This development favors all uranium stocks

More and more often, traditional search queries on the internet are not being submitted to Google or another search engine, but to ChatGPT or another AI application. This is practical, but has serious consequences, as researchers have determined that the answer to the question by the AI requires around ten times as much energy as the classic answer from a search engine.

In a study conducted by Best Brokers last year, it became impressively clear how much energy demand increases when AI is used. Best Brokers calculated that ChatGPT had around 200 million users per week in August 2024.

Assuming that each user asks the software 15 questions during this period and that each answer requires 2.9 watt hours (Wh) of energy, the daily volume of 428 million queries results in an energy requirement of around 1.2 million kilowatt hours (kWh).

In the USA, nuclear power is the solution to the problem

By comparison, a simple Google search requires around 0.3 watt hours. If you extrapolate these values to a whole year, the AI queries result in a total energy requirement of 450 million kilowatt hours. More than six million electric vehicles could be fully charged with this amount of electricity

This means that running AI applications is not only practical, but also very energy-intensive. What’s more, this energy must be available at all times, as requests are also made at night or when there is no wind, meaning that the operators of the data centers in which the AI is used must have base load-capable power.

Anyone familiar with these figures will understand why Microsoft decided last year to resume purchasing electricity from the currently shut-down Unit 1 of the Three Mile Island nuclear power plant near Harrisburg for a period of 20 years from 2028. The electricity is to be base-load capable. But it should also be CO2-free. For Microsoft and all other operators of large data centers, there is no way around nuclear energy.

More modern AI versions require even more power

However, the massive power consumption during operation is not the only problem for AI developers and data center operators. Training the new AI units also requires a lot of power. This became clear during the changeover from ChatGPT-3 to ChatGPT-4.

As the newer version is much more complex, large amounts of electricity were needed before the launch to prepare for the changeover and train the software. This required 62 million kilowatt hours. This corresponds to 48 times the amount of electricity required for the introduction of the previous model.

It is unlikely that this trend will change with the introduction of new models. Rather, it is to be feared that every innovation will be accompanied by a drastic increase in the amount of electricity required to train and operate the AI, and ChatGPT is currently just one of the new AI applications that are currently active on the market at the same time and vying for users.

The hunt for base load-capable nuclear power has long since begun

It is therefore not surprising that Amazon is already purchasing electricity from a nuclear power plant in Pennsylvania and Oracle is working on a data center that will be operated by three smaller SMR (Small Modular Reactors) nuclear reactors. The US investment bank Goldman Sachs estimates that the electricity requirements of data centers will rise to eight percent of total US electricity demand by 2030. The current figure is just three percent.

Since the US is not dreaming the German Greens’ dream of being able to run an industrialized country on renewable energy, it is obvious that building more nuclear power plants will be the preferred option for solving the huge electricity problem.

As obvious as it is to rely on nuclear energy with base load capability, it is also clear that uranium, which is indispensable for the operation of nuclear power plants, must be available. As this is not guaranteed, not only are further increases in uranium prices to be expected. It is also imperative that significantly more money is invested in the exploration of new uranium deposits. Without this exploration, there will be no projects and mines to produce the uranium needed in the newly built nuclear power plants in the 2030s and 2040s.

Investors who want to bet on the uranium trend should therefore not only look at the shares of the major uranium producers Cameco and Kazatomprom, but also keep an eye on up-and-coming junior companies such as Uranium Energy or Energy Fuels and smaller exploration companies such as Aero Energy.