Powering AI: The growing energy challenge

Powering AI: AI holds great potential in various fields, from coding to driving, but widespread adoption could lead to an energy demand surpassing some nations. Alex de Vries, founder of Digiconomist, warns about the energy footprint of AI in a commentary for Joule

Since 2022, generative AI, including OpenAI’s ChatGPT, has experienced explosive growth. However, training these models is a resource-intensive process, with companies like Hugging Face reporting that their text-generating AI consumes as much energy as 40 American homes in a year during training, about 433 megawatt-hours (MWh).

Powering AI: The energy-intensive training phase

AI’s energy consumption doesn’t stop at training. De Vries’s analysis reveals that when an AI generates data based on prompts, it expends significant computing power and energy. For example, ChatGPT could consume 564 MWh of electricity daily.

Google’s ambitious AI integration

Improving AI hardware and software efficiency may drive greater energy consumption. As AI becomes more efficient, it becomes more accessible and used in more applications, following Jevons’ Paradox, leading to a net increase in resource use.

Google integrates generative AI into its services, such as email and search. De Vries estimates that if every Google search employed AI, it would require 29.2 TWh of power annually, equal to Ireland’s annual electricity consumption. While this scenario seems unlikely in the short term, rapid growth in AI server production may change the landscape.

The growing energy challenge for AI

By 2027, global AI-related electricity consumption could surge by 85 to 134 TWh annually, comparable to the power needs of countries like the Netherlands, Argentina, and Sweden. Improvements in AI efficiency and repurposing computer processing chips for AI use could further escalate electricity demand. De Vries emphasises the need for cautious AI application due to its energy intensity.

We must manage AI’s energy consumption

In conclusion, the potential energy demands of AI are a critical concern. While AI offers incredible possibilities, from revolutionising industries to enhancing daily life, we must carefully manage its energy consumption. Improving efficiency should be balanced with the realisation that increased accessibility can lead to more significant usage, creating an energy paradox.

Companies, researchers, and policymakers must prioritise energy-efficient AI development as we progress. By doing so, we can harness the power of AI while mitigating its environmental impact, ensuring that this transformative technology serves us without compromising our planet’s resources.