Data centres worsen water crisis

We’ve not had much growth here, or across much of the developed world – the US excepted – for a while, and what little we have seen has tended to arrive in forms that most of us don’t immediately see the benefits from.

Data centres, and AI in general, may not mean much in the way of growth. Goldman Sachs put out a fascinating paper a couple of months back, which City AM summarised: “Goldman Sachs has published a paper stating that AI’s productivity benefits and returns are likely to be significantly more limited than anticipated, while its power demands will be so great that utility companies will have to spend almost 40 per cent more in the next three years to keep up with demand.” 

Essentially we could be investing a very large amount of money, and committing very significant resources, to stuff that doesn’t actually do very much for the broader economy. This is a longstanding problem in economics, known as the Solow Paradox, after Robert Solow, who as far back as 1987 said: “How come I can see the IT revolution everywhere, except in the productivity statistics?”

Jammed

But big picture aside, the specifics on data centres look bad. These things produce woefully limited local economic benefits. 

The planned data centre in Teesside, to be operated by Blackstone and announced with great fanfare by Rachel Reeves, the British chancellor, last autumn, will cost $10bn. For that $10bn investment, it’s expected to create 4,000 jobs locally – or, in other words, a cost of about $2.5m for every job created.

This isn’t going to be public money but of course the data centre itself, which will be one of the largest in the world, will use a huge amount of public resource – electricity, and, as we’re all now finding out water. 

That means more grid connections for both, more pressure on the system for both – all for an investment that will create a tiny number of jobs relative to the amounts put in.

This shouldn’t surprise us – data centres are massively, wildly capital rather than labour intensive investments. In other words, to work they depend far more on the banks of servers jammed full of overheating silicon chips than they do on people being there to operate them. 

Power

And the profits, which have (until recently) been astronomically for the key, monopolistic operators, flow not only out of the local area but typically out of the country all together. 

The Digital Services Tax, introduced by the last government to try and claw back some of the money flooding out of the country to mostly US Big Tech, does see some of those profits returned to the public purse. 

But keep an eye on what happens to it in upcoming negotiations with the Trump administration over a trade deal – my guess is Trump will demand its removal in return for a US-UK trade deal.

None of this means we shouldn’t be investing in data centres and computing power of every sort. 

Yimby

The National Engineering Policy Centre has some excellent proposals on how to make data centres more environmentally sustainable, from top to bottom – taking in the problem of e-waste (Britain produces more per capita than any other country except Norway!), options of recycling water, reusing waste heat from data centres and so on. 

But we really do need a much more sophisticated conversation about the purpose of these things – one that the arrival of DeepSeek is forcing on all of us. If it’s possible to now build sophisticated large language models without the massive infrastructure and resource use US Big Tech insists on, there’s a range of possibilities here. 

We can start to think about democratising and spreading the technology, using smaller, simpler models on the sorts of task we know AI is good at – things like drug discovery, or localised environmental modelling.

What we’re getting instead, from the government, the industry, and with some noises off from the so-called yimby movement is the exact opposite of this.

This Author

Dr James Meadway is an economist and former political advisor. This article is based on a transcript of an episode of Meadway’s podcast, Macrodose.