ComputerWeekly

Government science advisers: ‘No framework for datacentre demand’


The UK has no agreed method for working out how much onshore datacentre capacity it needs, and delays in securing grid connections are a key factor in where new sites go.

That’s according to the Government Office for Science, which published a report of a meeting chaired by chief scientific adviser Angela McLean in February. The report represents “the combined views of roundtable participants” and is explicitly “not a statement of government policy” but does provide useful insight into the datacentre debate at national planning level.

According to the report, the group found “no obvious frameworks to determine how much onshore datacentre capacity the UK will need in the future”, and said demand was “very hard to estimate”. Sizing decisions should start from what UK capacity is essential – where offshore provision is not an option, it said – and from tasks datacentres perform across the economy, rather than from sector-by-sector demand.

“Water is not a major constraint on datacentre siting in the UK,” it said. The notes state that datacentres can be cooled without water using closed-loop systems and ventilation, given the UK’s climate, and that Amazon Web Services’ facilities reportedly run waterless for around 97% of the year. It notes the trade-off is power, and using less water means using more electricity.

The roundtable recommended the UK build a national interactive map of every datacentre, plotting sites alongside fibre routes, power lines and other infrastructure. The aim is to give developers, planners and communities a shared picture of where capacity sits, from hyperscale campuses down to small edge sites.

The notes draw a line between workloads that need to stay in the UK and those that do not. Regulated public services and industries under data protection rules such as GDPR generally require onshore sites, it said, and there are “strong arguments” for onshore capacity to serve national security and defence. Ultra-low-latency tasks – high-frequency trading and growing financial artificial intelligence (AI) use among them – also tie capacity to specific locations.

Much of the rest can go elsewhere, says the report. The notes say many workloads tolerate some latency or are not sensitive to data location, so they can use offshore facilities or, in future, centralised mega datacentres. They cite Ukraine’s experience of hostile actors as a reason to favour offshore or distributed capacity for resilience.

Grid connections are cited as a pinch point. The notes say delays in establishing a grid connection are a key factor in siting and “risk undermining the UK’s digital growth objectives”. They float removing the government requirement for new power generation to be connected to the grid as a way to cut delays and improve national resilience.

Planning is another gap identified. There is currently no national planning guidance for datacentre siting, and the notes point to strategies in the Netherlands and Saudi Arabia as references. They argue clearer guidance would give developers and communities more certainty, and that public acceptance depends on people seeing a national need, clear local benefits and developers they trust.

The timing coincides with lively debate decisions on planning, grid connections and datacentre location. 

The debate nationally lacks reliable figures. Computer Weekly analysis published this week has highlighted the dearth of reliable data around datacentre planning, in particular with regard to electricity draw, water use, and even the numbers of datacentres in existence and in the pipeline.

Computer Weekly has spent several months attempting to build reliable estimates. It has drawn on the Barbour ABI construction database and Energy Performance Certificate registers and found 88 verified existing UK datacentre sites with a combined capacity of about 1.6GW, against a committed pipeline of 173 schemes and about 14.6GW. The recommendation for a national map echoes that work. The numbers are contested because no shared, verified inventory exists.

Demand uncertainty runs through the meeting notes. The roundtable said future capacity would be variable and geographic, would depend on balance between small firms and large consumers, and would hinge on the pace of AI adoption – which itself depends on domestic datacentre capacity. It acknowledged that over-provisioning may be necessary to absorb that uncertainty, and that AI and non-AI services are already blurring as cloud sites evolve into combined general and AI facilities.

It acknowledged that workloads differ sharply. Training AI models is generally less latency-sensitive than running them, and for some large language models, demand has shifted from training to inference, which the notes say carries a significantly greater environmental impact at scale. Image processing for agriculture and transport is among the fastest-growing uses and typically needs localised capacity because of the volume of data involved.

Nine people took part in the roundtable. Alongside Angela McLean, the roundtable included Clive Richardson (Amazon Web Services), Dawn Childs (Pure Data Centres Group), Dimitra Simeonidou (University of Bristol), Eve Schooler (University of Oxford), John Booth (BCS, The Chartered Institute for IT), Luisa Cardani (TechUK), Patrick Devine-Wright (Exeter University) and Tim Watson (Loughborough University).

No grid or energy system operator sat at the table, and no water or planning regulator, even though grid delays and planning are two of the report’s central themes. The notes also carry the caution that they are advice, not a decision. The government has not committed to the map, to planning guidance or to changing grid connection rules.

The eight-month gap between meeting and publication points to a slow clearance loop in a fast-moving area. In that time, the UK pressed ahead with AI growth zones and datacentre planning decisions against capacity and demand figures that, by the advisers’ own account, nobody has a framework to test.

Waste heat and climate add further constraints. The notes say reuse needs datacentres sited near infrastructure that can use the heat, and that opportunities are limited without district heat networks and when sites sit on former industrial or greenfield land. Possible use cases include laundry, food preparation and greenhouses. Siting must also adapt to flooding, temperature and water availability.



Source link