Elizabeth Kolbert writes today for The New Yorker on the alarming energy demands of AI, based on analyses carried out by Alex de Vries, founder of Digiconomist. In 2016—Kolbert reports—, Alex de Vries stumbled upon the alarming fact that a single Bitcoin transaction required as much energy as an average American household consumes in a day. Working for a consulting firm at the time and also running a blog called Digiconomist focused on cryptocurrency risks, the revelation about energy consumption deeply troubled him. Despite his efforts, he couldn't find any substantial data on the matter, prompting him to initiate his research. As a result, he developed the Bitcoin Energy Consumption Index, showcasing on Digiconomist that Bitcoin mining's energy consumption had escalated to a staggering 145 billion kilowatt-hours annually. The figure surpassed the total energy usage of the Netherlands, while the electricity production emitted over 81 million tons of CO2, which exceeds the annual emissions of a nation like Morocco. Moreover, de Vries expanded his analysis to include the electronic waste and water consumption associated with Bitcoin mining, highlighting significant environmental concerns.

Concerns about energy consumption extended beyond cryptocurrencies for de Vries, especially with the obvious rise of Artificial Intelligence (AI). He noticed AI's potential for significant energy use and its growth trajectory, prompting him to address AI sustainability on his blog. In a study published in the sustainable energy journal Joule, de Vries—now employed by the central bank of the Netherlands—projected a significant increase in electricity consumption by Google if it integrates generative AI into every search, surpassing the combined energy consumption of several countries. He argued that such technological advancements pose a challenge to environmental sustainability, a sentiment echoed by OpenAI's CEO, Sam Altman, who also expressed concerns about AI's energy demands during a public appearance at Davos.

The International Energy Agency's announcement of a rise in global CO2 emissions in 2023 to over 37 billion metric tons underscored the urgent challenge of achieving net-zero emissions. Despite data centers contributing a relatively smaller portion to the overall emissions, the increasing reliance on AI and soaring Bitcoin values raise questions about the feasibility of reaching net-zero targets. The inherent energy demands of mining cryptocurrencies and running advanced AI models, such as those used by ChatGPT, require significant energy, leading to calls for innovative solutions to meet these needs sustainably.

De Vries criticizes the slow response in addressing the energy consumption issues associated with AI, compared to the long journey towards recognizing the environmental impacts of cryptocurrency mining. He advocated for immediate disclosure requirements as a realistic policy approach in the short to medium term, expressing disappointment over the delayed acknowledgement of AI's energy implications despite previous experiences with cryptocurrency mining.