Artificial Intelligence: Insatiable Vampire Hungry for Electricity

The topic of artificial intelligence (AI) is currently being discussed with increasing frequency and has fueled heated debates about the need to “delegate” to AI some tasks, or even entire industries, that can be performed without human intervention and presence. Artificial intelligence is no longer only capable of finding, grouping, and cataloging information present on the web in increasingly faster and more efficient ways, but thanks to advances in its computational capabilities, AI can now perform tasks previously unimaginable, such as simultaneous interpretation of conversations in real time, creating photos of people or landscapes from a brief description, or even composing and creating music or videos without musicians or actors. The concept and term artificial intelligence originated in 1956, but the popularity of this technology began to grow with the increase of the amount of processed data, the development of more and more complex algorithms to manage masses of data, and the increase of computing power of servers and computers on which this data is stored and processed. What we call AI today is more properly defined as “generative” artificial intelligence, which is an evolution of “machine learning” and “deep learning,” the foundation of data analysis and data processing on which AI relies and from which it extrapolates the information it needs to function.

While attention is paid to the vast computing power that can make humans “obsolete” and “unnecessary” in some areas, we tend to overlook the impact AI is having and will have on energy consumption. The processes that artificial intelligence performs are actually very energy-intensive. According to Goldman Sachs statistics, processing a ChatGPT query requires on average ten times more electricity than it takes to perform a search on Google. The US Department of Energy (DOE) compared the energy consumption of offices and data centers and found that data centers use 10 to 50 times more energy than a typical commercial building. The European Union has published a study, which shows that in 2022, the electricity consumption of European data centers, estimated at between 45 and 65 TWh, was about double that of the entire telecommunications sector, estimated at between 25 and 30 TWh.

According to International Energy Agency (IEA), the demand for electricity to run servers located in data centers has been growing since 2010 to the extent that it accounts for approximately 3% of the global electricity consumption. According to IEA estimates, these volumes should double by 2026, exceeding 1000 TWh in the most pessimistic scenario. Depending on the number of data centers to be commissioned in the future, on energy efficiency improvements, and on the development and proliferation of artificial intelligence and cryptocurrencies, electricity demand could reach values between 650 and 1050 TWh, up from 460 TWh in 2022. Again, according to IEA data, 1% of greenhouse gases can be attributed to the generation of the electricity needed to keep the data centers scattered around the world running.

The distribution of servers and data centers is not at all homogeneous; moreover, it is quite highly concentrated. The USA, given the presence of the world’s largest IT companies on its territory, is the country with the largest number of data centers. The analysis titled “The U.S. Needs a Bigger Energy Boat: Putting the Sheer Magnitude of Forecasted Energy Demand into Perspective” by consulting firm Pickering Energy Partners (PEP) highlighted that the United States has approximately 5381 data centers in operation, 10 times more than the 521 in Germany, which ranks second in the world, and 12 times more than China’s 449. The PEP study also highlights two other very important data points for assessing the impact of AI on future electricity demand, namely the pace of R&D and the level of its effective utilization. In the USA, for example, AI has a 43% development rate, but its utilization rate is only 25%. Germany’s development and diffusion rates are in line with the world average at 44% and 34% respectively, while China’s are 58%, the highest in the world, and 30%, one of the lowest in the world. By analyzing this data, the US consulting firm wanted to emphasize that the impact of artificial intelligence on electricity demand can still be considered limited, and therefore the US government’s energy consumption projections are not very realistic. In fact, the PEP’s conservative estimate calls for doubling the current electricity consumption in the United States, which could reach 8400 TWh. In Europe, the situation in the electricity sector is somewhat different. While Europe is, in essence, the second region in the world in terms of concentration of operational data centers, their sizes are not comparable to those of data centers in the USA. Nevertheless, the weight of their consumption in the total electricity demand is not insignificant and amounts to 2.7%. It should also be noted that this percentage is estimated to grow and could reach 3.2% by 2030, a 28% increase compared to 2018. Despite the European Union’s efforts to curb the growing consumption of data centers through the approval of the Energy Efficiency Directive, which sets a minimum 11% reduction in electricity consumption, by 2030 consumption for data center operations could reach about 100 TWh, which is twice as much as in 2022.

Thus, the spreading use of artificial intelligence seems to represent another factor capable of putting further pressure on the global electrification plans. However, in the political context that aims to reduce the weight of hydrocarbons in the global energy mix in favor of renewables, new questions about achieving energy security arise.

Renewable energy sources, given the technologies available today, are currently not only virtually incapable of offsetting and guaranteeing the amount of energy that will no longer be produced using hydrocarbons, but will not even be able to handle the production of the additional electricity needed by data centers. The United States can once again provide a prime example of this. According to the US Department of Energy’s “On the Path to 100% Clean Electricity” analysis, the US government has determined that by 2030, the country should be able to generate at least 80% of its electricity from renewable sources. Data from the Energy Information Administration (EIA) for the end of 2023 showed, however, that the target is still quite far away, as 894 TWh was produced out of 4178 TWh from renewable sources, which accounts for approximately 22% of all US energy production. Assuming that future electricity demand remains unchanged, alternative sources are expected to generate 3400 TWh, about 3.5 times what they are capable of producing today. If the growth rate of energy demand, based on the degree of AI adoption, were between 3.7% and 15% per year, the renewable electricity generation capacity should reach a conservative estimate of 4500 TWh. Planned US generating capacity from alternative sources, however, does not seem likely to keep pace with the growth in potential energy needs, given that the EIA’s 2024 additions are “only” 45 GWh, or 0.045 TWh, of which 36 GWh is attributable to solar generation and 8 GWh to wind generation.

Even as far as Europe is concerned, there are doubts about the ability of renewable sources to generate the additional electricity needed to phase out hydrocarbons and meet growing demand. Nevertheless, the European Commission has increased its 2030 targets for energy production from renewable sources, bringing them from 32% to at least 42.5%, almost doubling from 23% in 2022. All this despite the fact that renewable sources accounted for 50% of the electricity generation mix in the first half of 2024, adding 56 GWh of solar energy and 16 GWh of wind energy in 2023. It is noteworthy that this was possible given the forced abandonment of natural gas and coal, but also and above all, due to the decrease in demand for electricity itself, which in 2023 recorded a decrease of about 6.5%, 186 TWh, compared to the same period in 2021, following the reduction in energy-intensive industrial production of the European Union. Total electricity demand in Europe, again at the end of 2023, was around 2697 TWh, which is the lowest level recorded since 2001. However, production capacity from renewable sources is still far from the targets set. The average annual growth of production from renewable sources in Europe was about 1%, however, according to the REPowerEU plan, by 2030, generating capacity should increase from about 614 GWh in 2023 to 1236 GWh from solar and wind power, namely from 260 GWh and 221 GWh in 2023 to 700 GWh and 400 GWh respectively.

The figures thus seem to indicate how we are increasingly approaching a crossroads where the political ambitions of divestment from fossil fuels will have to clash with actual production capacity and how they will be able to meet the electricity demand destined to grow following the electrification and digitalization of the world we live in. According to the IEA, the goal of tripling generation capacity from alternative sources is currently questionable. According to Fatih Birol, executive director of the IEA, it is up to national governments to provide greater incentives to move towards a more sustainable energy system, thereby reducing pollution. However, there can be no doubt that political action by governments is not enough, as the environmental targets set are too ambitious and do not really take into account the obstacles, physical and/or financial, that need to be overcome and the effort required to achieve the targets, whether medium-term (2030) or long-term (2050).