RSSAmplifier

Rod McLaren: Words that work · Oct 13, 2025

UK data centres and AI hosting will need a lot of energy

0
Sign in to vote or save

holdfastprojects.com

13 October 2025

What are grid operators, UK Gov and industry saying about future electricity demand for data centre hosting? Spoiler: everyone says it’s growing a lot.

NESO, the grid operator

In March 2022, the UKs National Energy System Operator (NESO) thought data centre energy demand would grow from 3.6 TWh/year in 2020 to ~13 - 35 TWh/year by 2050. (To give a broad comparison, UK as a whole generated 228 TWh in 2024, a number that’s down on 2022 and 2023.)

By November 2024, NESOs Clean Power 2030 plan increased the projected demand to 22 TWh in 2030 and 62 TWh in 2050.

And in July 2025, NESOs 2050 projections are at 30 TWh/year in their lowest growth falling behind” scenario and 71 TWh/year in their highest growth electric engagement” scenario. Detail in Pathways to Net Zero v4 July 2025. So the projected figure has doubled since 2022. 1

UK Gov and industry analysis

DSITs Compute roadmap July 2025 forecasts that the UK will need at least 6 GW of AI-capable data centre capacity by 2030. (To give a broad comparison, UK as a whole was generating 31 GW at the time of writing, and over the last year averaged 27 GW.)

Hold on, what’s the difference between GW and TWh?

Gigawatts (GW) are a measure of power, or the total of amount of energy being supplied to the country right now. If the energy system was like a bath, GW is how much the bath tap is turned on.

Terawatt hours (TWh) are a measure of the total amount of energy over a period of time. TWh is like how full a bath is after it has been filling with water for a while. The typical household in England, Scotland and Wales uses 2,700 kWh of electricity and 11,500 kWh of gas in a year.

And to turn GW of power back into the TWh of energy, allow me to handwave a bit: if you multiply 6 GW by 8,760 hours, and a data centre utilisation rate of 70% 2, you get a number of 37-ish TWh a year.

And the House of Commons Library Data centres: planning policy, sustainability, and resilience briefing August 2025 was slightly less ambitious, citing the Clean Power 2030 numbers and projecting 3.3 - 6.3 GW capacity.

techUK’s report from Nov 2024 Foundations For The Future: How Data Centres Can Supercharge UK Economic Growth suggests that demand for data centres is set to increase at a much faster rate than it ever has before, with projections of future growth at between 10-20% a year”.

The wider global picture for context

EU/Ember Energy: Europe’s AI ambitions at risk of gridlock and Grids for data centres: ambitious grid planning can win Europe’s AI race.

Texas/Powwr: 7.6GW today, 78GW by 2031: The largest percentage of projected demand growth comes from the state’s data center sector, which is expected to rise to 78GW by 2031.”

Global/IEA: Energy supply for AI: 1,000 TWh in 2030 and 1,300 TWh in 2035 in the Base Case”.

Global/Bloomberg: BloombergNEF expects global electricity demand from data centers to rise to 1,200 terawatt-hours by 2035 and 3,700 terawatt-hours by 2050 - 2025 New Energy Outlook.

UK announcements September 2025

Some recent headlines:

To get a sense of the current interest in data centres, all of those announcements were in September 2025. Just one month!

So it’s clear AI hosting is driving a fair bit of the growth in data centre energy demand, though it won’t be driving all of it because some of it goes to non-AI services like YouTube and that timesheet system they make you use.

And obviously there’s a big difference (and gap in time) between announcements and deployments, but even so it makes me wonder if government projections are too conservative.


  1. For comparison, 2023 total UK energy use was 263 TWh, according to the Clean Power 2030 Annex 1 Electricity demand and supply analysis, and which also predicts a 5-fold growth in data centre electricity demand from today out to 2030”.↩︎

  2. Data centre utilisation rates: here’s Data Center Frontier citing the 2024 United States Data Center Energy Usage Report. It says AI training is ~80% and inference is ~40%, and you might say inference is a bigger use case than training, and conclude ~50%? However that piece also says AI servers operate at 80-90% utilisation, with non-AI use cases lower. So I’m calling it 70%, with a shrug.↩︎

Read the original on holdfastprojects.com

Comments

Nothing yet. Say the first thing.

    Sign in to join the conversation.