiny is healthy, and infrastructure providers should be held to the standards they publicly set. But the evidence from OpenAI’s new AI campuses does not support the idea that growing AI use must place enormous pressure on household water supplies. That outcome is not inherent to AI; it is largely a design choice.