The Myth That AI Will Boil the Oceans

Every wave of AI data center announcements brings a matching wave of dread. The latest, SoftBank's €75 billion plan for France, was met in some quarters with a familiar warning: that the AI build-out will drain national grids, spike household electricity bills and consume water by the reservoir until the planet pays the price. The fear is not baseless. In its loudest form, though, it is considerably overstated.

What the numbers actually show

Start with electricity, because that is the real story. The International Energy Agency projects global data center power use roughly doubling, from about 415 terawatt-hours in 2024 to around 945 terawatt-hours by 2030. Goldman Sachs Research expects a 165 percent rise in data center power demand over a similar window. These are large numbers and they are growing fast. They are not, however, the whole grid. Even at 945 terawatt-hours, data centers would account for a low single-digit share of global electricity use, well behind heating, heavy industry and transport.

Water draws similar alarm and deserves similar care. United States data centers consumed an estimated 17.5 billion gallons of water directly in 2023, according to figures cited by IEEE Spectrum. That sounds enormous until you set it beside agriculture, which uses orders of magnitude more, or against the water embedded in everyday goods. The phrase boiling the oceans is rhetoric, not measurement.

Where the worry is justified

None of this means the build-out is harmless, and the honest version of the story is local rather than planetary. AI capacity clusters in a handful of places, and those places feel real strain. Analysts expect Virginia's data centers to claim somewhere between 41 and 59 percent of the state's electricity by 2030. A community sitting next to a new gigawatt campus can face genuine competition for power and water, even as the global averages stay modest. A recent arXiv study makes exactly this point, warning that concentrated siting drives regional grid stress that national figures hide.

On the bills people actually pay, the picture is muddier than the fear suggests. Consumer Reports notes that evidence linking data centers to higher household electricity rates is thin, and that in some markets large new customers have helped spread fixed grid costs and nudged rates down. In others, ratepayers do appear to be shouldering part of the infrastructure cost. It depends on the regulator, the utility and the deal.

The honest reading

So the doomsday framing fails, but the complacent one fails too. AI's energy appetite is real, rising and worth watching, especially where the machines crowd together. What it is not is an extinction-level drain on the world's water and power. The useful response is the boring one: site these centers where clean power is plentiful, hold utilities and regulators to transparent accounting, and keep measuring. France's nuclear-heavy grid, the reason SoftBank chose it, is a hint that the industry knows this already. The oceans are not boiling. The grid in your own county still deserves a close eye.