In April, the International Energy Agency published a report on the ever-growing links between energy and artificial intelligence (AI). It projected that, by 2030, electricity demand from data centres will double globally – and that AI will be the most significant driver of this. Current projections suggest only half of this demand will be met by renewable energy sources.

The greenhouse gas emissions associated with energy use by data centres have raised concerns because of their potential to exacerbate the impacts of the climate crisis. 

Such data centres ‘can be great gobblers of water because of the cooling processes they require,’ says Michael Showalter, an officer of the IBA Environment, Health and Safety Law Committee. They can also compete with local communities for energy and water. 

The UN Environment Programme (UNEP) highlights that, according to estimates, global demand for water resulting from AI may reach 4.2–6.6 billion cubic metres in 2027. For reference, this would represent over half of the UK’s annual water use in 2023.

Additionally, the computer chips integral to AI servers are often made from rare metals, while the UNEP notes that as AI scales, it’ll generate ‘higher volumes of electronic waste,' of which only a relatively small amount is currently recycled.  

In the US, this means that in practice they’re established within emerging technology hubs. For example, there’s a 25 square mile area in northern Virginia containing nearly 200 data centres – and that number is increasing. According to Aurora Energy Research, fast-rising demand for electricity from data centres in the northern part of the state could ‘put significant strain on the region’s capacity reserves and transmission grid.’ 

One of the challenges presented by these clusters of data centres, says Showalter, is that ‘you then end up loading an awful lot onto a power grid and a locality very fast, which means figuring out how to deal with that. So, either you build out the grid or you build your own power source and essentially create a cul-de-sac of your own energy use.’

Els Reynaers is Co-Chair of the IBA Environment, Health and Safety Law Committee. She highlights that, currently, the electricity demands of AI data centres also vary greatly depending on the jurisdiction. There’s a well-developed focus on constructing such facilities in the US, and even in India, ‘there is a significant growth forecast,’ she says.

A global challenge in terms of tackling the potential environmental impact of the operation of AI data centres is the relative lack of targeted and binding legislation in this area, says Reynaers, who’s also a partner at MV Kini in Mumbai. ‘It is not particularly regulated in a mandatory manner in my jurisdiction or even in the US at a federal level, although some states have adopted more tailored energy efficiency standards,’ she says. Only a handful of countries, such as Germany, have any specific laws in place, which go even beyond the requirements of the EU’s Energy Efficiency Directive, adds Reynaers.

Singapore, meanwhile, has committed to assess the growth of data centres in a sustainable manner, launching its Green Data Centre Roadmap in 2023.

Outside of legislative frameworks, instruments such as the International Telecommunication Union’s standards for AI and the environment, as well as an UNESCO recommendation on the ethics of AI, have been introduced. 

‘In the US, we don’t have a comprehensive climate change policy,’ says Showalter. And since the second administration of President Donald Trump began in January, this has been coupled with a strident policy focus on rolling back energy regulation and building up the domestic power industry in areas including fossil fuels. 

However, commentators identify a willingness by the largest companies responsible for building AI-driven data centres to address environmental concerns. ‘The US technology sector largely comes from California, and those companies have roots in those issues,’ says Showalter.

He flags the expansion in initiatives such as district energy – in which excess heat from data centres is used to warm office buildings – or Microsoft’s contract with nuclear plant operator Constellation Energy to buy power for its data centres. And there’s also the potential inherent in the self-improving nature of AI to develop new ways of sourcing sustainable energy, updating power grids and making consumption more efficient.  

‘These companies are taking it seriously because they understand that securing energy supply is a critical task for them,’ says Chernyavsky, who’s a partner at Sayenko Kharenko in Kyiv. ‘It’s in their interests to find a solution’ in order to prevent governments potentially blocking the development of data centres, he adds.

Reynaers believes that scrutiny applied to the development of AI data centres at this stage is positive, as a means of catching environmental issues early on. She compares this with the development of many environmental laws, which have often been adopted in response to harms already underway, as scientific awareness has emerged much later. Reynaers shares feedback she has received from certain clients operating in this field, who would welcome the clarity of AI data centre-specific laws, ‘as long as it is as harmonised as possible across jurisdictions.’

‘It’s good that we criticise in a constructive manner,’ says Reynaers, ‘but the next step is that the scientific research needs to be translated into specific regulations for industry to comply with, while giving them sufficient time to implement it. Otherwise, you’re asking them to do so on a voluntary basis and that all becomes more ad hoc.’

The author thanks the IBA’s Legal Policy & Research Unit for their assistance with this article.

Image credit: Gorodenkoff/AdobeStock.com