AI Uses as Much Energy as a Small Country, Usage Expected to Double by 2026

Artificial Intelligence (AI) is revolutionizing our world, driving advancements in various fields, but this progress comes with significant energy demands. According to the International Energy Agency (IEA), data centers, cryptocurrency, and AI together accounted for nearly 2% of global energy demand in 2022. Projections suggest that this demand could double by 2026, equating to the electricity consumption of Japan.

This digital age, where machines guide many aspects of our lives, comes with environmental trade-offs. AI, particularly generative AI, requires enormous energy both for training and operation. For instance, training a large language model like OpenAI’s GPT-3 consumes approximately 1,300 megawatt-hours (MWh) of electricity, which is equivalent to the annual energy consumption of about 130 U.S. homes. Moreover, a single ChatGPT request uses 2.9 watt-hours of electricity, significantly higher than the 0.3 watt-hours used by a Google search.

In a recent interview with Sasha Luccioni, lead climate researcher at Hugging Face, she highlighted the challenges associated with AI’s energy consumption. Luccioni noted that transitioning from traditional AI to generative AI can increase energy usage by 30 to 40 times for the same task. She emphasized the need for transparency in AI deployment, as users are often unaware of the energy implications of their interactions with AI systems.

Data storage and AI model training are significant contributors to this energy demand. Modern AI models require extensive data and computational power, leading to increased use of energy-intensive GPUs. These GPUs, while more powerful, also consume more energy, creating a vicious cycle of escalating demand.

Additionally, the infrastructure supporting AI, such as data centers, also has substantial environmental impacts. These centers require continuous cooling to prevent overheating, often relying on water circulation, further intensifying their resource use. As AI applications scale up, these centers increasingly rely on nonrenewable energy sources like coal and natural gas, exacerbating their carbon footprint.

To mitigate these impacts, Luccioni advocates for providing consumers with information to make energy-efficient choices. She is working on developing an Energy Star rating for AI models, which could help users select models based on their energy efficiency.

Furthermore, the concept of “digital sobriety,” popularized in France, encourages consumers to consider the necessity of their digital interactions. Luccioni suggests evaluating whether we need new gadgets or AI-powered conveniences, promoting mindful consumption of technology.

As AI continues to advance, balancing its benefits with its environmental costs will be crucial. By prioritizing energy-efficient practices and technologies, we can work towards a greener, more sustainable future for AI.