Nuclear Energy Option for Powering AI -- Is it safe and clean?

The rapid growth of AI, particularly Generative AI (GenAI) in the recent years has accelerated the demand for energy. This is the result of a very high and growing demand for computational power, the need for which arises during training and fine-tuning the GenAI models as well as for using the model for inferencing (responding to the prompts). According to a report by the World Economic Forum (WEF), the computational power required for AI is doubling every 100 days. To meet this increasing computational power requirement, massive data centres are needed, which require energy for cooling, power supply and server operations. According to one estimate, the energy consumption by data centres will triple by 2030. The demand for energy would be much higher if there were no constraints, as at present, on the availability of AI chips and data centre equipment. Thus, the world appears to be heading towards an energy crisis as a result of all the energy-intensive activities related to AI. Almost all the energy required for this purpose comes from fossil fuel based power generation which has a very high carbon footprint and which therefore adversely impacts sustainability of the planet.

While the threat to sustainability is critical and of concern to the companies developing newer and more powerful AI models, the likely energy crisis in the near future is a matter of equal or probably a greater concern for them. The Big Tech companies seek urgent solutions to this challenge as it is akin to an existential risk to them. Apart from the sheer quantum of energy required for AI, it is important to have an uninterrupted supply of concentrated and consistent energy with high power density, since AI data centers require steady power 24X7. To achieve this goal, it is necessary for them to explore other sources of energy. The renewable energy sources such as solar and wind energy are not found to be suitable because these are unable to provide reliable, uninterrupted high density energy suited to the needs of AI. The AI industry is therefore seriously exploring the nuclear option to supply energy of the kind that can meet its demand as well as be a source of clean energy to address the sustainability challenge.

Pros and Cons of Nuclear Option

Some of the advantages of using nuclear energy for meeting AI’s energy demand are as follows:

a) A nuclear reactor can provide a massive amount of reliable, predictable, stable and scalable, 24/7 power supply consistently to run energy-hungry AI applications.

b) Nuclear energy has virtually zero greenhouse gas emission impact. Therefore, its environmental impact is drastically low and it is considered a clean energy option.

c) Due to the compact fuel and modular plant, nuclear reactors provide substantial flexibility to generate power incrementally. Thus, scaling up or down to meet changing demands for energy is facilitated.

d) Nuclear energy can be integrated with other energy sources to boost reliability and sustainability.

However, the disadvantages include the following:

a) The upfront costs in building the requisite infrastructure are very high. The costs can be three to four times higher than that of solar or wind power plants of similar capacity.

b) It can take decades to install nuclear power plants owing to the need for absolutely safe design and complexity of the nuclear reactors. Regulations aimed at fool-proof security and safety of the nuclear energy plants, significantly contribute to expanding the timeline to construct these plants and make them operational.

c) The nuclear plants require uranium as a fuel, which is a scarce resource. Therefore, the possibility of fuel shortage in the future can be very high.

d) Nuclear waste created during the operation of nuclear plants leads to a risk of radioactive emissions as no satisfactory solution has yet been found for safely storing the nuclear waste.

Public Perception regarding Safety of Nuclear Plants

Apart from the havoc caused by the dropping of nuclear bombs in Japan in the second world war, the history of mankind has witnessed at least three major nuclear accidents. These are the Chernobyl nuclear plant accident in the erstwhile Soviet Union (1986), the Three Mile Island nuclear disaster in the US (1976) and the Fukushima Daiichi nuclear accident in Japan (2011). Such nuclear disasters have created a fear of the nuclear energy among general public on issues such as power plants melting down and highly radioactive nuclear waste, which can be very harmful to human kind. In the unfortunate possible incidents of accidents or deliberate sabotage, the loss to human life can be catastrophic due to the emission of unsafe radiations. The nuclear plants can be vulnerable to malicious attacks due to terrorism and acts of war. Therefore, the public perception and concerns about nuclear safety, nuclear waste management, and nuclear proliferation cannot be ignored.

However, despite horror stories from the past and the potential safety risk, nuclear energy is being viewed as a viable option, thanks to technological advances in design of nuclear reactors and better safety systems in place. A key technological development is the advent of small modular reactors (SMRs). These reactors are smaller than traditional reactors in terms of their energy generation capacity, but can be constructed faster, since these are planned to be assembled from prefabricated parts. The SMR based nuclear plants are more efficient, safer, scalable and faster to deploy and their modular design provides a greater flexibility. However, a counter-argument is that SMRs may produce more nuclear waste, use fuel less efficiently and may be less safe than conventional nuclear plants. Also, these are unlikely to be cheaper than conventional plants.

Nuclear Energy Ventures by Big Tech companies

In the recent years, big tech companies such as Google, Amazon and Microsoft have made major investments in nuclear energy ventures. Google has partnered with Kairos Power to set up nuclear plants using SMRs to power its AI data centers. Amazon has signed agreements with Energy Northwest and Dominion Energy to develop SMR projects and is also planning to build a data center next to Talon Energy’s nuclear facility in Pennsylvania. Microsoft, in collaboration with Constellation Energy, plans to revive the traditional Three Miles Island nuclear power plant in the US. The development of nuclear energy options for meeting energy needs of AI is being actively supported by the US Govt.

The International Atomic Energy Agency (IAEA) has projected a total investment of nearly US$100 billion to meet the surging demand for nuclear energy by 2050. Though this is a small fraction of the global energy infrastructure investment, it is a big jump from the current levels of investment in nuclear energy owing to the massive demand solely attributable to AI.

Other Options of Energy Sources

Other options being explored by the AI industry include advanced nuclear fission, geothermal energy and nuclear fusion as these options can also match AI’s demand for concentrated power, 24X7 reliably. While the nuclear fusion technology is under development, advanced fission and geothermal energy options are ready for deployment at small scale. It has been reported for instance that Sam Altman of OpenAI is investing into start-up companies engaged in developing next generation nuclear fission and nuclear fusion technologies.

Conclusion

AI has an insatiable demand for energy and the nuclear energy option appears to be a great strategic move. However, it is also important to address public concerns about the perils of nuclear technology. At the same time, energy efficiency and innovation in AI development are crucial so that energy demand can be moderated. Given the rapid pace at which AI is developing, it is quite conceivable that new innovations in AI may substantially reduce the demand for energy in the future. AI may also be helpful in bringing about innovations to enhance safety and optimize operations of the nuclear power plants. Thus, both AI and nuclear energy options to power AI can have a symbiotic relationship and both need to develop concurrently.

References

1. Britney Nguyen, “A look at the energy companies OpenAI's Sam Altman is backing”, Quartz, October 21, 2024, https://qz.com/nuclear-energy-helion-exowatt-oklo-sam-altman-ai-1851676383

2. Carl Franzen, “Why AI is going nuclear?”, Venture Beat, October 18, 2024, https://venturebeat.com/data-infrastructure/why-ai-is-going-nuclear/

3. Dan Meyer, “Amazon, Google pick nuclear option to power AI data center demands”, SDX Central, October 22, 2024, https://www.sdxcentral.com/articles/analysis/amazon-google-pick-nuclear-option-to-power-ai-data-center-demands/2024/10/

4. Danielle Rios , “Why AI’s future might depend on nuclear energy”, Telcodr blog, April 23, 2024, https://telcodr.com/insights/ai-and-nuclear-energy-blog/

5. Davide Castelvecchi, “Will AI’s huge energy demands spur a nuclear renaissance?”, Nature.com. October 25, 2024, https://www.nature.com/articles/d41586-024-03490-3

6. Jason Bordoff and Jared Dunnmon, “America’s AI Leadership Depends on Energy”, Foreign Policy, October 14, 2024, https://foreignpolicy.com/2024/10/14/ai-artificial-intelligence-energy-demand-nuclear-solar-wind/

7. John Moore, “Generative AI energy consumption grows, but ROI role unclear”, TechTarget, Sep 30, 2024, https://www.techtarget.com/searchcio/feature/Generative-AI-energy-consumption-grows-but-ROI-role-unclear

8. Lucas Mearian, “Desperate for power, AI hosts turn to nuclear industry”, Computerworld, July 8, 2024, https://www.computerworld.com/article/2514119/desperate-for-power-ai-hosts-turn-to-nuclear-industry.html

9. Matt Marshal, “Meta engineer: Only two nuclear power plants needed to fuel AI inference next year”, Venture Beat, November 13, 2023, https://venturebeat.com/ai/meta-engineer-only-two-nuclear-power-plants-needed-to-fuel-ai-inference-next-year/

10. Michael Kearney and Lisa Hansmann, “Why artificial intelligence and clean energy need each other”, Technology Review, October 8, 2024, https://www.technologyreview.com/2024/10/08/1105165/why-artificial-intelligence-and-clean-energy-need-each-other/