Currently, there is a rapid increase in the popularity of images created with artificial intelligence (AI) in the style of the 1980s on Instagram. Users upload their regular photos into AI tools to give them a look characteristic of the 80s decade, including clothing, hairstyles, background, and imitation of old cameras. A new image variant is generated in mere moments.
Although this process takes only a few seconds for users, the AI system performs significant computational work. When a large number of people simultaneously request the creation of such images, the AI experiences increased load. This requires intensive work from graphics processing units (GPUs) and servers located in data centers, leading to an increase in electricity consumption and server heating.
Creating photos using AI is not just applying a filter to an existing picture. The AI analyzes the uploaded photo and the given text prompt to form a completely new image. This process involves numerous complex calculations on AI servers. According to one study, generating one AI image requires an average of about 2.9 Wh of electricity, although the demand varies depending on the specific AI model.
If a person stopped after creating one photo, the energy consumption would be lower. However, within trends, people often generate many pictures: they may not like the first version, so they continue generating, change prompts, or try other styles, forcing the AI to create new images repeatedly.
The GPUs and other computer chips used to create AI photos operate continuously, releasing heat. Therefore, it is critically important to keep the servers in data centers in a cooled state, for which various cooling systems are used. The choice of a specific system depends on the location and the technology of the data center itself.
According to the International Energy Agency, global data centers consumed about 460 TWh of electricity in 2024. It is projected that this figure could reach 945 TWh by 2030. The growing use of AI is one of the main reasons for this increase. It is important to understand that this statistic covers all types of operations in data centers worldwide, not just AI photos, but it demonstrates the general trend of increasing demand for computing power and energy with the rise of AI usage.
In some data centers, water is used to reduce the heat emitted by servers. In such systems, water helps dissipate heat outside. In some cases, some water evaporates. Nevertheless, the use of water is not uniform across all data centers and can vary depending on the region, climate, and the cooling system employed.
According to a Lawrence Berkeley National Laboratory report for 2024, the average water consumption in American data centers in 2023 was approximately 0.36 liters of water per kWh of IT energy. This figure represents an average for the entire data center industry.
The more people create images using AI, the more requests the AI receives. A higher number of requests means more intensive server operation. Increased computing power requires more electricity, which leads to higher server temperatures and, consequently, the need for enhanced cooling. In data centers that use water cooling, this can lead to an increase in water consumption. Thus, behind one 80s-style AI photo lies a complex system involving servers, electricity, and cooling systems. Although the exact amount of water used because of this single trend is not yet known, the growth in the number of AI images inevitably leads to an increase in computational and cooling load on data centers.