Is AI the key to nuclear renaissance?

AI and hunger for energy

We have already written about the revolutionary changes that artificial intelligence (AI) brings to decision-making processes in the nuclear sector, particularly in the domain of safety procedures. The application of AI can enhance operational efficiency and safety, as well as improve decision-making in the broader nuclear sector, including nuclear medicine, electricity production, and nuclear weapons management. However, a topic that needs to be addressed concerns the relationship between the development of AI technology and the increase in energy consumption. At first glance, this connection may seem tenuous, but deeper analysis reveals a strong causal relationship between these two processes.

The global acceptance and widespread use of artificial intelligence are greatly affecting worldwide energy demands and the environmental costs associated with reaching this level of technological advancement. As AI technologies become more sophisticated, their operational costs become more significant, contributing to energy-intensive growth that increases the environmental footprint. The reason for this is the vast expansion in the size and complexity of AI models, which demands intensive hardware and extensive data centre operations. These extremely high-energy consumption data centres are powered mainly by AI computation.

Training large AI models, particularly deep learning, requires vast amounts of computational power. These powerful models are trained on millions of datasets, often going through an iterative process, making it resource-intensive and energy-consuming in nature. This computational intensity tends to be higher with large-scale AI implementations. Most large-scale natural language processing models or computer vision applications depend on the availability of thousands of very powerful GPUs, if not more specialised hardware. All of this requires long, very energy-devouring training runs.

There is a direct correlation between the exponential increase in model parameters and the increase in the computational demand and, thus, higher energy consumption and carbon emissions. Energy demand is increasing due to the hardware infrastructure required to power AI, such as data centres and purpose-built processors, consuming large quantities of electricity for computation, cooling, and other operational needs. Data centres have already seen a 160% increase in energy use due to AI and cloud computing technology, putting immense strain on power grids. As of March 2024, there are about 11,800 data centres present in the world. Out of these, the USA took the lead with 5,381 data centres, followed by Germany (521), the United Kingdom (514), and China (449), with a constant tendency to increase their number. The data centre industry contributes 1 to 2 percent of global greenhouse gas emissions, and cloud computing is growing by 10 to 30 percent in terms of energy consumption each year. According to Morgan Stanley research conducted to gauge the expected growth in the global data centre industry, the industry will generate emissions equivalent to about 2.5 billion metric tons of carbon dioxide by the end of 2030. The same report indicated that Big Tech companies, such as Google, Microsoft, Meta, and Amazon, are the culprits in pushing for the establishment of data centres as they grow their capability in AI and cloud computing. It is important to emphasise that we are still in the early stages of AI development. For AI systems to evolve, they actually need more training that involves bombarding the software with data, which, according to some predictions, will lead to data centres consuming more energy than India by 2030.

In the global race for AI dominance, tech giants spare no effort in securing the necessary energy resources. However, they are simultaneously constrained by the imperative to seek sustainable solutions with minimal carbon footprints. As the most optimal solution to the energy-decarbonisation dilemma, nuclear energy has emerged as a compelling alternative.

Nuclear energy, decarbonisation, and energy consumption

Solar and wind power constitute the cornerstones of clean energy, yet they cease to produce power when the sun stops shining and the wind stops blowing. The intermittency of sun and wind clashes with data centres’ need for uninterrupted power. Nuclear energy is known as a steady energy source with an extremely low carbon footprint when compared to other sources of electricity. As stated by the United Nations Economic Commission for Europe (UNECE), the amount of CO₂-equivalent emitted per kilowatt hour of energy produced (gCO₂e/kWh) by nuclear power in its entire lifecycle is somewhere between 5.1 to 6.4 grams. These figures are lower than those set by wind, with 7.8 to 20 gCO₂e/kWh, and are significantly lower than those from coal-fired power plants, which range from 753 to 1,095 gCO₂e/kWh. Due to the non-burning of fossil fuels in its functioning, nuclear energy generates a minimal amount of carbon emissions. Some emissions from power plant construction, operation, and decommissioning do occur, but these emissions are relatively small when compared to those arising from other energy sources. Hence, these attributes make nuclear energy a very strong contender in the fight against greenhouse gas emissions and climate change.

AI is projected to contribute USD 15-20 trillion to the global economy by 2030, driven by rapid adoption and efficiency gains. But there is a catch. A single ChatGPT request uses about 2.9 watt-hours of electricity! Compared to that, the power required is tenfold that consumed by a Google search, and it is enough power to light a 60-watt incandescent bulb for 3 minutes. Advanced applications such as image, audio, and video generation push these demands even higher. Multiply that by millions of such requests in a day, and you will see the enormous amount of energy required to support an artificial intelligence system on a daily basis. Plans exist for gaining efficiency, as custom chips are being developed for handling specific workloads with greater efficiency than a traditional GPU. However, there is usually a great increase in computational and energy demand with each groundbreaking new model. The emerging appetite for generative AI, in part, could account for as much as 40 percent of the increased electrical requirements between now to 2030. Hyperscale facilities, such as Microsoft’s USD 100 billion ‘Stargate’ supercomputer, may require up to five gigawatts to operate. Nuclear energy, which provides emissions-free baseload electricity, has emerged as the preferred solution. Bill Gates, Jeff Bezos, Elon Musk, Mark Zuckerberg, Sam Altman, and other tech leaders argue that nuclear fission—and aspirational fusion projects—are critical to decarbonising AI infrastructure while maintaining growth.

Big Tech in the race for nuclear energy

By 2030, Microsoft commits to becoming carbon-negative, but its emissions jumped 30% from 2020 to 2023, largely due to the construction of new data centres for AI. Furthermore, by 2050, the company plans to eliminate from the environment all carbon it has emitted—both directly and through electricity usage—since its founding in 1975. Following the Chernobyl disaster, public confidence in nuclear energy experienced a general decline, despite a consistent global increase in the number of operational reactors. However, the convergence of heightened concerns over fossil fuel depletion and substantial pollution emissions has reinstated nuclear energy as a pivotal component in the contemporary energy policy discourse.

In Pennsylvania, Microsoft intends to refurbish the Three Mile Island nuclear power plant, which had a partial meltdown in 1979 and never operated again. A similar reactor located on the same island in the Susquehanna River was brought back online six years after the accident but was shut down in 2019 when its owner, Constellation Energy, could not secure any subsidies from the State of Pennsylvania. After the rapid development of AI, Microsoft made a deal with this company to reopen the reactor and get 100 percent of the electricity. Bill Gates founded the TerraPower company to invest in advanced nuclear energy to meet electricity needs and mitigate climate change. On the other side, tech giant Google is working with startup Kairos Power to launch advanced reactors in the next decade. Amazon is not far behind in its attempt to secure a nuclear power source for its sprawling data centres, having purchased a new data centre next to the Susquehanna Steam Electric Station in Salem Township. Amazon has pledged to invest in small modular reactors designed by xEnergy in Washington state. Meta’s request to put an AI data centre near a nuclear power plant encountered a problem with the discovery of a rare bee species at the site. But Mark Zuckerberg is definitely not giving up. Meanwhile, in Michigan, Holtec International remains on track to restart operations of the Palisades nuclear power plant in October 2025, with a plan to add 600 mW of small modular reactor capacity by 2030. Oracle is developing an AI data centre powered by three SMRs, which Larry Ellison calls ‘probably the most bizarre’ thing he has ever done but believes it must be done in order to satisfy AI’s ‘crazy’ energy demands. OpenAI CEO Sam Altman has backed nuclear startup Oklo Inc. to provide power for a Las Vegas-based data centre.

Leaders of global technological progress have realised that AI is doomed in its quest to change humanity without the introduction of nuclear energy as a primary source of energy. Tech companies have a vision of the world based on AI with a clean and steady power source, so the symbiosis between nuclear technologies and AI is something that will constantly strengthen in the prospective future.