The insatiable energy thirst of artificial intelligence (AI) continues to test global infrastructure. In December, the US Department of Energy published a report with forecasts that left previous estimates short: data center consumptionthe processor-packed facilities where models are trained, data housed, and calculations made possible by AI tools are executed, has tripled over the past decade and will triple again through 2028. According to its projections, The amount of electricity they will need will be at least 325 terawatt hours (TWh), that is, more than what entire countries such as Spain (246 TWh), the United Kingdom consume in a year (287 TWh) or Italy (298 TWh).
Assuming that an average of 50% of the capacity of these data centers is used, the Department of Energy report, prepared for the Government by the Lawrence Berkeley National Laboratory (LBNL), specifies that the United States will need an installed capacity of between 74 and 132 GW to power these key infrastructures for digitalization and AI in 2028, more than what an entire country like Spain generates (125.6 GW by 2023). Thus, the United States will dedicate between 6.7% and 12% of all electricity consumed to this growing industry.
These forecasts could fall short after the announcement this week by the president of the United States, Donald Trump, of investments worth 500 billion dollars (about 480 billion euros) in four years to promote AI. That money, which Trump described as “the largest artificial intelligence infrastructure project, by far, in history,” will primarily go toward building data centers and the power plants to power them. The capital for the project, named Stargate, will come from SoftBank, OpenAI, Oracle and the Abu Dhabi sovereign fund MGX. Trump said he will use emergency declarations and executive orders to ensure the construction and uninterrupted access to energy they require. One of the 41 executive orders that the Republican signed on his first day of his new term will involve the review of all federal regulations that impose an “undue burden” on the development or use of various energy sources, particularly coal, oil, natural gas, nuclear energy, the latter being one of the solutions considered by large technology companies to ensure a continuous energy supply in their data centers.
“The Department of Energy’s projection is very striking: data centers, driven mainly by AI, can reach 12% of total US electricity consumption by 2028, while the most common forecasts pointed to 10% by 2030,” says Shaolei Ren, associate professor of electrical and computer engineering at the University of California, Riverside and specialist in AI sustainability. “This means that, if data centers specialized in AI continue to be built without substantial improvements in their energy efficiency, the overload of the electrical grid will arrive much sooner than expected,” adds the academic.
“The results presented in this report show that the electrical consumption of data centers in the United States is growing at an accelerated rate,” he warns. the Department of Energy document. “The compound annual growth rate was 7% between 2014 and 2018, 18% between 2018 and 2023, and will be between 13% and 27% between 2023 and 2028.”
Another interesting aspect of the Department of Energy report, according to Ren, has to do with its data center water consumption estimatesa resource that some facilities use to cool the processors. According to the document, in 2023 US data centers used about 66 billion liters of water, part of which evaporated. The calculation does not take into account the water used to generate energy: it only takes into account what was used in cooling systems. By 2028, the figure could reach 124 billion liters. “Even using conservative models, the study projects that the water consumption of US data centers in 2028 could double or even quadruple the level of 2023, which is an alarming sign,” says the expert.
The great energy guzzler
The US is not prepared to respond to the growth in demand that will lead to the rise of generative AI, which is behind tools such as ChatGPT, Copilot or Gemini. OpenAI CEO Sam Altman believes the technology his company has done so much to make mainstream will cause an energy crisis. “There is no way to get there without drastic changes,” the executive told an attentive audience of businessmen and statesmen during a conference at the 2024 Davos Forum.
These “drastic changes” already have a face and eyes. The sector believes that The solution is in the so-called advanced nuclear energythe name under which pocket nuclear reactors and nuclear fusion are included. Companies like Amazon, Google, Microsoft or Meta have already stated that they welcome this solution to power their data centers. Nuclear energy, by not depending on environmental factors (such as the sun, the wind or the amount of water in reservoirs), can ensure an uninterrupted supply. Some of them, such as Microsoft, have already closed supply agreements with nuclear plants. that they will return
The outgoing Secretary of Energy, Jennifer Granholm, met in March of last year with representatives of some of the aforementioned companies to explore imaginative formulas to respond to this large increase in energy demand. One of the options that was put on the table was the development of data centers with small modular reactors (SMR), according to Axios. It has not emerged that any agreement has been closed, although the first executive orders from Trump and the Stargate alliance suggest that it would not be unreasonable for these types of projects to obtain the green light.
The situation in Spain
Spain is establishing itself as a pole of attraction for data centers. Microsoft, Amazon or Meta are some of the large technology companies that already have or are developing their own projects in the country.
The latest projections on the energy consumption of this industry are not as large as those of the United States, but they reflect that we are facing a sector in full growth. The specialized consulting firm DNV believes that Spain’s data centers will go from having an electrical load capacity of about 900MW (the 2024 estimate) to about 1,350 in 2030. That is, 50% more energy will be needed.
The figures are larger the further forward we look. “Currently, the energy demand of data centers and AI represents 13% of the electrical demand of commercial equipment, and this figure will have increased to 44% in 2050,” says the consultancy in its report. Energy Transition Outlook Spain 2024and warns: “This demand represents a considerable challenge for the Spanish electricity grid, which will have to guarantee a constant and sustainable supply of energy.”
