For the last decade, environmental experts have been pointing out a specific field of concern, regarding the environmental footprint. That is the Green IT, which means the footprint coming from  any device, from smartphones to the cloud. And not because it represents the biggest percentage of that footprint, but because it is growing very quickly. And since 2022, with the widespread use of AI, even quicker.

That is why a new specialization comes up: the Green AI.

AI is more than just a new technology, it is already part of our daily lives. New proposals of accessible applications are available, more and more. And we are using them in many different ways, like designing, writing, coding… in order to enhance the manufacturing, transportation, science, marketing, cybersecurity… 

Although AI has been an important field of study for almost 80 years, since the end of 2022 a new universe of generative AI is accessible to almost all of us. It’s the LLM or Large Language Models, together with their chat interface, which uses RLHF or Reinforcement Learning from Human Feedback. 

As a consequence, the usage of this technology in USA households, for instance, has grown up to 40% in 2025. And it keeps skyrocketing. The figures are also growing regarding business AI usage worldwide.

And all that AI training, these queries, these prompts that we are elaborating, imply a huge technological deployment. More and bigger datacenters, with more and faster GPUs, connected to users through a wider bandwidth.

A datacenter is where all servers, a specific type of computer which provides information to other computers, are located. 

A GPU, on the other hand, is a Graphics Processing Unit and it works in a similar way as the CPU or Central Processing Unit. The CPU is the kernel of a computer, the brain, so to speak. Its where the processors are. These processors take care of all the activity of the computer and are connected to the RAM memory through the motherboard. 

In the case of GPU it’s slightly different. It means Graphics Processing Unit, and as its name tells us, it’s where all processes related to images, video, etc., are processed. The GPU is connected to the VRAM memory through the graphic card. And the graphic card is attached to the motherboard. The GPUs being more efficient in processes related to AI, these are the ones used in datacenters to work with it.

In any case, bigger datacenters and more and faster GPUs means more energy and water consumption.

Moreover, at first, LLM models were specialized in a specific field. But many of the proposals available today respond to the General Purpose AI model. That means these new LLMs are not specialized anymore, but can solve any kind of problem, related to any kind of subject. For that, the model is going to need an amount of data around 10000 times bigger than before, in order to have all the elements and to be able to combine them. 

That means also a lot more of training and, therefore, again, more energy and water consumption.

Actually, datacenters in Ireland, a country with a big presence of tech companies, are already using more than 10% of the electric energy of the whole country. And it is becoming a major concern. Where is the extra energy the AI will need going to come from? How will that affect users, whether it is companies or households? Will our electricity bills grow because of that? That may eventually happen.

Because more energy in the system means more infrastructure, which means more investment and a bigger complexity of the energy grid.

According to French engineer Marc Jancovici, specialized in energy, and very concerned about the environmental footprint, the safest, fastest and cheapest way to fix the problem is betting on the nuclear energy again. After decades of negative feedback from the public opinion and pullback from many governments, this technology seems to be living some kind of resurrection. 

Many environmentalists, of course, don’t agree with Mr. Jancovici. But the man has been raising awareness about the climate change and the usage of fossil energies for so many years now, so the debate is worthwhile.

There is the problem of nuclear waste, what to do with it, where to put it. Nuclear waste is extremely toxic and it keeps on being so for centuries. There are also the potential accidents. The Zaporizhzhia power plant, for instance, has been in the middle of combats caused by the Russian invasion of Ukraine. Conflicts and natural disasters happen and nuclear plants have a long life, so that must be taken into consideration.

On the other hand, the necessary technologies to build and maintain the new plants may be concentrated in a handful of players, which awakens the ghost of monopolistic behavior. Europe has had a painful enough experience with the end of Mainstream 2, the huge gas pipeline that used to  connect Russia to Germany.

But according to Mr. Jancovici, the alternative energies, such as solar or wind, are not going to be enough for the huge amount of energy necessary for the whole economy, including AI.

The discussion has just begun and it promises interesting conversations.

In any case, the Green AI is here to stay and it will focus on the optimization, which means using as little energy as possible, to achieve as much as possible. That means environmentally-conscious practices, energy-efficient algorithms and transparency, among other matters