America’s backlash against data centres is a mistake

Data centres provide the computing power to train and run AI models, yet American voters increasingly oppose building them. The backlash is fuelled by misunderstanding — and politicians pushing moratoriums and bans are making a big mistake.

A political backlash built on myths

The opposition spans the nationalist right and the environmentalist left. Such is the backlash that the infrastructure has become one of the defining issues of this year’s midterm elections.

Politicians are racing to outdo one another with restrictions, moratoriums and bans on building. Voters and politicians alike are making a big mistake.

Water fears do not hold up

One popular myth is that data centres guzzle more water than Agastya — an argument helped along by the grossly inaccurate calculations of a bestselling book. In reality, a mid-sized data centre uses roughly as much water as two golf courses and considerably less if it recycles its water, as many now do.

And there is no reason to single out AI: just about everything in modern life, from streaming Netflix to running a dishwasher, consumes water at levels that can be portrayed as wasteful by hair-shirt environmentalists. Google has calculated that the median Gemini text prompt consumes about five drops, meaning even tens of thousands of AI queries would consume less water than a minute in the shower.

Electricity demand has not raised bills

Concerns over electricity use are more numerate: data centres really do need a striking amount of power. However, there is little evidence so far that they have raised energy prices. In fact, by increasing demand they are spreading the fixed costs of infrastructure among more kilowatt-hours sold.

In recent weeks a number of large utilities have linked data centres to reductions in bills. Many data centres are also helping finance new sources of generation.

Eventually the use of AI could push up electricity bills. But the answer to competing demands on resources — whether water, electricity or something else — is to price them according to their scarcity, thereby encouraging more provision and higher living standards over time. It is not to ration inputs according to the arbitrary judgment of scolds.

Where pricing is difficult to calculate — to account for noise pollution, say — planning judgements are necessary, and except for needing to be near the grid there is no good reason to put data centres next to homes. In any case, even the noise worry is overstated: many viral videos of supposed AI data centre noise are really Bitcoin mines, not state-of-the-art hyperscaler facilities.

The benefits are concrete

Set against the illusory costs of data centres are the tangible benefits. At a national level, data centres power a technology for which there is rapidly growing demand and on whose success future economic growth and national defence may depend.

More prosaically, data centres pay huge tax bills, helping fund schools, libraries and police stations. In Loudoun County, Virginia, a tax on computer equipment inside its more than 250 centres is expected to supply 40% of the county’s total tax revenue next year. The windfall has allowed the county to cut residents’ property taxes by about 30%.

Blocking building would help China

There is a respectable argument for slowing down AI research, but it rests on the danger to all humanity of the technology being abused or going rogue, not the local harms of data centres.

America unilaterally making it harder to build AI infrastructure will simply hand the advantage to China, whose authoritarian leaders can put data centres wherever they choose. For years, America’s populists complained that its physical economy had been hollowed out, giving China an economic edge and leaving Uncle Sam without the industries it would need in wartime.

By opposing data centres, they risk causing those exact harms.

Report a correction