The Data Centre Problem Bigger Than Water or Power: the Money

"I still think that the actual scale of those agreements tends to be pretty paltry compared to the scale of the data centre itself," he said.

Communities often don't decide where that money goes, either. A developer might prefer to "throw a few million dollars into the fire department", Green said, rather than tackle noise reduction, which residents may care about far more.

How risky is the money behind the boom?

Here's where it gets uncomfortable. In June, economists Jonathan and Jessica Wachter calculated that annual spending on AI infrastructure could exceed £750bn ($1tr) by 2027. For the bet to pay off, AI must generate equally colossal returns through services, new products or productivity gains.

Assuming investors expect a 15 per cent return, the Wachters calculated that these data centres would need to become 2.7 times as productive for the numbers to work. If the gains don't materialise, they warned, the firms "risk bankruptcy".

And much of this is borrowed money, not current profits. If AI disappoints, companies would face both expensive, underperforming data centres and heavy debts. If enough hit that wall at once, the shock could ripple through the wider economy.

There's a precedent. Economic historian Liaquat Ahamed has compared the AI boom to the great US railroad bubble of the 19th century, when borrowed money financed a frenzy that doubled America's rail network in just seven years after the Civil War. When the bubble burst in 1873, it helped drag the entire US economy into the Long Depression. Railroad construction then averaged around 2.5 per cent of the US economy; AI infrastructure spending is now expected to reach 2.8 per cent.

The company behind that planned data centre down your road may go bust before a single brick is laid. But an abandoned building site could be the least of anyone's worries.

Will we even need all these data centres?

Nobody knows how much computing power AI will need, and the hardware is moving fast. Chips that run AI have roughly doubled in performance every two years.

"The chips that Nvidia put out on laptops today are more capable than the ones four years ago, or two years ago, that required 10 times as much energy and [went] into the big servers," said Prof Carl Boettiger, an associate professor at the University of California, Berkeley. AI models themselves are shrinking too, he notes, doing similar work with far less computing power.

Yet progress cuts both ways. Cheaper AI means we'll likely use much more of it, and total demand for computing power could grow fast enough that we need more data centres, not fewer.

Green agrees we'll use more AI — just not necessarily the giant kind. Much of today's building boom is designed for the most powerful models, but smaller, cheaper models will soon handle everyday tasks nearly as well.

"Why pay such a premium for the most sophisticated model when you can, for 10 per cent of the cost, get 80 per cent of the performance?" he asks.

If enough people reach that conclusion, many planned data centres could end up as 'stranded assets' — humming away, half-empty, long after the hype has died. So what does this change for the people living next door? Possibly nothing yet. The promised tax money may arrive, or the developer may collapse first. Either way, the water and power bills are the smaller gamble.