The development and use of AI models such as ChatGPT raise significant environmental challenges, particularly due to their high energy consumption. A single ChatGPT training session can require up to 50 gigawatt-hours of electricity, equivalent to the consumption of an entire city like San Francisco for three days. This essential training process allows artificial intelligence to understand and predict effectively, but results in a substantial energy footprint.

ChatGPT’s Energy Consumption: A Significant Ecological Impact

Artificial intelligence offers enormous benefits, but it comes with a significant environmental cost. Training models like ChatGPT requires a massive amount of energy. In a single training session, ChatGPT’s energy consumption is equivalent to that of a city like San Francisco for three days. This article examines how this consumption is distributed, its ecological impact, and its implementation in technological infrastructures. The Impact of ChatGPT’s Energy Consumption

Artificial intelligence models like ChatGPT require significant energy consumption to operate properly. A single ChatGPT training session can consume approximately 50 gigawatt-hours. This is enough to power a city the size of San Francisco for three days, demonstrating the magnitude of the energy required to train these systems.

How Energy Is Consumed ChatGPT’s energy consumption can be divided into two broad categories: the energy required for daily queries and the energy used for the initial training of its models. For queries, ChatGPT processes millions of requests each day, requiring computing resources stored in data centers. Energy-intensive. During training, the model is immersed in vast data sets to analyze and predict complex patterns.

ChatGPT Training: An Ecological Constraint

This training process contributes significantly to the carbon footprint of modern information technology. Using this much energy for a single training session is equivalent to charging a cell phone (15 Wh) approximately 3.3 billion times. For example, the initial training of GPT-4 expended enough energy to equal the annual usage of approximately 4,000 French households. Server Consumption In addition to energy consumption during training, high-performance servers play a central role in the daily operation of ChatGPT. To operate efficiently, a ChatGPT deployment on platforms like Google Search would require over 500,000 servers. This massive infrastructure requires annual energy consumption comparable to that of the entire country of Ireland. Challenges in Estimating Actual Consumption

Large tech companies such as Google, Microsoft, and Meta are often reluctant to share detailed information on their energy consumption, making it difficult to estimate the overall environmental cost. However, researchers such as Alex de Vries-Gao of the Free University of Amsterdam are attempting to provide estimates based on indirect calculations to raise awareness of these environmental issues.