The Environmental Cost of Generative AI: A Growing Concern

The Rising Popularity of Generative AI

Generative AI has seen a significant surge in popularity since ChatGPT's public debut in 2022. A survey by French pollsters Ifop found that 70% of 18- to 24-year-olds in France use generative AI, while in America, a Morning Consult poll revealed that 65% of 13- to 17-year-olds and nearly half of the general population use this technology 1

1

2

2

. OpenAI claims that ChatGPT alone has 300 million weekly users, generating one billion requests daily 1

1

.

Energy Consumption: A Growing Concern

The energy demands of generative AI are substantial and increasing. Each request to OpenAI's chatbot consumes 2.9 watt-hours of electricity, which is ten times more than a Google search 1

1

2

2

. Data centers, crucial for AI operations, accounted for 1.4% of global electricity consumption in 2023 1

1

2

2

.

With massive investments planned in generative AI, this figure is expected to reach 3% by 2030, equivalent to 1,000 terawatt-hours (TWh) - comparable to the combined annual consumption of France and Germany 1

1

2

2

. The International Energy Agency (IEA) forecasts a more than 75% increase in data center power consumption by 2026 compared to 2022 levels, reaching 800 TWh 1

1

.

Carbon Footprint of AI Training

Training large language models (LLMs) that power chatbots has a significant carbon footprint. Researchers at the University of Massachusetts Amherst estimated in 2019 that training one LLM generates around 300 tonnes of greenhouse gas carbon dioxide, equivalent to 125 return flights between New York and Beijing 1

1

2

2

. Oxford University researchers later estimated 224 tonnes of CO2 for a single training session of OpenAI's GPT-3 model 1

1

2

2

.

Water Consumption in AI Operations

Generative AI also consumes substantial amounts of water, primarily for cooling computer hardware. GPT-3 requires approximately half a liter of water to generate between 10 and 50 responses 1

1

2

2

. Researchers project that increased AI demand for water could amount to between 4.2 billion and 6.6 billion cubic meters by 2030, which is four to six times Denmark's annual water consumption 1

1

2

2

.

Electronic Waste: A Growing Problem

The rapid advancement of AI technology is contributing to electronic waste. A study from Nature Computational Science journal reported that generative AI applications produced around 2,600 tonnes of electronic waste in 2023, including graphics cards, servers, and memory chips 1

1

2

2

. If current trends continue, this could escalate to 2.5 million tonnes by 2030, equivalent to about 13.3 billion discarded smartphones 1

1

2

2

.

Challenges in Measuring Environmental Impact

Despite these estimates, researchers face challenges in accurately assessing generative AI's overall environmental impact. Experts point to a lack of information on how models are produced and an absence of global measurement standards 1

1

2

2

. This uncertainty underscores the need for greater transparency and standardized reporting in the AI industry.

Future Concerns and Potential Shortages

As AI applications continue to grow, there are concerns about potential resource shortages. American consultancy Gartner predicts that up to 40% of data centers built for AI applications could face electricity shortages by 2027 due to vast power demands 1

1

2

2

. Additionally, the production of AI hardware requires rare metals, often mined in Africa through potentially polluting processes 1

1

2

2

.

As the global AI summit in Paris approaches, these environmental concerns are set to be a key topic of discussion, highlighting the urgent need for sustainable practices in the rapidly evolving field of generative AI.