Sam Altman, general director of OpenAI, recognized it during his visit to the Davos summit (Switzerland) last January, the Achilles heel of artificial intelligence is its energy consumption. His hope, a breakthrough in nuclear fusion which will provide clean and infinite energy. Promethean dream. In the meantime, it is rather dirty energy that comes to its rescue, coal.
Since the start of the year in the United States, major electricity producers have announced, one after the other, the postponement of their plans to close coal-fired power plants. THE Financial Times actually list it. In Wisconsin, Alliant Energy is postponing the conversion of a coal-fired power plant to gas by three years, and FirstEnergy is pushing back its goal of phasing out coal beyond 2030.
According to a study by Standard & Poor’s, the drop in electricity production from coal will be 40% lower than expected in 2023. Under these conditions, there is little chance that the total output targets of this energy in 2035, decreed by the American environment agency, can be achieved. Twenty-five American states have already initiated proceedings against this decision.
The origin of this electrical panic lies at the heart of the chips. Those from Nvidia, the specialist in the field, heat up ten times more than a usual microprocessor. In other words, ChatGPT consumes ten times more energy than Google’s search engine. However, the major digital players, Microsoft, Amazon and Google, are in the process of deploying tens of billions of data centers adapted to this new technology throughout the world.
Another challenge for the energy transition
THE Wall Street Journal takes the example of Northern Virginia, the American state best equipped in the world in this area. The 250 data factories it owns on its territory already consume 4,000 megawatts, enough to power 1 million homes. Current projects will increase this consumption to more than 11,000 megawatts. As a result, according to the newspaper, data centers are expected to represent 9% of electricity demand in 2030, twice as much as previous forecasts.
This is excellent for electricity companies whose profits are soaring, but which poses an additional challenge for the energy transition. We had anticipated the gradual switch of cars to electricity, but not that artificial intelligence was going to get ahead of it with such vigor. Creating as many new problems as we wanted to solve is also the curse of Prometheus.