IEA report brings AI’s environmental impact into sharp focus

James Talbot and Michael Pocock of Pinsent Masons were commenting after the International Energy Agency (IEA) warned that a surge in the use of AI over the coming years could place pressure on energy systems and lead to an increase in carbon emissions. This, the IEA said, is because of the increase in the number of data centres that would need to be operated and how they would be powered.

According to the baseline assumption adopted by the IEA in its new report on energy and AI (304-page / 7.7MB PDF), global electricity consumption for data centres will double by 2030 and represent just under 3% of total global electricity consumption in 2030. In an alternative scenario the IEA has considered, where there is a more resilient supply chain and greater flexibility, in data centre location, powering and operations, there would be faster data centre deployment and stronger growth in AI adoption – and data centres would account for 4.4% of global electricity demand by 2035.

The anticipated growth in data centre development would, the IEA said, just be one factor in driving overall growth in total global electricity consumption over the period, with growth in industrial output, the electrification of transport, and space cooling each being responsible for a greater proportion of the overall growth under the IEA’s projections.

The IEA said that wind and solar power, as well as nuclear, are expected to play an increasing role in meeting the energy demands of new data centres in the years ahead. However, it said that the greater number of data centres built in response to demand for AI over the next decade would also lead to increased use of fossil fuels, like coal and natural gas. This, it said, would result in a knock-on increase in carbon emissions stemming from the operation of data centres and buck the general trend expected in other sectors, where carbon emissions are expected to reduce over time.

According to the IEA, however, AI has the potential to drive efficiencies and innovation across sectors that can more than offset the increased emissions expected to come from increased use of the technology.

It said, for example, that AI can help optimise manufacturing processes for their energy needs, make fossil fuel-powered plants more efficient to operate, and achieve efficiencies in the way buildings and heated and cooled. However, it cited access to data, the absence of digital infrastructure and skills, regulatory and security restrictions, and social or cultural obstacles, as potential barriers to use of AI in those ways.

James Talbot of Pinsent Masons, an expert in technology law, said: “The IEA’s message is clear: AI itself has the potential to be part of the solution in addressing the increasing environmental impact of its ever-increasing adoption. Demand for compute results in overheating of underlying hardware, but such compute can be used to identify and manage efficiencies to manage such overheating as well. In work with clients, Pinsent Masons has already seen how AI can speedily interrogate business processes and identify efficiencies more effectively than human or more dated computing methods and the potential for compute hungry processes being applied with a view to lessen their own impact is cause for optimism when considering how the technology can be deployed in practice and how negative trade-offs will be managed.”

“While businesses have been rushing to explore how they might benefit from using AI, there is a friction with their responsibilities and commitments around ESG which, as awareness of AI’s environmental impact increases, may act as a potential barrier to innovation. This is because it may become impractical or prohibitively expensive to power AI over time and do so without conflicting with environmental targets such as net zero,” he said.

“For businesses, the globalisation of data processing via cloud computing does not mean they can simply offshore their sustainability obligations. The conflict between the desire to use AI to its full potential while meeting ESG requirements can be managed if they develop effective decarbonisation and other environmental strategies,” Talbot said.

According to Pocock, the IEA report highlighted other factors that developers and investors behind data centre and energy infrastructure projects need to consider when thinking about how to respond to growth in the use of AI.

The report detailed how localised pressures on energy systems can build as a result of the relative concentration of data centre development around cities or existing energy infrastructure.

Analysis by the IEA also envisages around 20% of all data centre projects in the pipeline for construction by 2030 globally being delayed as a result of constraints with electricity grids. The IEA called for data centre connection applications to be rationalised by grid operators and advised data centre developers to “maximise the buildout of data centres in areas of high power and grid availability and explore strategies to incentivise their operational flexibility”.

Pocock said: “Co-locating data centres next to available high power supplies is crucial but it is not a decision which can be made in isolation. Planning and environmental considerations also need to be carefully considered, including the likelihood and timeframe for securing requisite consents and permissions. There can often be long lead in times in these processes, which is something the IEA report also recognises. This is an area which, in the UK, the government has been keen to address, and it has already updated planning policies with a view to more easily facilitating development of gigafactories, data centres and digital infrastructure.”

“The recent creation of AI growth zones should also support this drive as well as the mooted planning reforms to bring these types of development within the Nationally Significant Infrastructure Projects (NSIP) regime, with changes anticipated to come forward this summer. The NSIP regime could be a game changer in terms of speeding up and giving more certainty to the overall process, because it is a more comprehensive consenting process which is ideal for dealing with projects where co-location is a key factor – which is the case with data centres and high energy supply requirements,” he said.

Pocock added: “For now there is no magic wand to wave through data centre projects from a planning and environmental perspective and there remain risks of delay and uncertainty in the processes, but the direction of travel is positive and the consenting system will hopefully become more streamlined going forwards.”

The UK government has established an AI Energy Council to consider how the energy system in Britain can support growth in AI and increased demand for computing power. The body, which includes representatives from across the energy and technology sectors, met for the first time earlier this week.