AI's Electricity Consumption Rapidly Increasing, Soon to Match a Nation's Power Usage
-
Artificial Intelligence (AI) has become a crucial part of modern technology, but a recent analysis warns that AI's electricity consumption is skyrocketing and will soon match the power usage of an entire country.
A study published by data scientist Alex de Vries from Vrije Universiteit Amsterdam in the journal Joule estimates that by 2027, AI data centers could require 850 to 1,340 terawatt-hours of electricity annually. This is equivalent to the yearly power consumption of countries like Argentina, the Netherlands, or Sweden, accounting for 0.5% of global electricity demand. This trend is concerning, as similar energy consumption in the cryptocurrency sector has already drawn widespread criticism.
Image credit: AI-generated image, licensed by MidjourneyParticularly worrying is the secrecy surrounding energy usage by AI companies like OpenAI, making it difficult to determine their exact power consumption. Alex de Vries estimated the electricity demand by analyzing sales of Nvidia A100 servers, which account for 95% of AI industry infrastructure. He described these Nvidia servers as "energy monsters" due to their massive power requirements.
This issue has drawn significant attention from experts, who are calling for a reevaluation of investments in the AI sector, especially the high-energy operations of tech giants like OpenAI and Google. Roberto Verdecchia, an assistant professor at the University of Florence, stated, "Perhaps we need to slow down and start applying the solutions we already have. We should not only focus on improving model accuracy and speed but also deeply consider the environmental costs of our resource consumption."
California may be the first region to take action. Governor Gavin Newsom recently signed two major climate disclosure bills, requiring companies to report their carbon emissions by 2026. Despite increased regulatory scrutiny, the AI sector remains largely self-regulated, and companies may continue their high-energy operations.
However, given the current high energy consumption, there are economic incentives to reduce these costs, and technological progress is essential. Considering its significant environmental footprint, any breakthrough cannot come too soon. Sustainable development in the AI field requires more innovation and attention to address the growing electricity demands.