Explainer
Fighting the Power Deficiency: The AI Energy Crisis
Is AI contributing to solving the climate crisis or to making it worse? Either way, the increase in AI applications goes hand in hand with the need for additional data centers, for which energy resources are currently lacking.
If you would like to learn more about the topic, please get in touch with:
Dr. Angela Müller
Executive Director AlgorithmWatch CH | Executive Board Member AlgorithmWatch
Predictions about the positive environmental impacts of «game-changing» AI innovations are to be taken with a grain of salt.The sector is not known for eagerly validating that their interventions have worked at the scale promised. For this lack of consistency, the sustainability expert Vlad Coroamăhas coined the term «Chronic Potentialitis,» as the AI industry is generally more interested in what might be than what actually is the case. However, projected innovations are often used to devalue criticism.
A reoccurring argument pops up in the debate: If deploying AI has positive and negative environmental impacts, the benefits will outweigh the environmental costs. This happened at the Breakthrough Energy Summit in London in 2024, where Bill Gates urged environmentalists and governments to «not go overboard» on concerns about the huge amounts of power required to run new generative AI systems.
«Everything is going to be fine»
AI development is one of the factors driving the rise of the global energy demand. In order to meet this demand, new data centers need to be built. During the Breakthrough Energy Summit, Bill Gates also said that these data centers will drive a rise in global electricity usage between 2-6 percent. As a rise in global energy consumption would aggravate the climate crisis, are data centers a matter of concern?
Gates says no. «The question is, will AI accelerate a more than 6 percent reduction? And the answer is: certainly.» AI is increasingly being applied to accelerate innovation and technology development in numerous fields, including the energy sector. For example, researchers are using AI to accelerate the discovery of promising battery chemistries. AI models themselves are becoming more efficient. According to a 2019 study by Microsoft and PwC, AI has the potential to reduce global greenhouse gas emissions by one and a half to four percent by 2030. It is not known from where Gates takes his new estimate.
The projected reduction is not the only Microsoft-related number that has become bigger. In its 2024 Environmental Sustainability Report, Microsoft admitted that its greenhouse gas emissions had risen by almost a third since 2020, in large part due to the construction of data centers.
Among Big Tech companies, Microsoft is not alone in its struggle to stick to prior climate commitments. Google's aggressive pursuit of AI has led to a 48 percent increase in its greenhouse gas emissions over the past five years, as the company revealed in its 2024 Environmental Report. The emissions increased at a rate of 13 percent in 2023. Google linked this growth to its AI initiatives, citing electricity consumption by data centers and supply chain emissions as primary drivers.
Unlike Bill Gates, Google pointed to «significant uncertainty» in reaching sustainability targets, particularly noting «the uncertainty around the future environmental impact of AI, which is complex and difficult to predict.» Reaching its «extremely ambitious» goal of net zero emissions by 2030 would be at risk.
Empty promises: Big Tech’s sustainability failure
Big Tech’s stark increase in emissions in the last years, with no hard data available that would support the widespread narrative of AI being a sustainability booster, underscores the growing tension between the tech industry's declared AI ambitions and its actual climate responsibilities. As Big Tech invests more in AI, the public pays more attention on the environmental costs of powering and cooling massive data centers.
A 2024 report by the US Department of Energy estimates that data center load growth has tripled over the past decade and is projected to double or triple by 2028. AI is expected to be the biggest driver of US data center-related load growth in the near future. While the likes of Amazon and Microsoft have signed long-term power purchase agreements with wind and solar power generators, those deals «typically do not match electricity demand hour by hour with local resources,» the US agency said, meaning that there was «no guarantee that all electricity-related greenhouse gas emissions are offset» by such agreements. The US estimate for future energy consumption by data centers is particularly important as the US is a critical hub of AI development.
AI energy demand: detrimental to the energy system transformation
Data centers are vast collections of computer servers forming the backbone for AI services in storing, processing, and distributing data. AI requires an extensive number of these servers. In 2022, data centers collectively accounted for about 1 percent of global electricity demand. In large economies, such as China, the European Union, the United Kingdom, and the United States, data centers are estimated to account for 2-4 percent of the electricity demand.
However, such numbers hide local challenges, as data centers tend to cluster together. They are typically much more spatially concentrated than other similarly energy-intensive infrastructures. In at least five states in the United States, data centers have already surpassed 10 percent of total electricity consumption. In Ireland, the share is over 20 percent of all metered electricity consumption. With major data center campuses currently under development, this could lead to considerable strain on local grids.
In early 2024, OpenAI chief executive Sam Altman admitted that the AI industry is heading for an energy crisis. Meeting this electricity demand with emissions-intensive sources of electricity could also throw regional energy transition targets off track. His solution is nuclear fusion. In 2021, Altman started investing in the nuclear fusion company Helion Energy. Google, Amazon, and Microsoft are also investing in new nuclear power stations, but it can take decades to get them running. Apart from more fundamental concerns in regard to the security, sustainability, and cost-effectiveness of nuclear energy, relying on it does not answer the problem at hand. It is time to avoid any irreversible effects of the climate crisis as soon as possible. A technology whose potentially positive effects would only be seen many years later cannot be a suitable measure in the face of current challenges. We need an immediate answer to AI’s energy demand. Nuclear power does not help at all at this point.
Big Tech’s proclaimed net-zero goals also compel them to buy so-called «renewable energy credits» (RECs), which supposedly support an additional generation of renewable electricity. This arrangement has a major flaw: Paying for renewable electricity generation that is to occur at some point, somewhere in the world, does not even begin to cover the amount of electricity that the company consumes in a specific place at any given time. Even if a solar farm operator receives additional money for producing electricity it was going to generate anyway, a server farm’s emissions from electricity provided by a natural gas power plant elsewhere are not reversed.
«The reality on the ground is that its data centers are driving up demand for fossil fuels,» argued a report from Amazon Employees for Climate Justice, a group of workers that demand the company take more aggressive climate action. The organization said that a significant share of Amazon’s RECs is not contributing to developing new projects. The group estimates that 78 percent of Amazon’s US energy comes from nonrenewable sources and accuses the company of using «creative accounting» to claim it has reached its clean-electricity goals.
Even if Big Tech sticks to its commitment to renewable energy, there are companies that most certainly will not: oil and gas companies. AI is being used for fossil fuel extraction right now. For example, the management consultancy EY stated on their website (the survey referenced on the page has been taken down but is available on the Internet Archive) that «more than 92 percent of oil and gas companies are either currently investing in AI or plan to in the next two years.» A 2024 report details Microsoft aggressively chasing multiple hundred-million-dollar deals to accelerate fossil fuel extraction with various oil super majors like ExxonMobil, Chevron, and Shell. The question is: Can AI really enrich fossil-fuel companies and fight climate change at the same time?
Globally, the regions with the most projected data center capacity tend to be areas where a larger share of electricity generation comes from fossil fuels. In the US, where most of the data center capacity is located, it tends to be in areas with easy access to gas or coal, such as North Virginia. Fossil energy generation from coal and gas still makes up most of the electricity generation nationwide. China, likely the nation with the next highest amount of data center capacity, also has a coal-heavy energy grid.
Energy consumption is not the only footprint
Another issue is the impact transfer, which means the shifting of environmental burdens. If data centers, for example, reduce their electricity consumption by cooling with water, the energy consumption is reduced at the expense of increased water consumption. Or materials that achieve better energy efficiency might lead to an increased material footprint during manufacturing.
The water footprint has become associated with data center water consumption in recent years. A byproduct of running servers is heat, and data centers need to keep hardware cool, which is achieved by using water. In 2023 researchers from the University of Texas and Riverside estimated that training OpenAI’s ChatGPT3 Large Language Model in Microsoft’s data centers used around 700,000 liters of fresh drinking water. In many cases, water is drawn from aquifers in regions that are already at risk of drought and water shortage. As data centers use so much electricity, the water consumed during the generation of electricity is itself significant. In the USA, around 40 percent of all the water drawn from lakes or rivers is used to cool nuclear and fossil fuel power plants during their operation, making it the largest single use of freshwater in the country.
And there’s the environmental impact from obtaining minerals to manufacture the electronics used in data centers. Various chemically and energy-intensive processes are needed to extract such minerals, oftentimes resulting in toxic tailing pools or other forms of waste, making land unusable. The waste from electronics is often unrecyclable and ends up in dumps. Even in Europe and the USA, where recycling is a common practice, electronics recycling rates are generally below 20 percent. Electrical waste is named by the Baker Institute at Rice University as the fastest growing waste stream globally.
Possible solutions for a more sustainable energy management
What aspects must be taken into account when addressing the increasing power consumption of AI?
Mandating low energy use and reduced resource consumption from digital infrastructure providers might be an effective measure as Big Tech has currently little incentive to follow a sustainability agenda. The companies’ commitments are mostly voluntary. Public and private investment plans should be tailored to privilege resource-efficient AI. Tech companies could ensure that upcoming data centers are only built when their energy demand is met with renewable energy that is locally produced and matched to the energy use of their data centers.
To be aware of the full scope of the problem, tech companies must be required to assess, measure, and reduce the environmental impact of AI along the value chain in alignment with the best available science and to make this assessment transparent. AI infrastructure providers must also disclose information about the development, resource consumption, and impacts of data centers before they are builtso that utilities and grid planners can accurately plan for future energy needs. Carbon offsets and other false solutions that do not genuinely reduce emissions should be rejected by regulatory authorities.
Tech companies must ensure that the mining for raw materials in their supply chain does not harm the environment or local communities, and ensure any new data centers built will not deplete water and land needed for people. The operators must release data on the environmental impact of their hardware development and transportation as well as information about thedisposal of hardware.