AI Model Training Needs Additional Record Levels Of Electricity

Leaving aside the direct use of AI by consumers and businesses and there is another aspect of its growth that will take up an unfathomable amount of energy, almost all of which will be electricity. AI models must be trained. Training is critical to the ability of AI to do virtually all of its analysis and output. 

EPRI estimates that the training aspect of the AI process will take up as much electricity as that of entire cities. “Epoch AI forecasts that training a leading model could require more than 4 gigawatts (GW) of power by 2030—enough to power millions of U.S. homes.”

This energy use will come even though there is a chance that AI will solve some of its energy problems via the creation of energy models, which are beyond the capacity of homes to create. 

More Than Server Farms

The energy training needs are beyond the massive draw required to run server farms. “The energy demands of training cutting-edge AI models are doubling annually, soon rivaling the output of the largest nuclear power plants,” said Jaime Sevilla, director of Epoch AI.

The challenge is that some forecasters show that server farms alone could use 50% of all US electricity by the end of the decade. With training, electricity needs are added; that number could go up.