Around the world, people are demanding a halt to the proliferation of data centres — racks of computers, housed in warehouses, that are powering the boom in artificial intelligence. Communities worry that existing infrastructure will not be able to handle the surge in demand for electricity and water, and that data centres’ energy demands will cause environmental and health problems, including increases in air pollution and greenhouse-gas emissions. The question of whether AI is going to stall climate action is explored in the first of a three-part Nature Podcast (Nature https://doi.org/rzqs; 2026).
Is the climate totally cooked? What the data say
People have many concerns about the way AI is being developed, including how it might affect safety, security and jobs. Where energy is concerned, however, any added demand for computing power can, and must, be met in smart, sustainable ways, not by burning more fossil fuels. Achieving this will take some creative thinking, and researchers need to be closely involved. Many argue that there are alternatives to relying on energy-intensive data centres, such as AI models that can run on individual computers or institutional servers, thereby consuming much less power than those run using cloud services. Efforts to develop such options need to be better funded.
Future projections of data centres’ energy demands are hard to make, because they are based on assumptions that could change. If current global trends continue, electricity consumption by data centres will go on rising. Between 2017 and 2024, it grew by around 12% per year. Figures for 2024 from the International Energy Agency (IEA), based in Paris, show most of this demand coming from three regions: the United States (45%), China (25%) and Europe (15%).
The Electric Power Research Institute, a non-profit organization in Palo Alto, California, has projected that the continued growth of US data centres could add up to 94 gigawatts of demand by 2030, equivalent to 17% of the country’s electricity consumption. Data centres currently consume more than 1.5% of the world’s electricity supply, and the IEA’s projections indicate that this figure will more than double in the next few years. By 2030, data centres could consume more electricity than Japan, the agency says.
Why scientists should lead the shift away from AI mega data centres
There are serious concerns that this extra demand will be powered mainly by fossil fuels, when carbon emissions are already at a record high. Research published this week in Nature Sustainability estimates that some 400 data centres in the United States accounted for around 26 megatonnes of carbon dioxide emissions between May 2024 and April 2025, more than those produced by many countries (G. Guidi et al. Nature Sustain. https://doi.org/hcsptc; 2026). And according to a study published last week in Nature Climate Change, data centres’ energy demands could prolong the lifespan of coal-fired power stations (C. Li et al. Nature Clim. Change https://doi.org/rzqr; 2026).
But none of this is inevitable. Governments could mandate that new energy demand be met with renewable sources. And regulators could look at alternative ways to manage existing electric supplies.
Many nations’ electric grid systems are designed to accommodate surges in demand, whether to power extra cooling on hot summer days or extra heating during winter cold snaps. Much of the time, there is excess capacity. This excess could be used to reduce the amount of new energy needed to power data centres.
How much energy will AI really consume? The good, the bad and the unknown
One solution is to curtail data-centre operations during periods of peak electricity consumption, and ramp up at times of lower demand. An analysis released in 2025 by researchers at Duke University in Durham, North Carolina, found that the US electric grid could accommodate 98 gigawatts of extra demand in this way, without adding any new generation capacity (see go.nature.com/4cdeknm). That 98 gigawatts would be enough to cover most, if not all, of the expected expansion in US data centres over the next few years.
There is potential for this to become a reality. In June, as part of a larger effort to modernize the grid, the US Federal Energy Regulatory Commission instructed grid operators to draw up a plan for how they will cope with extra demand from data centres. The technology industry is understandably paying close attention to how it might mitigate the problems of powering data centres. Earlier this month, OpenAI announced what it is calling a “grid protectors program”, a project to consult with US grid operators, regulators and policymakers to see how the company can work around peaks in power demand.
Meanwhile, Google is paying households and businesses in parts of the Eastern United States to reduce their electricity consumption during periods of high demand. Similar proposals are being considered in other countries.
Why open-source generative AI models are an ethical way forward for science
These are all tentative moves in the right direction, although arguably progress is not being made fast enough. Policymakers and data-centre operators must engage seriously with those who oppose the centres’ expansion. At a minimum, any new energy sources developed to power these centres must be renewable. Existing energy sources can also be better deployed, and researchers working to develop alternatives to energy-intensive data centres should be better supported. More peer-reviewed evidence is needed, especially for countries outside the United States.
There are smarter and more sustainable ways to power the growth of AI, but to achieve them, concerted and coordinated action will be needed. In this way, the harm posed to people and the planet can be minimized, and AI technologies can benefit everyone.