In recent months, ChatGPT has become highly popular and has transformed the world of web writing. Despite its effectiveness and comprehensiveness, the use of this artificial intelligence system also contributes to its environmental impact and thus it is important to be aware of its digital footprint.

What is ChatGPT?

ChatGPT (Chat Generative Pre-Trained Transformer) is a program created by OpenAI, an artificial intelligence research organization, used for dialog creation.

It is a particular type of chatbot initially created to improve customer service in terms of speed and accuracy, but now possesses far more potential than just chat management.

ChatGPT is a language model developed for human interaction in a conversational manner. It is a pre-trained text generator that utilizes natural language processing (NLP) and a vast database of texts, websites, and articles to mimic human style and respond to inputs.

The main feature of ChatGPT is to generate as “natural” responses as possible in a text box. It is capable of writing a code, article, translation, debug, or complex document [1].

The environmental impact of machine-learning models

As pointed out by Ron Schmelzer in Forbes, although AI (Artificial Intelligence) models are also seen as a way to combat climate change in terms of immediately finding information to act preemptively, these same digital systems require an enormous amount of energy.

This is precisely why machine learning models, and more generally the whole world of the Internet, are increasingly the focus of attention because of their impressive consumption of electricity.

In terms of data processing alone, industries using AI technology pollute as much as the aviation industry. In fact, training an AI system can emit more than 113 tons of carbon dioxide [2] which, in the case of Chat GPT, reaches as much as 522 tCO2e [3].

How much does Chat GPT pollute?

As useful and cutting-edge as ChatGPT may be, in a brief analysis Chris Pointon estimates that this artificial intelligence system could emit about 3.8 tons of CO2e every single day![4]

The same author reports that “since the Internet is the largest coal-fired machine on the planet, it is likely that anything that gets a lot of traffic will emit a lot of CO₂,” so his doubts are well founded.

For Chat GPT training alone, a level of emissions equal to that of a car traveling as long as a round trip to the Moon has been estimated. In fact, some digital scientists have calculated that this “training” required to operate Chat GPT generates CO2 at the same levels as a car traveling 700,000 km, which is about twice the distance between Earth and the Moon[5].

Clearly, Pointon also reports, these are rough calculations since there is limited information available. The calculations made are to be taken with a grain of salt, since the Chat GPT manufacturer itself does not reveal all that is necessary to go into more detail.

That is why in another article Kasper Groes, with the same premises of inaccuracy, makes a similar calculation but arriving at different results, according to which ChatGPT pollutes much less.

Although therefore the carbon footprint of ChatGPT cannot be estimated to the millimeter, what is certain is that it requires a huge amount of data and electricity for its normal operation. 

How to reduce the impact of large AI models

As mentioned so far, artificial intelligence-based models must process a lot of data, inevitably increasing the need for computing power and electricity, along with the need to cool data centers. Because of their high energy requirements, artificial intelligence systems can have high financial and environmental costs, leaving a significant carbon footprint.

So how can we try to reduce their environmental impact? From the perspective of the users of these software, it is clear that the scope for action is very narrow in terms of reducing environmental impact; providers, on the other hand, have several options for limiting their digital footprint.  

  • “Compressing” large models. During processing, a large AI system can be compressed, thus transferring the same database into a smaller model, which will then need less power consumption[6].
  • Select only the necessary data. Selecting only the most relevant data for training or effectively modifying existing models for a new task dramatically reduces the cost of training a model, making AI more sustainable.
  • Invest in the renewable. With a view to digital sustainability, it is essential to rely on cloud providers that ensure that power is supplied from renewable energy sources, so as to minimize CO2 emissions.

Responsible choices even in the digital world

As noted in other articles (see also What is and Why is Digital Sustainability Important), recent studies and research are increasingly highlighting the need for both consumers and companies to make responsible digital sustainability decisions.

Exactly as one tries to switch from gasoline to electric cars, or as one purchases products with reduced environmental impact, it is critical to make responsible choices to reduce the Digital Carbon Footprint.


[2]Cornell University, 2019

[3]Pointon C., The Carbon Footprint of ChatGPT,, 2022 

[4]Pointon C., The Carbon Footprint of ChatGPT,, 2022 

[5]The Register, 2020

[6]Pointon C., The Carbon Footprint of ChatGPT,, 2022