UNESCO Report: Large AI Models Put Huge Pressure on Global Energy and Water Resources

Generative artificial intelligence (AI), in its current form, is becoming increasingly environmentally costly, UNESCO warns in a recent report. The high energy and water consumption of large-scale models raises serious concerns about the sustainability of this type of technology - especially in the context of the accelerated growth of global use.

• Large models, huge impact

In the report titled "Smarter, smaller, more powerful: Resource-efficient generative AI and the future of digital transformation”, UNESCO warns that tools like ChatGPT could consume over 300 GWh per year, equivalent to the electricity consumption of more than three million people in countries like Ethiopia. At the same time, the data centers that support these systems use large amounts of drinking water for cooling - a consumption that could triple by 2027, according to estimates included in the study.

• The solution? Smaller, more efficient models

Leona Verdadero, an AI expert at UNESCO and co-author of the report, emphasizes that large-scale models are not only inefficient, but also unaffordable for many regions of the world due to high infrastructure and energy requirements. UNESCO proposes a paradigm shift: moving from "gigantic” models to smaller, more sustainable models that efficiently respond to everyday needs. To demonstrate the viability of this idea, the team analyzed the performance of the open-source LLaMA-8B model (with 8 billion parameters), comparing it with smaller models with 30 million parameters. The result: for tasks such as machine translation, text summarization or questions and answers, the small models had comparable performance, with up to 90% less energy consumption.

• A problem of global transparency and equity

Another critical aspect highlighted by the report is the lack of transparency regarding the ecological footprint of AI models - especially those developed by large companies. Verdadero compared this situation to "driving a car without knowing how much fuel it consumes”.

The report also draws attention to the risk of deepening digital inequalities: large models are inaccessible to countries and regions with limited resources, which can lead to the exclusion of vulnerable communities from the benefits of digitalization.

• UNESCO recommendations

The report proposes a number of directions for future policies: Promoting and financing energy-efficient and open AI models; Mandatory transparency in reporting on the resource consumption of AI models; Developing an accessible and decentralized infrastructure for AI; Mandatory environmental impact assessments for large-scale models; Digital inclusion: supporting regions with underdeveloped technological infrastructure.

As generative artificial intelligence becomes a pillar of technological development, the choice between "bigger” and "more sustainable” becomes increasingly clear. UNESCO draws attention to the fact that the digital future must also be green, and smaller and smarter AI models can be the key to a fair and responsible digital transformation.