Keir Starmer this week launched a plan to bring a 20-fold increase in the amount of artificial intelligence (AI) computing power under public control by 2030.
But the race to build more electricity hungry AI datacentres over the next five years appears to work against another government target: to plug in enough low-carbon electricity projects to create a clean power system by the same date.
The green goal was already considered to be “at the outer limit of what’s achievable” by Fintan Slye, the chief executive of the National Energy System Operator (Neso), which is responsible for delivering the net zero target. But the enormous energy appetite of an AI boom has raised concerns that the government may end up derailing the clean power pledge just months after making it one of its key election promises.
Was the government on track to meet its green power goal?
To meet the 2030 target based on current power use, the government believes Britain needs to double its onshore wind, triple its solar power and quadruple its offshore wind capabilities. The UK is also considering hefty financing deals to support new nuclear projects.
Britain shut the last of its coal power plants last year and the government has promised that by the end of the decade it will use gas plants only sparingly to create a 95% carbon-free electricity system by 2030. However, gas is still relied on to plug the gaps when the wind does not blow and the sun does not shine. That’s what happened last week when the grid operator was forced to pay a pair of gas plants rates up to 100 times higher than normal market prices to generate electricity as temperatures fell below freezing and the UK’s green power generation slumped.
“All our modelling suggests that on the current trajectory the government’s [2030] target will be extremely stretching,” says Kate Mulvany, a principal consultant at Cornwall Insight, one of the UK’s most respected energy advisories. “We cannot see how these targets can be met with current schemes and policies.”
What impact would increasing AI use have?
Building and running the servers that form the central nervous system of AI is hugely energy-intensive, with ever more needed to train and run increasingly complex AI models. Globally, the electricity consumed by datacentres is expected to match the entire power consumption of Japan within the next two years, according to the International Energy Agency.
Mulvany says that always-on energy is needed to support datacentres, to match their continuous running hours. Emerging energy storage technologies could make it possible for wind and solar power to play this role in the future, she says, but in the near-term most markets with a large density of datacentres end up using more gas power: “It would absolutely pile pressure on already very difficult clean power targets.”
Will the grid infrastructure be able to cope with demand?
The government’s plans include the setting up of “AI growth zones” that will benefit from fast-track planning and infrastructure upgrades to accelerate the rollout of “clean” energy needed to power its datacentres. This is likely to include grid upgrades to the areas hosting these datacentres in order to handle the vast volumes of power that will need to be generated and consumed nearby.
However, the grid already needs to carry out an unprecedented level of energy infrastructure upgrades across the country to connect new energy projects to the grid and to plug in new housing developments and newly electrified factories and businesses.
The chief executive of the National Grid, John Pettigrew, has estimated that Britain needs to roll out five times as many pylons and underground lines in the next five years than has been achieved in the past 30 years to achieve this, requiring investments in the UK’s supply chains and training the necessary workers. AI would only add to this load.
“There are big competing priorities here, and they are all energy-intensive,” one industry source said.
Is the government ready for the challenge?
Starmer’s announcement included plans for a new AI energy council, which will be co-chaired by the technology secretary, Peter Kyle, and the energy secretary, Ed Miliband, to secure electricity generation for the scheme. The council plans to accelerate investment in low-carbon energy sources for datacentres, including renewables and small nuclear reactors.
Mulvany argues that ministers need to be honest about the tradeoff involved in focusing on AI to solve growth concerns. “Prioritising AI growth zones does imply a deprioritisation of something else, such as electrifying heavy industry or housebuilding,” she says. “It might be decided that AI is so important to the greater good that it should be pursued, but then we need to acknowledge that and have a clear idea of our national priorities – or risk underdelivering on multiple fronts.”
A Neso spokesperson said its 2030 clean power plans had already assumed a sharp fourfold increase in the energy demand required by AI datacentres by the end of the decade, but declined to clarify whether that included the government’s planned 20-fold increase in AI under public control by the end of the decade. The spokesperson added that the emergence of AI could also help reduce energy demand by making power systems more efficient, or by making use of the heat generated by datacentres to warm nearby homes.
“We are working closely with the government on how to best leverage the opportunity datacentres can bring from a technology and system management perspective,” they said.
A government spokesperson said the AI energy council would help the government to understand the energy demands “which will deliver on our new blueprint for AI”, including the use of “a range of clean-energy technologies such as small modular reactors that we can tap into”.
“We will ensure clean, renewable energy solutions are at the heart of powering AI, and datacentre operators in the UK are already committed to implementing energy-efficient measures under a climate change agreement,” he said.
