Data centres’ energy demand is increasing finds IEA

Energy demand of data centres globally may increase by up to 300% by 2026 the IEA has reported in its mid-year electricity update.

The IEA’s previous best estimate for the electricity consumption of data centres – excluding cryptocurrencies – was about 1-1.3% of global electricity demand in 2022 and its recent projections suggest it could rise to a range between 1.5-3%.

Despite this increase, the projected demand nevertheless remains relatively limited in global terms and in comparison with other sectors – that for EVs is projected to reach around 2% by 2026 while aluminium production consumes around 4%, for example – but the nature of locating data centres is that local bottlenecks are already emerging, the IEA states.

In Ireland, for example, 18% of electricity demand came from the data centre sector in 2022. In Singapore, data centres accounted for around 7% of national electricity use in 2020.

Have you read?
GridBeyond BESS to set up Dublin data centres for grid support
The good, the bad and the safety: AI and data in the energy sector

This, in turn, has led to restrictions on commissioning new data centres, such as in Ireland where grid connection conditions have been in place since late 2021.

In another example, in the Netherlands since January 2024 new hyperscale data centres are restricted in most locations due to grid constraints among other reasons.

The IEA points out that to circumvent grid connection challenges or to reduce dependency on the grid, some data centre providers are increasingly looking into on-site generation.

For example in China, where data centres and 5G networks are growing sources of electricity demand, technology company Tencent installed around 10MW of rooftop solar with battery storage at its Tianjin cloud data centre.

In the US, Amazon Web Services has acquired Talen Energy’s data centre connected to the Susquehanna nuclear power plant, with plans to expand its capacity.

In that connection, alongside renewables nuclear and geothermal are emerging as attractive low-emission options for powering data centres, the IEA adds.

In another example, Microsoft has become the first PPA signee on fusion power with Helion, with its plant planned to come online by 2028.

Microsoft also has secured a 10-year geothermal PPA in New Zealand and alongside AI company G42 is investing $1 billion in a geothermal-powered data centre in Kenya.

Stocktaking and future demand

The IEA’s mid-year update comments that electricity consumption of data centres has historically shown a wide range of estimates across sources, with a lack of reliable data in many countries contributing to substantial uncertainties in consumption levels.

For example, studies for the EU show that the share of data centres’ consumption in total electricity demand in 2022 could have ranged between 1.8% to 3.5%, while in the US 2022 estimates ranged between 1.3% to 4.5%.

“Global data centre deployment trends, evolution of technology towards higher power demanding servers and recent announcements of construction plans underline the importance of better stocktaking of data centres. Improving the collection of energy consumption data of the sector is important to correctly identify past developments. This in turn helps to better understand future trends in order to inform the public debate more accurately.”

Future demand projections also bring the additional challenge of the uncertainty of AI energy consumption, with the increasing adoption of AI applications, particularly generative AI.

This is most notable in the US, where 2030 projections range from 4% to 10% of national electricity use.

Against this, AI accelerators such as graphics processing units are also becoming increasingly energy efficient even as they are also more powerful. NVIDIA has reported its new generation of chips over the older generation is 25 times more energy efficient.

Despite this, as computation power and efficiency increase, some degree of rebound effects cannot be ruled out, since the corresponding decrease in the cost for computation can drive additional demand, the IEA concludes.