Electricity demand growth in IEA scenarios, in TWh between 2024 and 2030, due to data-centre expansion (dark blue) – including a scenario in which expansion happens faster (light blue) – as well as from other sectors (grey). Source: IEA energy and AI report.
If data-centre electricity use rose in line with the IEA’s faster-growth scenario, the facilities would be responsible for around 12 per cent of global demand growth overall.
While the IEA says “uncertainties widen” when considering electricity demand growth beyond 2030, it expects a continued – albeit slower – increase to 1,193TWh by 2035.
This would mean annual demand growth roughly halving, from around 90TWh per year out to 2030, down to less than 50TWh a year out to 2035.
While the global picture suggests a relatively modest role for data centres in driving near-future electricity demand growth, it could be far more pronounced in some countries.
Data centres are very geographically concentrated, both in terms of their global distribution and within leading countries. Today, nearly half of their electricity consumption takes place in the US, 25 per cent in China and 15 per cent in Europe, according to the IEA.
US data centres used around 4 per cent of the nation’s electricity in 2023 and this is set to rise to 7-12 per cent by 2028, according to analysis by the Lawrence Berkeley National Laboratory.
In Ireland – regarded as a European “tech hub” – around 21 per cent of the nation’s electricity is used for data centres. The IEA estimates that this share could rise to 32 per cent by 2026.
Data-centre electricity demand tends to be further localised in certain regions. In the US state of Virginia, these facilities already consume 26 per cent of electricity, while in the Irish capital, Dublin, the figure is 79 per cent, according to analysis by Oeko-Institute.
Much of the commentary on AI threatening climate goals comes from “advanced economies” in the global north, where the IEA estimates that, on average, a quarter of electricity demand growth by 2030 will be driven by data centres.
(In many of these countries, electricity demand has previously been flat or falling for years.)
Roughly half of the power demand growth in the US and Japan over the next five years is expected to come from data centres, according to the IEA, as shown in the figure below.
Share of electricity demand growth between 2024 and 2030, %, in the IEA’s central scenario for data-centre expansion, in select countries and country groupings. Source: IEA energy and AI report, IEA.
While there are some notable exceptions, such as Malaysia, data centres are set to be a relatively small portion of electricity demand growth in developing and emerging markets.
Around the world, electricity grids are under strain, with many developed countries, in particular, seeing long wait times for grid connections and new transmission lines. Data-centre growth is raising this pressure.
There are also growing concerns, notably in the US, about the impact data-centre growth could have on energy bills.
The IEA says that demand growth presents “advanced economies” with a “wake-up call” for the electricity sector to invest in infrastructure, otherwise “there is a risk that meeting data-centre load growth could entail trade-offs with other goals, such as electrification”.
4. Fossil-fuel use will likely expand to power data centres, but clean-energy supplies are set to grow faster
The extent to which data-centre growth increases emissions depends on which energy sources power those data centres.
Data centres can use power from the grid, in which case their electricity mix will reflect that of the region they are in and could therefore become cleaner as nations decarbonise.
They can also be powered by “captive” sources, built to supply specific facilities, such as solar panels, small nuclear reactors or gas turbines.
There are concerns that data-centre expansion will be used to justify the prolonged use of fossil fuels, “locking in” a future of elevated emissions.
Indeed, the likes of Shell have framed AI in such terms and some data-centre operators have been explicitly seeking gas connections to meet their electricity needs.
Currently, coal is the biggest single electricity source for data centres globally, largely due to the numerous facilities in China.
Overall, fossil fuels provide nearly 60 per cent of power to data centres, according to the IEA. Renewables meet 27 per cent of their electricity demand and nuclear another 15 per cent.
(These figures are based on the electricity these facilities consume, rather than any contracts they have to buy clean energy credits.)
In the IEA’s central scenario, by 2035 the ratio of the data-centre electricity mix switches from around 60 per cent fossil fuels and 40 per cent clean power to 60 per cent clean power and 40 per cent fossil fuels, as shown in the chart below.
This is expected to be driven primarily by the wider global expansion of renewables, although some projects will be funded directly by data-centre companies.
However, the IEA says significantly more gas and coal power would likely still be required to meet data-centre demand, both from ramping up existing plants and building new ones.
Annually global electricity generation, TWh, expected to supply data centres globally over 2024-2035, broken down by generation type, in the IEA’s central scenario. Low-carbon electricity sources are grey and fossil fuels are blue. required Source: IEA energy and AI report.
Gas-power generation for data centres is expected to more than double from 120TWh in 2024 to 293TWh in 2035, with much of this growth in the US, according to the IEA.
About 38GW of captive gas plants currently “in development” – roughly a quarter of all such projects – are planned to power data centres, according to Global Energy Monitor (GEM).
The US has doubled the amount of gas- and oil-fired capacity it has in development over the past year, driven partly by the energy demand of the “burgeoning AI industry”, according to GEM.
However, these projects are facing long lead times and “sharply” rising costs, with GEM noting, as a result, that many may never materialise.
5. There is a lot of uncertainty about how much data centres will expand
Currently, there are no comprehensive global datasets available on data-centre electricity consumption or emissions, with few governments mandating any reporting of such numbers.
All figures concerning the energy and climate impact of AI are therefore estimates.
The IEA has assessed hundreds of available estimates and forecasts, noting that even historical data can be “widely divergent”, due in part to a lack of common definitions.
On top of this, there are major uncertainties, including over how quickly AI will be adopted. Despite the enthusiastic uptake of generative AI by individuals and companies, some argue that the business case for continued, rapid growth may be weaker than suggested.
Another uncertainty is how energy-efficient AI will be. Experts have already identified efficiency improvements resulting from better chips, more efficient training algorithms and larger data centres, all of which could continue curbing electricity demand.
(Google has also reported a substantial drop in the electricity use required for individual AI search queries, which is already small compared to the power needed to train AI models.)
A final uncertainty is over how many proposed data centres will actually get built, with some speculative requests for grid capacity relating to plans that may never materialise.
As a result of these knowledge gaps, there have been numerous estimates of short-term electricity demand growth from data centres, which have produced very different results, as shown in the chart below.
Some estimates – such as one from the Gas Exporting Countries Forum arguing that more gas exports will be needed to fuel meteoric rises in electricity demand for AI – have been deemed less credible in reviews by independent experts.
Estimated electricity demand from data centres globally in 2023 and 2030, according to scenarios laid out by a range of analysts. Some scenarios have been extrapolated one or two years forward or backwards so that they cover the same timescale. The Liebreich Associates scenario is approximate, based on estimates of % total electricity demand. Sources: IEA, Goldman Sachs, Deloitte, Boston Consulting Group, Gas Exporting Countries Forum, Liebreich Associates, Semianalysis, Schneider Electric, International Data Corporation, Jeffries (2024).
Another area of great uncertainty concerns the impact that the application of AI could have on electricity use and emissions.
Some researchers have attempted to calculate how much AI could curb emissions, by helping to identify efficiency gains in other parts of the energy system, or by making technological breakthroughs.
In some “exploratory” analysis, the IEA says such gains could cancel out any extra data-centre emissions due to the growth of AI.
However, it adds that despite the AI hype, “there is currently no existing momentum of AI adoption that would unlock these emissions reductions”.
This story was published with permission from Carbon Brief.
