Carbon footprint of modern AI models growing catastrophically fast. According to the AI Index 2025 report, emissions from training neural networks are already comparable to industrial ones, such as oil refineries.
The oil and gas industry has long been under the scrutiny of environmentalists. Companies are introducing carbon capture technologies, switching to renewable energy and optimizing production to reduce harmful emissions. However, pressure on the industry continues unabated: giants face millions in fines, carbon taxes and strict public scrutiny.
But now the area of environmental debate is also at the center of artificial intelligence. New data shows that the carbon footprint of large AI models is rapidly increasing.
The AI Index 2025 report contains shocking numbers:
- Llama 3.1 405B became the leader in terms of emissions – 8,930 tons of CO₂. This is equivalent to the annual impact of 1,940 cars.
- Training GPT-3 resulted in emissions of 588 tons of carbon dioxide – the same amount produced by 127 cars per year.
- GPT-4 turned out to be even dirtier: 5,184 tons of CO₂, which is comparable to the annual emissions of an entire oil refinery.
AI technologies, which are supposed to help humanity, are now themselves becoming a serious environmental problem. Without implementing green solutions, the industry’s carbon footprint will continue to grow, approaching that of traditional industries.
Source: @rosatomogt
