Does AI need nuclear energy?

Data centers are the backbone of AI. They provide the vast computational power required to train and develop complex models. However, the energy requirements associated with this fast-growing industry pose critical challenges to our existing energy production systems. But what exactly are we talking about?

How much power does AI really need?

AI data centers need significant power due to the high energy demands of processing AI workloads. These workloads involve executing complex calculations necessary for training and running models on massive datasets, which require substantial computational resources. Alongside this, cooling systems are crucial, as hardware that consumes vast amounts of energy also generates considerable heat, making efficient and robust cooling essential for operational stability. In addition to the computing power, data centers rely on consistent and accessible storage, which increases the energy demand, as does the infrastructure supporting these storage systems. Furthermore, high-performance networking equipment is needed to transmit large volumes of data quickly, contributing more to the overall energy consumption.

However, not all AI models seem to be equally power-hungry. A breakthrough example is DeepSeek, a Chinese AI startup that has developed an AI model operating at a fraction of the energy cost compared to Western counterparts. While traditional AI models have driven up global electricity consumption, DeepSeek’s innovation shows that AI doesn’t have to be an environmental catastrophe. DeepSeek’s model was trained for just $5.5 million, using far fewer resources than models developed by tech giants like Meta, Microsoft, or OpenAI. This challenges the assumption that AI’s growing computational needs must lead to a proportional rise in energy consumption.

Despite these advancements, the rapid deployment of AI still presents a potential climate risk. Even with more efficient models like DeepSeek’s, the phenomenon of Jevons' Paradox suggests that efficiency gains may not reduce infrastructure spending but could instead fuel greater AI growth. This paradox raises the question of whether AI’s energy demands will ever truly decrease, or if they will continue to rise as efficiency improves.

As AI expands into sectors like healthcare and finance, the demand for processing power intensifies, putting greater pressure on the need for reliable and stable energy sources. According to the Electric Power Research Institute (EPRI), data centers could consume up to 9% of the electricity generated annually in the U.S. by 2030. This projected increase not only puts immense pressure on existing energy systems but also exacerbates environmental concerns related to high energy consumption.

Is a sustainable and stable power input possible?

While some data centers have begun integrating renewable energy sources like solar and wind, these solutions still have some limitations. Solar power generation is largely dependent on weather conditions and time of day, while wind energy can be unpredictable. However, new developments in energy storage technologies, such as advanced battery systems and grid-scale storage solutions, are helping to mitigate these challenges, allowing for more reliable power supply. Despite these advancements, data centers still require a constant and stable power supply to operate 24/7. Without reliable energy inputs, they risk downtime, potentially disrupting critical operations and incurring significant financial losses.

In addition, both sources often require extensive energy storage solutions to manage supply and demand effectively. The transition to renewables can involve significant infrastructural changes and investments, which may not keep pace with the urgent energy demands of AI.

To date, many data centers remain largely powered by fossil fuels, which are stable energy sources but contribute significantly to climate change through high carbon emissions. This reliance conflicts with the sustainability goals set by major tech companies and governments. Even as AI technology becomes more energy-efficient, the launch of innovations like DeepSeek, AI models reported to reduce computing power per query by 90%, raises doubts about how AI’s energy demands will evolve. As enterprise AI spending rises, energy demand could surpass renewable growth, locking in emissions and hindering climate goals. 

The challenge of nuclear energy

Nuclear energy appears a compelling solution to the energy demands of AI data centers. It produces large amounts of energy with minimal carbon emissions, making it a reliable source for the high and consistent power needs of data centers, also aligning with global decarbonization goals. 

However, traditional large nuclear reactors present several challenges, including safety concerns, public apprehension, and waste management issues. The safety of these reactors is a major concern, largely due to high-profile accidents like Chernobyl in 1986 and Fukushima in 2011, which have deeply influenced public perception of nuclear energy. Beyond the risks of accidents at nuclear facilities, there are growing concerns about the vulnerability of nuclear infrastructure, especially in today's geopolitical landscape. Recent conflicts, such as the war in Ukraine, have intensified these concerns, for example, in April 2024, the Zaporizhzhia Nuclear Power Plant in Ukraine was targeted by drone strikes, underscoring the risks that military activities pose to nuclear facilities. In addition, transportation-related incidents have raised public fears due to the potential for widespread contamination. The International Atomic Energy Agency reports that more than 50% of all thefts involving nuclear and other radioactive materials since 1995 have been linked to transportation-related incidents, further amplifying fears about the security of radioactive materials. 

Managing nuclear waste is another major challenge, as the byproducts of nuclear reactions, such as spent fuel rods and other radioactive materials, can remain hazardous to humans and the environment for thousands, or even tens of thousands, of years. These materials require careful handling and containment to prevent contamination of soil, water, and air. Finding secure, long-term storage solutions has proven difficult.Local communities and environmental groups often oppose proposed nuclear waste storage sites due to concerns about potential leaks, transport accidents, and the long-term reliability of containment systems. For instance, in the United States, the challenge of convincing people that it's safe to live near nuclear waste storage facilities is a significant social issue. 

Additionally, these reactors are expensive to build, often requiring billions of dollars in investment, and can take a decade or more to complete due to complex regulatory and construction requirements. The combination of high costs and waste management issues makes it difficult to establish these reactors, especially in regions that need immediate energy solutions or lack the financial resources to invest in such large-scale infrastructure.

The advantages and drawbacks of SMRs

The solution may come from Small Modular Reactors (SMRs). These advanced nuclear reactors are smaller, cheaper, and more compact compared to traditional nuclear power plants. They are designed to be factory-built, easily transported, and assembled on-site in modular units. While their lower emissions align with global climate objectives and could help address the energy demands of AI, the broader implications for energy systems remain complex. Investing in SMRs may create jobs within the energy sector, spanning construction to plant operation. Additionally, when integrated with renewable energy sources like solar, wind, and hydropower, SMRs could contribute to a more diverse energy mix. However, several challenges, including cost-effectiveness and public perception, still need to be addressed.

One of the primary challenges is the issue of economies of scale. Large reactors usually have cost advantages that make them cheaper than smaller ones. This could make it harder for SMRs to compete on a market level. Additionally, many SMRs’ designs are still in the development stage, which means there is limited operating experience to assess their long-term performance and reliability. The lack of established track records can create uncertainty for investors and operators, which may interfere with widespread adoption. One last challenge is public perception: despite its safety advancements, nuclear energy often faces skepticism and resistance due to historical accidents and the potential risks involved. In the UK, for example, environmental groups such as Greenpeace and Friends of the Earth have criticized plans to expand nuclear power, including the implementation of SMRs, arguing that these initiatives are costly, untested, and reliant on government support.

Lastly, SMRs face the same challenges as traditional nuclear reactors when it comes to waste management. The safe storage and long-term disposal of radioactive waste remains a significant hurdle, and addressing these concerns will be critical for the widespread acceptance and success of SMRs.

What does the future hold?

The interplay between AI, energy demands, and environmental impacts will be critical in shaping the future of data centers and broader energy and decarbonization policies. The adoption of SMRs may offer a promising solution to alleviate energy constraints and improve the overall resilience of energy systems. Growing interest in this technology is evident in major tech companies like Amazon and Google, which have recently signed significant deals to incorporate SMRs as a power source for their data centers. This not only shows their belief in the feasibility of SMRs but also highlights their efforts to secure approval for these potentially safer and more affordable reactors.

However, as tech giants begin to produce their own energy, a new question emerges: who will control the energy market if large corporations take on energy production themselves? This shift could have significant implications for the broader energy landscape, potentially concentrating power in the hands of a few major players.

The transition to SMRs and other innovative energy solutions will require collaboration among big tech companies, governments, and the nuclear industry. Policies that promote nuclear energy research, development, and deployment, alongside incentives for renewable energy investments, will be essential. Education and outreach initiatives can also help mitigate public fears and encourage greater acceptance of nuclear technologies. While society must carefully weigh the pros and cons, the sustainability and reliability offered by nuclear energy position it as a strong option. 

Are we entering an era of AI powered by nuclear energy? The question remains open, but one thing is certain: finding innovative energy solutions will be crucial for AI and for a sustainable and resilient future.

Check Data Hub™ for the sustainability performance of the world's most polluting companies based on their data center operations: Amazon (17.1 mtCO₂), Google (8.1 mtCO₂), Microsoft (3.5 mtCO₂), Meta (50.6 mtCO₂), Samsung...

illuminem Voices is a democratic space presenting the thoughts and opinions of leading Sustainability & Energy writers, their opinions do not necessarily represent those of illuminem.