Is AI sustainable?
When discussing artificial intelligence, attention is often directed toward its technological and programming aspects or the wide range of applications it enables. Much less frequently, however, we consider its broader impact on society and, above all, on the planet, together with the environmental consequences that accompany its rapid development.
​
At the intersection of artificial intelligence, ethics, and sustainability stands Dr. Sasha Luccioni, a leading researcher with a PhD in AI and over a decade of experience in both academic research and industry. Her work has earned international recognition, including being named among TIME Magazine’s 100 most influential people in AI and featured in Business Insider’s 2024 AI Power List. During the 2026 International AI Festival, she analyzed in depth the relationship between AI development and climate change.
The enthusiasm surrounding generative AI is understandable. From improving worker productivity to accelerating scientific discovery, its potential benefits are vast and compelling. Yet this rapid and widespread adoption of powerful AI systems comes at a significant environmental cost. The training and deployment of large-scale generative models dramatically increase electricity demand and water consumption. Models containing billions of parameters, such as OpenAI’s GPT-4, require enormous computational power, which translates into higher carbon dioxide emissions and increased pressure on already overburdened electrical grids.
In addition to energy consumption, AI systems rely heavily on water resources. Large amounts of water are required to cool the hardware used for training, fine-tuning, and operating these models, placing stress on municipal water supplies and potentially disrupting local ecosystems. It is estimated that typing just 100 words into a chatbot can consume the equivalent of three bottles of water. On a larger scale, companies such as Amazon operate more than 100 data centers worldwide, each hosting around 50,000 servers to support cloud computing services—an infrastructure that today generates a carbon footprint exceeding that of the aviation industry. Moreover, the training of GPT-3 alone is estimated to have produced approximately one ton of COâ‚‚ emissions.
​
This raises a crucial question: are AI agents truly sustainable? The answer depends largely on the design choices made during their development. Models built for specific tasks, such as small language models, have fewer parameters and therefore lower energy requirements, especially when trained efficiently and without unnecessary data. Studies comparing AI models by tracking their energy consumption have revealed substantial differences among reasoning systems. In particular the generation system of an AI model is extremely important: image-generation models consume far more energy than text-based ones because they generate content pixel by pixel, as a consequence video-generation models are even more energy-intensive. In fact, producing a 10-second video can require as much energy as charging a smartphone ten times.
So Artificial intelligence is neither inherently sustainable nor inherently harmful to the environment. Its impact depends on how it is designed, trained, and deployed. As AI continues to shape the future of society, the challenge lies in balancing innovation with responsibility. Reflecting on the environmental cost of digital progress invites us to rethink not only what AI can do, but also how—and at what price—we choose to use it.
-Gennaio 2026_