Amazon, Google and Microsoft are betting the future of artificial intelligence on nuclear power. It will be difficult

They say necessary strength. And if there’s one thing AI needs (besides the data to train it, of course), it’s energy. Lots of energy. Big tech companies are learning this the hard way. Their gigantic investments in new data centers are generating enormous supply, and many of them already have a clear idea of ​​how to solve this problem. With nuclear energy.

Amazon. Today we learned how AWS has signed an agreement with Dominion Energy to develop SMR (Small modular reactor) in Virginia (USA). This will not be the only project of this type: $500 million will be invested in just three projects in different US states. Each of them can generate 320 MW of electricity, so if the three projects become reality, the total capacity will be 960 MW.

Google. This week we also learned that Google has reached an agreement with Kairos Power to build seven SMR nuclear reactors by 2030. The project will add 500 MW of carbon-free electricity and power artificial intelligence data centers. The idea is to go even further: the company could commission more SMR reactors before 2035.

Microsoft. And be careful, because they are not the only ones. Microsoft has reached an agreement with Constellation Energy to reopen the Three Mile Island nuclear power plant, which has been closed since 2019. Constellation Energy is investing $1.6 billion to restore and modernize the plant. Once restored, the power plant will transfer its 835 MW of power to Microsoft data centers, which will, of course, need it to implement their artificial intelligence projects. In 2023, the company has already invested in a bet on the future: Sam Altman-backed Helion Energy and its promise of nuclear fusion.

Will SMRs become a cure? The energy consumption of data centers of companies such as Google or Microsoft is enormous: recent study shows that in both cases it amounted to 24 TWh, a figure close to the full annual consumption of countries such as Slovakia or Ecuador. The numbers used in these projects – 500 to 1000 MW – will undoubtedly help cover future needs in the short to medium term, but there is another problem.

There is a rush. These projects offer promising solutions, but these plants will not be available immediately. We have Microsoft’s example: Three Mile Island Nuclear Power Plant won’t be ready for at least four years. In China and South Korea, construction and commissioning of a new nuclear power plant takes about five years, and they are the fastest countries in this type of project.

Doubts about the future. It’s not entirely clear whether there are enough of them: a recent Goldman Sachs study found that US data centers will triple energy demand between 2023 and 2030. This will require the provision of 47 GW more energy. Some of this will no doubt come from renewable energy, and these projects will help, but it is likely that we will see more investment (public and private) in this area to meet this demand.

Love for nuclear power renewed. After years of bad publicity and a pessimistic vision of the future of nuclear power (in Spain we keep our heads down), many countries have resumed projects to build more plants of this type. The United States is a good example: it wants to launch 300 new modern reactors by 2050. The aforementioned energy needs driven by the development of generative artificial intelligence have sparked a sudden love affair between big tech and nuclear power.

Astronomical predictions. Global electricity consumption is expected to double from about 460 TWh in 2022 to more than 1,000 TWh in 2026, according to the International Energy Agency (IEA). This demand roughly corresponds to what all of Japan consumes in a year.

But things can go even further.. This prediction may not come true if generative AI becomes as popular as some analysts claim. In August, we learned that OpenAI already had 200 million weekly ChatGPT users, up from half that figure in November 2023. If usage increases dramatically, so will the energy demands of the data centers that provide us with all these answers and content.

But criticism of the nuclear program still exists.. Meanwhile, the controversy remains active, especially due to opposition from activists who highlight the environmental and safety risks posed by nuclear power plants. Greenpeace claims that “nuclear power is incredibly expensive, dangerous and slow to build.”

And be careful about competition from renewable energy sources.. Those who criticize nuclear energy do not consider it a viable alternative to renewable energy sources, but the truth is that the competition between both energy sources is obvious. The government support that renewables have received in recent years contrasts with this new push for nuclear power, at least from the private sector (and in public opinion). There could be an imbalance here: if new nuclear power plants displace cheaper renewable sources, electricity prices could rise.

Image | Lukasz Legocki

In Hatak | Artificial intelligence has already reached nuclear power plants. And it will change them forever