The ongoing development of artificial intelligence may become a significant problem for the global energy sector. Will the problems faced by the technology industry in terms of energy be solved by the era of nuclear renaissance? Or perhaps the funds invested in nuclear energy will never be realised?
It is worth starting our story with a few important dates.
- 12 July 1957: At the Santa Susana Field Laboratory in California, the SRE reactor, the first nuclear power source to supply electricity to American homes, is put into commercial use. This inconspicuous 20 MW device will be the first of more than 100 commercial nuclear reactors to be built in the United States by the end of the century.
- 28 March 1979: The Three Mile Island nuclear power plant (Pennsylvania) experiences the worst nuclear disaster in US history. The core of reactor number 2 partially melts down. No one was hurt, but the reactor never produced energy again. Reactor number 1, which was not involved in the disaster, continued to produce electricity until it was shut down in 2019.
- 27 May 1996: The last American nuclear power plant of the 20th century, Watts Bar in Tennessee, is commissioned. For the next 20 years, no commercial nuclear reactors will be built in the United States.
- 16 January 2020: Microsoft announces that by 2030, the company will produce not only zero but negative carbon footprint. In summary, they call this ambitious environmental strategy a ‘moonshot’.
- 30 November 2022: OpenAI launches ChatGPT. The premiere of the language model marks the beginning of the rapid development of AI-based tools. The next several months see a real technological leap forward. Industry giants such as Google and Microsoft launch their own AI-based services. The growth of the sector is accompanied by a huge demand for energy related to data centres.
- May 2024: Microsoft CEO Brad Smith states outright that the company’s goal of a negative carbon footprint by the end of the decade is no longer realistic. Referring to the metaphor of ‘shooting for the moon,’ he admits that in four years, ‘the moon has moved.’
- September 2024: Microsoft announces the signing of an agreement with Constellation Energy, the company that manages the inactive Unit 2 at the Three Mile Island power plant. Under the agreement, Constellation Energy will reactivate the unit, which failed nearly half a century earlier, and Microsoft will receive the energy produced by the reactor. The electricity will be used to power the tech giant’s data centres.
Nuclear energy at a turning point. How did the atomic dream end?
Between 1957 and 2000, more than 100 commercial nuclear reactors were opened in the United States. From 2000 to the present: only three. The situation in the American energy sector reflects a global trend. The nuclear era was coming to an end, new projects were no longer being developed in a matter of years, and the process of obtaining the necessary permits and designs could take a dozen or even several dozen years. As Poles, who have been waiting decades for the first nuclear power plant in our country, we know this best.
One megawatt of installed capacity in nuclear power costs between PLN 21 million and even PLN 43 million. The first Polish power plant of this type is to have a capacity of 3.75 GW and cost approximately PLN 115 billion. This gives us PLN 30 million per megawatt. For comparison, the gas unit at the Ostrołęka power plant (currently in the start-up phase) with a capacity of 745 MW cost PLN 2.85 billion, or PLN 3.8 million per megawatt.
Today, such costs in nuclear energy are the rule rather than the exception. Investments of this type almost always cost more than originally estimated. For example, the Finnish Olkiluoto 3 unit was supposed to cost EUR 3 billion, but its price ultimately rose to EUR 11 billion. The costs of building the French Flamanville 3 were initially estimated at EUR 3.3 billion. The final amount, after taking into account interest and inflation, is approximately EUR 23.7 billion.
Read more: Working as a wind turbine technician: how to minimise risks
Huge costs and long deadlines
There are several reasons for this. The construction of a nuclear power plant is a very complex technological undertaking. Today’s reactors are much safer than those that failed at Three Mile Island or Chernobyl. However, it is still necessary to ensure their safe installation and operating conditions, and this requires costs. The risk in the event of a potential failure is too great not to spend huge sums of money to minimise it. The entire process of construction and technology development is therefore quite tedious and subject to numerous controls.
The second factor is time. The construction of a nuclear power plant cannot begin overnight; first, all the formalities must be completed, building permits and environmental approvals must be obtained. It is also often necessary to change the law through appropriate acts or ministerial regulations. And as time passes without the investment being realised, its cost increases, partly due to inflation.
This is also the reason why many companies are reluctant to get involved in the construction of a nuclear power plant. In order to decide to sign contracts, invest funds, engage labour to create designs and documentation, and then wait several years before the time finally comes for implementation and real returns, one must have really large financial resources. Not every company is able to take such a step. It is not without reason that few nuclear projects are created without any financial support from the state. However, this may soon change.
‘The moon has moved’, or artificial intelligence at an energy turning point
‘The moon has moved’. The words of the Microsoft CEO quoted at the beginning of this article perfectly illustrate the situation in which the global energy sector finds itself in the era of rapid AI development. Most countries in the global north seemed to be on a straightforward, albeit sometimes bumpy and long, road to environmental neutrality. Increasingly intensive investments in renewable energy sources, modernisation of the energy grid, bold corporate strategies for a negative carbon footprint within a decade. Of course, there were complications, and few people really expected that most of the plans and promises would be fulfilled on time, but the horizon of truly clean energy and industry was visible.
And then came the development of generative AI. This type of artificial intelligence must perform thousands of complex calculations to create text, images or video material. This happens in large data centres, which tech giants are increasingly eager to build around the world. Huge amounts of electricity are needed to power the servers that run large language models. How much exactly?
AI like a small country
Sources disagree on this. According to a 2024 EPRI study, a single ChatGPT query consumes 10 times more electricity than a single Google search. Since this data was compiled, the search engine owner has already implemented an AI-generated summary, displayed at the top of the page for almost every search, so it can be assumed that this result has levelled out upwards. Interestingly, Dutch scientist Alex de Vries already considered such a scenario in 2023. In his study, he writes that once AI queries are integrated into every Google search, the company will consume approximately 22.8 TWh of energy per year. This is comparable to the energy consumed by Ireland. De Vries also states that, depending on the model and complexity of the command, a single query to artificial intelligence consumes between 3 and 9 watt-hours.
Read more: How to cool production halls without air conditioning and save hundreds of thousands annually?
More and more people are using artificial intelligence. Companies are eager to integrate language models into their daily operations, and generative AI is used in almost all industries. Independent artificial intelligence agents are even being created that communicate with other models completely on their own. Such rapid growth in demand for AI creates a huge demand for energy. According to estimates by the World Economic Forum, as early as next year, the entire artificial intelligence sector could consume as much as 4% of the world’s electricity, which is roughly the same as Japan.
Companies themselves, especially those in the technology industry, also admit this. Microsoft’s ‘moon shift’ and the need to revise its plans for carbon neutrality is just one example. In 2024, Google admitted that its emissions had grown by 48 per cent over the past five years, with the main reason being the increase in energy demand associated with the rapid growth of the AI sector.
AI and nuclear energy. Has the technology industry found a solution?
Tech giants need to get their electricity from somewhere to maintain their increasingly integrated AI infrastructure. Nuclear energy seems to be the ideal direction for development here. There are several reasons for this.
Firstly: ecology. Nuclear energy, although not fully renewable, is still many times more environmentally friendly than fossil fuel-based sources. The energy efficiency of radioactive elements leaves coal and gas far behind. That is why nuclear energy is now widely considered a low-emission and environmentally friendly source of energy.
Secondly: resources. We mentioned earlier how high the costs of building reactors are, which is a major problem for nuclear energy. Few companies can afford such expenses. It is possible that the tech giants belong to this select group. The enormous fortunes built by the owners of Google and Facebook could provide solid financial security for future nuclear investments.
Will nuclear power be able to power AI?
The combination of AI and nuclear energy seems to be a perfect match. However, there always has to be a ‘but,’ and this case is no different.
The biggest problem seems to be time. Big tech companies have already come to terms with the fact that the zero-emission strategy of just a few years ago will have to be shelved. However, just because the ‘moon has moved’ does not mean that it has completely disappeared from view. It may take many years to make AI an energy-efficient industry. However, as mentioned above, nuclear reactors are not usually built quickly.
Unfortunately, time is working against companies in the artificial intelligence industry. Environmental issues aside, there has been talk for some time that the inflated AI bubble will soon burst. Tech giants have poured millions of dollars into this segment, but public scepticism and the slowdown in technological development are calling these investments into question. In addition to Microsoft, Meta and Alphabet (Google) have also invested in nuclear energy to power AI. But when nuclear power plants connected to large server rooms finally start operating in a few years’ time, will they have anything to power? Or will the moon move again, this time in a completely different direction?
