Unveiling the Numbers: The Environmental Footprint of Generative AI

0
24
Generative AI's environmental impact in figures

The Hidden Costs of Generative AI: An Environmental Perspective

Growing Concerns Ahead of the Paris Summit

The rapid rise of generative artificial intelligence (AI) has sparked discussions about its ecological footprint, a key topic addressed at the global summit in Paris on February 10-11, 2025.

Energy Consumption: A Stark Comparison

Every request made to OpenAI’s chatbot consumes approximately 2.9 watt-hours of electricity. This figure is ten times higher than the energy consumed for a single Google search, as reported by the International Energy Agency (IEA).

OpenAI has indicated that ChatGPT now boasts 300 million weekly users, generating a staggering one billion requests daily.

Ubiquity of Generative AI

Since its prominent emergence in 2022, generative AI has exploded in usage, with numerous chatbots available to the public. A survey by French pollsters Ifop found that 70% of 18- to 24-year-olds in France use generative AI. Similarly, a Morning Consult poll in the United States indicated that 65% of 13- to 17-year-olds engage with this technology, with nearly half of the overall population using it as well.

Data Centres: The Powerhouse of AI

Generative AI primarily relies on data centres laden with massive reserves of information and computational capabilities. In 2023, these data centres accounted for nearly 1.4% of global electricity consumption, as indicated by a Deloitte study.

Projections suggest this figure will climb to 3% by 2030 – an equivalent of 1,000 terawatt-hours (TWh), matching the combined annual electricity usage of France and Germany.

Soaring Energy Needs

The IEA anticipates a more than 75% increase in data centre power consumption by 2026, aiming for a total of 800 TWh compared to 2022. Alarmingly, American consultancy Gartner warns that up to 40% of data centres built for AI could face electricity shortages by 2027.

Financial and Environmental Costs of Training AI

Training large language models (LLMs) generates significant greenhouse gas emissions. Researchers from the University of Massachusetts Amherst estimated that training a single model emits around 300 tonnes of CO2, approximately the same amount produced by 125 round-trip flights between New York and Beijing. A subsequent study by Oxford University revealed the figure to be around 224 tonnes for the training of OpenAI’s GPT-3 model.

The arduous task of training numerous models to enhance technology compounds these emissions, making it difficult to assess the overall environmental impact comprehensively.

The Water Footprint of AI

Another critical resource required for generative AI is water, primarily for cooling hardware. Researchers at the University of California Riverside have estimated that generating 10 to 50 responses from GPT-3 necessitates about half a litre of water.

Projected AI-driven water consumption may reach between 4.2 billion and 6.6 billion cubic metres (155 billion – 233 billion cubic feet), equivalent to four to six times Denmark’s annual water usage.

The Electronic Waste Crisis

Generative AI also contributes to electronic waste, producing approximately 2,600 tonnes of discarded materials like graphics cards, servers, and memory chips in 2023, according to the Nature Computational Science journal. Without intervention, this amount could swell to 2.5 million tonnes by 2030, akin to discarding around 13.3 billion smartphones.

The manufacturing of AI hardware relies on rare metals, with their extraction often involving environmentally detrimental mining practices, particularly in Africa.

Conclusion

The rise of generative AI presents significant ecological challenges that demand critical attention. As technology continues to evolve, balancing innovation with environmental sustainability will be key to ensuring a healthier planet for future generations.

Questions and Answers

  1. What was the primary topic discussed at the global summit in Paris in February 2025?
    The primary topic was the ecological footprint of generative artificial intelligence (AI) and its environmental impacts.
  2. How much electricity does OpenAI’s chatbot consume per request compared to a Google search?
    OpenAI’s chatbot consumes 2.9 watt-hours per request, which is ten times more than a Google search.
  3. What is the projected increase in data centre power consumption by 2026?
    The IEA forecasts more than a 75% increase in data centre power consumption by 2026, reaching approximately 800 TWh.
  4. How much CO2 is emitted during the training of large language models?
    Training a large language model can emit around 300 tonnes of CO2, comparable to 125 round-trip flights between New York and Beijing.
  5. What is the anticipated volume of electronic waste from generative AI applications by 2030?
    The anticipated volume of electronic waste could reach 2.5 million tonnes by 2030, equivalent to about 13.3 billion discarded smartphones.

source