I think that's a very easy question for businesses. If they do everything on edge and on prim they will save a lot of money on the energy efficiency but also have more control of their data and their AI as well as their information and I peak so it will save the environments but also save on hardware cost and cloud cost and it will save more energy also not being shot up into the cloud back to the US to be computed and then back here to the EU so I think overall it just makes more sense to to bring a lot of things including expertise of AI in-house but also the hardware in-house, so that you don't have back doors you don't have data leaks and you have more control over your business. Currently, EU has SustainML.eu to start researching this as well as a contact of mine Professor Raghavendra (Raghav) Selvan who wrote and compiled a great research book for Sustainable AI . We may be folly to jevons paradox and will fall for more efficient ways to save energy and resources only to continue operation past planetary boundaries.
Artificial intelligence systems consume a huge amount of energy and resources, both during their training phase and in their daily operation. Currently, data centers are the facilities that house the computing infrastructure necessary for AI models to be trained and deployed, and these require intensive cooling systems and a constant electricity supply, among other things. All of this raises a significant dilemma regarding innovation in the field of artificial intelligence and its commitment to environmental sustainability. Below, I present some proposals that could contribute to addressing this challenge: - Renewable energy: Migrating data centers that support AI systems to renewable energy sources Clean energy, such as solar, wind or hydroelectric power, would significantly reduce the CO₂ emissions associated with its operation. - Algorithmic efficiency: Designing more compact and optimized AI models that are capable of delivering equivalent results with lower consumption of computational resources. Model reuse: Instead of training new models from scratch, leverage existing models - pre-trained and adapt them, thus avoiding training from scratch which is especially costly in terms of energy.
This is one of the lesser-known aspects of AI. End users often perceive AI as something intangible that exists in “the cloud,” but that cloud is actually made up of enormous buildings filled with servers (data centers) that consume industrial quantities of electricity and thousands of litters of water for cooling. The problem has two phases: training and use. Training cutting-edge models such as GPT-5.2 or Gemini 3 requires massive computing power that leaves a carbon footprint equivalent to that of a small city operating at full capacity. But daily use (inference) is the “silent killer”: every time millions of people generate an image or request complex text, consumptions skyrockets. To balance this, I believe that the solution is not to slow down innovation, but to focus on real efficiency. Today, we are sometimes “using a sledgehammer to crack a nut”: we use a gigantic, expensive models for trivial tasks such as summarizing a simple email. The future must involve smaller, specialized models (SLMs) that consume much less, and optimizing hardware so that each prompt does not have such a high environmental cost for the planet.
In short, by becoming better at producing energy. Astronomy measures a civilization’s technological advancement based on their consumption of energy and it’s not really farfetched. It was the need to consume energy that sparked humans to come up with better and safer ways to harness it, and we can already see the effects of the AI fever today: There has never been bigger breakthroughs in fusion energy, as this last decade, driven by the ever-growing energy needs. Fusion, if harnessed, could be the ultimate, clean, almost infinite energy source that modern physics can feasible produce, and it would mean no more carbon emissions, no more radioactive waste. Even if we don’t reach fusion, we’ve also been finding better alternatives than water (as superheated steam) to harness the power of nuclear fission, something that was a staple for almost a century already.
The truth is, I’m not personally very familiar with what measures can be taken at this stage to balance the use of AI with the protection of nature and the environment. This largely depends on the decisions that large corporations make regarding how they power their generators and the cost they bear for everything they do and continue to expand. If people in leadership positions think more about the footprint they leave behind and about alternative methods to replace current ones, perhaps a solution to this problem will be found. For example, using renewable energy or reusing pre-trained models would reduce unnecessary energy consumption. Of course, as a tool, I believe that artificial intelligence can be used to protect nature—through observations, research, and helping scientists predict the future.
