This article was written for the Business section of the Herald as part of the newspaper’s series on hyperscale data centres. It lays out the implications of the vote and the changes to EIA requirements from Wednesday 16th September
This article was published in the Herald Scottish AI data centre pause: ‘We badly need this approach’
On Wednesday MSPs voted for a pause of up to 12 months on hyperscale data centre decisions until planning guidance is produced. This came after an announcement from the Scottish Government that morning that all data centres over 50MW would be required to have an Environmental Impact Assessment.
This is a very tangible change in policy and one that Local Authorities, communities and business will be mulling over what this means for them.
As a start, we haven’t come across anyone yet who thinks that hyperscale AI data centres shouldn’t have EIAs. Hyperscale AI data centres are not like our existing data centre infrastructure which runs our economy, cloud computing etc. Developments under 50MW (which is still a very big data centre) will sit with the existing regime.
Large Language Model (LLM) workloads require huge amounts of energy and the proposals for Scotland are absolutely vast. Most of our existing data centres are between 1-5MW, with one at 24MW, but these behemoths are 200-1000MW. Even the Edinburgh University supercomputer, one of the biggest in the world and delivering vital machine learning AI research, is just 39MW. The environmental cost of this energy use would be huge. Combined with the locations they are being proposed and the space required, there are very significant environmental impacts. This is why the DCA, the trade body for UK Data Centres, wrote in a report published in July ‘The DCA also believes Environmental Impact Assessment should form a standard component of evaluation for all major data centre developments.’
The Direction issued by the Government on Wednesday brings hyperscale data centres in line with the requirements on developers of other extremely large and potentially environmentally-damaging developments.
Secondly: the pause in planning applications while guidance is produced allows the strategic approach referred to in the Programme for Government and for the planning guidance to be produced. As the Minister, Hannah Mary Goodlad said in the debate “this is a chance to get it right from the beginning”.
We badly need this strategic approach, though. The developments coming through planning will come with serious economic costs. Twelve of the sites in the planning system are on Green Belt sites or farmland, including at least two on grade 2 farmland, the best Scotland has. Two of the sites are on Brownfield sites (Ravenscraig and Greenock IBM site) but these had been allocated for 1450 homes between them. None of the sites in the planning system are on a priority list of potential sites drawn up by the UK Government in 2021 and updated in 2023.
The Hunterston proposal is a good example of the lack of strategic approach. It’s on agricultural land, just yards from a 900 year-old castle, when the Hunterston coal yards, which are on that priority list, sit only 200m to the north of the site.
We have heard concerns of housing developers both in Scotland and in England, at the impact that data centre energy demand has on them. The Edinburgh Gyle project would badly impact the attractiveness and desirability of the housing, leisure and evening economy developments being planned and built there. Three local councils in London – Ealing, Hillingdon and Hounslow – have had to halt or delay new housing developments given the pressure existing data centres are putting on the grid locally. And no one wants to energy prices to rise, as they have done everywhere that data centres proliferate.This is due to two factors: first, the expense of the large amounts of additional energy generating and transmission infrastructure required, and secondly increased reliance on gas which is a more expensive way of generating energy than renewables.
The insatiable demand of hyperscale data centres for components is impacting business in the US and globally through the increased costs of materials, especially GPU, the processors that AI data centres require for the heavy computer workloads, batteries, generation and grid components, as manufacturers shift to producing for AI data centres.
And this is before we have got to the lack of jobs they create. Not only do data centres employ, on US industry data, just one job per $17m spent, but the industry itself says the large language models that these data centres will power are likely to decimate employment, especially for young people. We have looked at international evidence here, and it’s clear that most data centre developers are hugely exaggerating the jobs they would create. An FoI to the UK Government showed that, of the estimates of jobs provided for the data centres in the AI growth zone, less than 10% of the figure were actually in the data centres.
It is difficult to see how our politicians can keep talking about potential benefits for jobs and the economy of data centres when the US tech bosses, who know the most about this technology, keep telling us how few employees will be needed, and how employment levels will plummet as a result of AI.
Building hyperscale AI datacentres is not a shortcut to producing a flourishing tech sector in Scotland. If Ministers really wanted to benefit Scotland’s tech sector, they would invest in our cutting edge university research in machine learning, and Scotland-based innovators, learning from the success of Edinburgh, rather than rolling out the red carpet for US big tech.