At the start of the year, the International Energy Agency (IEA) forecast that global electricity demand is expected to more than double from 2022 to 2026, with AI (and cryptocurrency) playing a significant part in that rise – given global electricity consumption of 460 TWh.

The IEA forecasts that electricity consumption by data centres in the European Union (EU) will be 30% higher than 2023 levels (about 100 TWh), as new data facilities are commissioned in line with increased digitalisation and AI in particular. Ireland and Denmark alone make up 20% of the expected increase in EU data centre demand to 2026. And there are moves by the European Commission towards regulating data centre sustainability across the EU. Specific action is underway in some regions, such as the ad hoc moratorium on building new data centres in Dublin.

There are currently over 8,000 data centres globally, with about 33% located in the US, 16% in Europe and about 10% in China. Some 1,240 data centres were located in Europe (as of 2022), mostly concentrated in major financial centres including London, Frankfurt, Amsterdam, Paris and Dublin. With significant additional data centres planned, electricity consumption in the EU is expected to reach 150 TWh by 2026.

US data centre electricity demand is forecast to grow rapidly in coming years, rising from 200 TWh in 2022 to about 260 TWh in 2026 – accounting for 6% of US electricity demand.

China’s State Grid Energy Research Institute estimates that electricity demand in the country’s data centres could double to 400 TWh by 2030, compared to 2020. To meet this rocketing demand, regulations are being updated to promote sustainable practices in current and future data centres, aligned with decarbonisation strategies.

Why are data centres expanding so fast?
We all play a part. Every time phone data is backed up to the cloud it is stored in data centres. Moreover, 5G, cloud-based storage, Big Data and the Internet of Things (IoT) play a vital role in essential infrastructure worldwide, supporting everything from financial transactions to social media, industry and government operations. Indeed, the IEA estimates that data centres account for over 1% of global electricity utilisation.

As mentioned, both AI and blockchain technologies such as cryptocurrency are power-hungry operations. The US Energy Information Administration calculated that bitcoin mining accounted for 2% of US electricity demand in 2023 – and that figure is climbing rapidly.

What’s more, it is very energy-intensive to ‘train’ AI models. For example, a ChatGPT query uses 10 times more energy than a Google enquiry, David Proctor, a Vice President at the Electric Power Research Institute told Time magazine. He reckons that 10–20% of data centre energy in the US is currently used by AI, with a ‘significant increase’ on the horizon.

More and more powerful generative AI models are being developed, and researchers estimate that the computational power required to train these models is doubling every nine months.

The IEA predicts that within two years, data centres could consume the same amount of energy as Sweden or Germany’s total electricity demand. This voracious appetite for energy by AI is not just a matter of vast power consumption and GHG emissions – depending on whether the power is sourced from traditional fossil fuels (commonly) or renewables – but also consumes massive reservoirs of fresh water. A study published by UC Riverside estimates that global AI demand could cause data centres to use 1tn gallons of fresh water by 2027 for cooling purposes.

According to analysis by The Guardian (15.9.2024), tech giant GHG emissions may be 7.6 times higher than official figures on account of AI-led data centres.

Despite the likes of Google and Microsoft pledging to run ‘carbon-free by 2030’ in 2020, they and other tech giants, like AWS, have taken a step backwards. Microsoft actually increased GHG emissions by a third last year in pursuit of its AI development programme, and plans to double its data centre capacity, having invested billions in OpenAI.

In 2023, Google’s total data centre electricity consumption grew by 17%, despite an express strategy to maintain a 100% global renewable energy match. Google’s 24/7 Carbon Energy Pledge clearly states: ‘From 2010 to 2023, we signed more than 115 agreements totalling over 14 GW of clean energy generation capacity… now we’ve set a goal to run on 24/7 carbon-free energy on every grid where we operate by 2030’. This involves cutting 50% of its Scope 1, 2 and 3 emissions by 2030, investing in nature-based and technology-based carbon removals.

Indeed, environmental concerns about data centres are rising around the world as more people, industries and governments look to house an ever-growing mountain of data. The UK recently declared data centres as ‘critical infrastructure’ – a vital part of modern digital economies.

However, there are concerns that there isn’t sufficient power generation or transmission capacity in the pipeline to fuel all the data centres under development. There is a suspicion that the tech giants are gambling on sufficient build-up of the necessary infrastructure to satisfy their escalating data centre needs. Moreover, better transparency will be vital, as there is still a lack of understanding of the environmental impacts of AI.

The IEA predicts that within two years, data centres could consume the same amount of energy as Sweden or Germany’s total electricity demand.
 

What measures are underway to improve data centre sustainability?
Currently the data centres rely heavily on fossil fuelled power generation, although many companies are negotiating power purchase agreements with renewable energy suppliers, where possible. AI promoters and entrepreneurs suggest that ‘innovative technology’ will play a vital role for combatting climate change.

For example, Google claims that its DeepMind AI helped reduce the company’s data centre cooling by up to 40%. Furthermore, efficiency improvements in chip hardware are set to have a big impact. The world’s largest chipmaker NVIDIA recently launched a new line of graphics processing units (GPUs) with 25 times lower energy consumption than previous models.

Digital technology may hold the key to more sustainability in this sector. Steven Brown, Vice President of Digital Energy Solutions at Schneider Electric told Data Centre Magazine: ‘At Schneider Electric, we call it Electricity 4.0, using digital technology to become more efficient and then electrifying where we haven’t yet to decarbonise… helping microgrid infrastructure and smart grid infrastructure.’

Brown suggests there will be a re-emergence of ‘edge infrastructure’, where data centres will deploy workloads close to where they are being used. ‘Once a model is trained, that has to be deployed as an inference model where you get much more agility – such as for cloud AI. Building at the edge requires the kind of critical infrastructure to protect it that you might expect in a data centre environment,’ he says. Indeed, the edge computing model can reduce power demand significantly.

On the road towards better sustainability, Dave King, Principal Senior Product Engineer at Cadence, suggests that ‘digital twin technology’ has a key role to play in developing more energy-efficient and sustainable data centres. ‘Using virtual digital replicas of physical data centres, more efficient power management and optimisation of operating systems, such as cooling, can be implemented, helping facility managers reduce the carbon footprint of AI,’ he says.

Companies like CyrusOne are designing projects to scale AI infrastructure effectively. According to Tom Kingham, Design Lead Europe & Japan at CyrusOne: ‘We see a stronger focus on sustainability, with innovations reducing physical space requirements and carbon footprints, reflecting a broader trend towards greener data centre operations,’ he told Data Centre Magazine.

According to the IEA’s 2024 Electricity Report, the primary drivers of data centre electricity demand are the cooling systems and the servers themselves, which each typically account for 40% of the total consumption. The remaining 20% is consumed by the power supply system, storage devices and communication equipment. Adoption of high-efficiency cooling systems has the potential to reduce electricity demand in data centres by 10%. A further 20% reduction in consumption can be achieved with ‘direct-to-chip’ water cooling and low viscous fluids to cool all other components.

Looking longer term, quantum computers (which deliver faster processing power than supercomputers, while consuming less energy) could reduce electricity demand of the sector if supported by efficient cooling systems. But they need to be cooled to temperatures at absolute zero (–273°C), while supercomputers can operate at 21°C.

Data centres are also becoming more sustainable and their operations more efficient with the move towards hyperscale data centres (exceeding 5,000 servers and 10,000 sq ft), without a significant increase in electricity consumption. The global market for hyperscale data centres is projected to be $212bn by 2026 (double that of 2023).

In 2023, Sweden announced plans for a nuclear-powered data centre using small modular reactors (SMRs).

Another promising area of research for decarbonising data centre operations involves ‘time and location shifting’ of electricity demand, using new software to temporarily shift power loads with ‘carbon-aware’ models that relocate data centre workloads to regions with lower carbon intensity at selected times – while also increasing operational affordability and lowering emissions by 34% round the clock. Research shows that this approach, along with other on-site energy efficiency measures, can help data centres achieve 64% carbon-free energy consumption, says the IEA.

Indeed, there is an argument that there is a need for data centre operators to optimise and fine-tune AI models as well as invest in more energy efficient AI accelerator hardware, so AI becomes a more effective solution rather than a challenge to the environment and climate change.

• Further reading: ‘AI and renewables: What’s the potential and reality?’ Artificial intelligence and machine learning have the potential to support, optimise and accelerate the development of renewables in the energy transition. New Energy World Features Editor Brian Davis offers an insight into a raft of AI applications – and also road tests ChatGPT.
• Intelligent use of data offers material benefit to every aspect of energy and utilities, explains James Forrest, Global Industry Leader for Energy & Utilities at Capgemini.