It has been approximated that ChatGPT is utilizing over half a million kilowatt-hours of electricity daily. Photograph by Mark Felix / AFP / Getty
In 2016, Alex de Vries came across information stating that a single bitcoin transaction consumes as much energy as an average American household uses in a day. At that time, de Vries, a Dutch national, was employed at a consulting firm. Alongside his job, he maintained a blog named Digiconomist, focusing on the risks associated with cryptocurrency investments. The energy consumption statistic deeply troubled him.
Upon realizing the lack of available data on this topic, de Vries, then twenty-seven years old, took it upon himself to conduct the research. He established the Bitcoin Energy Consumption Index, which revealed that bitcoin mining currently accounts for a hundred and forty-five billion kilowatt-hours of electricity annually. This surpasses the energy usage of the entire Netherlands and results in eighty-one million tons of CO2 emissions, exceeding the annual emissions of countries like Morocco. Additionally, de Vries began monitoring the electronic waste generated by bitcoin mining, equating to an iPhone’s worth for each transaction, and the substantial water usage, around two trillion liters per year, primarily for cooling the mining servers and managing e-waste from outdated servers.
Concerned about the energy-intensive nature of Artificial Intelligence (A.I.), de Vries expanded his focus to include A.I. sustainability. In a recent publication in Joule, a sustainable energy journal, de Vries, now employed at the Netherlands’ central bank, estimated that integrating generative A.I. into every Google search could escalate its electricity consumption to approximately twenty-nine billion kilowatt-hours annually. This surpasses the energy consumption of several countries, such as Kenya, Guatemala, and Croatia.
De Vries emphasized the significant environmental sustainability challenges posed by A.I., echoing sentiments expressed by prominent A.I. advocate Sam Altman, CEO of OpenAI. Altman highlighted the underestimated energy requirements of A.I. technology and the necessity for breakthrough solutions like fusion or drastically cheaper solar energy with storage on a massive scale.
The recent announcement by the International Energy Agency revealing a rise in global energy-related CO2 emissions in 2023, exceeding thirty-seven billion metric tons, underscores the global challenge of achieving net-zero emissions. Despite efforts to reduce emissions, the surge in emissions, particularly from century-old technologies like the internal-combustion engine, poses a significant obstacle. As A.I. adoption increases and bitcoin prices soar, the dilemma remains: How can the world achieve net zero while continuously introducing energy-intensive technologies?
The energy consumption in mining cryptocurrencies like bitcoin is notably high due to the competitive nature of solving cryptographic puzzles for acquiring these currencies. This process demands substantial computing power, leading to the establishment of server farms in regions with inexpensive electricity. Following China’s ban on crypto mining in 2021, the U.S. has emerged as a primary player in this field. However, challenges persist, such as the U.S. Department of Energy’s efforts to regulate mining energy use facing legal obstacles.
Similarly, the energy requirements of Artificial Intelligence, such as the ChatGPT model, are substantial due to the extensive processing involved in handling vast amounts of information. ChatGPT’s estimated consumption of over half a million kilowatt-hours of electricity daily reflects the scale of its operations, surpassing the average daily consumption of a U.S. household.
While A.I. could potentially mitigate some of the energy challenges it presents by enhancing renewable-energy efficiency, the pace of such advancements may not match the escalating energy demands of A.I. technology. This underscores the urgency for technological breakthroughs, as emphasized by experts like Altman.
De Vries advocates for policy interventions, particularly disclosure requirements, to address the energy-intensive nature of A.I. and cryptocurrency mining. He stresses the importance of learning from past experiences with cryptocurrency mining and implementing proactive measures to tackle the energy implications of emerging technologies like A.I.
