Since generative AI models first debuted last year, its pace of innovation has been relentless. Models like ChatGPT and Sora have sprung up in rapid succession, significantly enhancing work efficiency and transforming people’s daily lives. However, this technological revolution has also spotlighted a crucial issue: the escalating energy demand of AI servers. According to the International Energy Agency (IEA), global data centers consumed an astonishing 460 terawatt-hours (TWh) of energy in 2022. By 2026, this figure is expected to surge to between 620 and 1,050 TWh, an increase equivalent to Sweden's annual energy consumption.

In response to this surge in energy demand, tech companies are actively pursuing reliable and clean energy solutions. At the recent Davos Forum, Sam Altman, the CEO of OpenAI, emphasized the urgent need for significant breakthroughs in the energy sector to support the future growth of AI. One potential solution to this looming energy crisis, in Altman’s opinion, is the use of nuclear fission technology.

Nuclear fission, as an efficient and clean form of energy, holds immense potential to bridge the gap in energy supply. However, the use of nuclear energy also comes with a series of risks and concerns. Issues such as nuclear radiation leaks and the storage and disposal of nuclear waste have become significant challenges that hinder the large-scale application of nuclear power. In this edition of Poseidon’s Market Foresight, we will delve into the advantages and risks of nuclear energy, as well as explore its feasibility as a power supply for the AI revolution.

History of Nuclear Energy

The history of nuclear power generation dates back to the early 20th century. In 1938, the ground-breaking discovery of nuclear fission by the German scientist Otto Hahn laid the foundation for nuclear power generation. Hahn's discovery revealed that when a neutron collides with a uranium nucleus, it can split into two lighter nuclei and release tremendous energy. This breakthrough opened the door to practical applications of nuclear energy. Subsequently, research and applications in nuclear energy advanced rapidly during the Second World War. In 1954, the Soviet Union established the world's first commercial nuclear power plant, the Obninsk Nuclear Power Plant.

Since then, with continuous technological advancements, nuclear power generation has been widely adopted globally. Currently, 32 countries worldwide are utilizing nuclear power plants for electricity generation, with the majority located in Europe, North America, and East Asia. Nuclear power accounts for approximately one-tenth of the global electricity production. According to the International Atomic Energy Agency (IAEA), the United States is currently the largest producer of nuclear power, with nuclear power generation accounting for 18% of its total electricity generation. In Europe, several countries, including France, Slovakia, Hungary, and Belgium, have a nuclear power generation share exceeding 40%. The reason for these countries' heavy reliance on nuclear power generation is their scarcity of traditional energy resources such as oil, gas, and coal. Nuclear fuel, with its high energy density, stable operation, and low greenhouse gas emissions, provides an effective pathway for these countries to achieve energy independence.

Globally, the primary source of electricity still comes from fossil fuels, particularly coal and natural gas, accounting for about two-thirds of the total. Over the past few decades, renewable energy sources such as solar, wind, and hydroelectric power have made significant progress, with their share in electricity supply increasing steadily from 20% in 1985 to approximately 30% currently. However, despite the increasing share of renewable energy, the proportion of nuclear power generation has shown a consistent decline, falling from 15% in 1985 to below 10% currently. This trend can be attributed to several factors.

First of all, safety risks, low operational efficiency, and high maintenance costs associated with aging nuclear power plants have prompted some countries to shut down these units to avoid potential risks and losses. Moreover, the high construction and operating costs of nuclear power plants, coupled with long construction cycles, have all limited their application in developing countries.

Furthermore, public concerns about nuclear safety are also a significant factor contributing to the decline in the proportion of nuclear power generation. Historical nuclear accidents, such as the Chernobyl and Fukushima, have resulted in public skepticism about the safety of nuclear power plants. Risks such as potential fuel rod leaks, mishandling of nuclear waste have garnered widespread public attention and concern.

Nuclear Power Revival Sparked by the AI Industry

With the increasing threat of climate change, governments worldwide are re-evaluating the importance of nuclear energy as a potential solution for stable power generation. At the 28th United Nations Climate Change Conference (COP28), countries unanimously agreed to accelerate the phase-out of fossil fuels in order to achieve net-zero emissions by 2050 or earlier. To achieve this goal, the conference called on the global community to accelerate the development of zero- and low-emission energy technologies, including nuclear and renewable energy. This was the first time that the conference recognized the important role nuclear energy can play in reducing global carbon emissions. At the same time, 24 countries led by the United States, the United Kingdom, and France jointly signed the Declaration to Triple Nuclear Energy, aiming to triple nuclear power generation capacity by 2050.

However, accomplishing a threefold increase in nuclear power generation capacity is not an easy task. It requires the governments that have signed the declaration to accelerate constructions of new nuclear power plants and make substantial financial commitments. Given that existing nuclear reactor projects often face long-term construction delays and significant cost overruns, many experts are skeptical about whether the target can be achieved. In addition, it is estimated that by 2050, 270 nuclear reactors will need to be shut down as they approach the end of their lifespan, which means that 10 new reactors need to be built each year just to maintain the current nuclear power generation capacity. To achieve a threefold increase in nuclear energy capacity, not only is it necessary to increase capital investment, but also significantly accelerate the pace of nuclear power plant construction.

Currently, traditional energy sources still dominate the global energy supply. To completely transition away from fossil fuels by 2050, substantial development of alternative clean energy sources is imperative to bridge the gap between supply and demand. At the same time, with the escalating utilization of AI, global electricity demand is set to surge, posing profound challenges to the energy infrastructure.

Based on the current trend of AI development, analysts predict that by 2027, the annual shipment volume of global AI servers is expected to reach 1.5 million units. These servers, operating at full capacity, will consume vast amounts of electricity. Research conducted by VU Amsterdam University in the Netherlands reveals that when 1.5 million AI servers operate at maximum capacity, they will consume a minimum of 85 terawatt-hours of electricity annually, equivalent to the annual energy demand of the Netherlands. This would require the entire output of about 15 average-size nuclear power plants. In addition to the huge energy consumption required for data center operations, cooling the data centers to ensure their stable operation, as well as the energy required to produce AI chips, will all pose significant challenges to future energy supply.

Currently, it is widely recognized that fossil fuels alone cannot satisfy the escalating energy demands of AI, necessitating a pivotal role for non-fossil energy sources in the advancement of computing power. In response to this, major technology firms have embarked on an arms race to secure non-fossil energy resources.

Microsoft publicly recruited a nuclear technology expert last September, with the relevant position titled "Principal Program Manager for Nuclear Technologies." This role will be responsible for refining and implementing energy strategies for global small modular reactors (SMRs) and microreactors, leading technical evaluations for the integration of SMRs and microreactors, and powering Microsoft's cloud and AI data centers. Nuclear energy is inexpensive, and unlike other technologies such as solar energy, it can generate electricity continuously and stably 24 hours a day. Furthermore, nuclear energy is clean and can produce energy without emitting much greenhouse gases, which will help reduce the carbon footprint of data centers. Sam Altman has invested heavily in two nuclear energy startups, Helion and Oklo. Helion, which focuses on nuclear fusion, has signed a power supply agreement with Microsoft, while Oklo, which focuses on nuclear fission, has completed its public listing.

Tech giants such as Microsoft, Google, and Amazon have all invested in nuclear energy development. Despite its many benefits, nuclear energy can also pose a series of new challenges in the handling of radioactive waste and the supply chain for uranium. If cloud computing companies hope to rely on SMRs to power their data centers, they will need more high-assay low-enriched uranium (HALEU) fuel than traditional reactors. Currently, Russia has been a major global supplier of HALEU. The United States is pushing to establish a domestic uranium supply chain, but communities near uranium mines and refining plants are already protesting. Additionally, the long-term disposal of nuclear waste remains unresolved.

In addition to nuclear energy, major technology companies are also turning their attention to new and renewable energy sources such as solar, wind, geothermal, and hydrogen. Major tech companies have invested in hundreds of solar and wind energy projects. However, due to their inability to provide stable power around the clock, companies are exploring new energy alternatives. Geothermal and hydrogen are promising clean energy sources, but they are not yet widely used. Google began developing new geothermal energy in 2021 as part of its goal to achieve zero-carbon emissions by 2030. This initiative, in partnership with startup Fervo, has begun powering two Google data centers at the end of last year. Fervo's new geothermal technology involves drilling near geothermal rocks to introduce heated water sources to generate electricity, although the current power supply is still relatively small. In terms of hydrogen energy, companies such as Caterpillar and Honda have announced collaborations with technology companies to supply hydrogen fuel cells as clean backup power sources for data centers. Honda recently began a hydrogen energy pilot project at a factory near Los Angeles International Airport, powering a small data center.

Conclusion

With the rapid advancement of AI technology, the short-term obstacle to its further progress lies in computing power, while the long-term challenge revolves around energy supply. Fueled by the COP28 Conference, countries have converged on the pivotal role of nuclear energy in the shift towards clean energy. Nuclear energy’s role in the AI revolution will continue to evolve. Whether it is through significant technological breakthroughs by nuclear energy startups or increased investment and support from governments, these trends have the potential to drive the future development of nuclear energy. Whether nuclear energy and AI will form a mutually beneficial partnership, or nuclear energy will merely provide limited support in addressing AI's substantial energy consumption, remains to be seen.