The Global Environmental Footprint of AI in 2025.

In 2025, we have seen the surge of artificial intelligence (AI), especially Generative AI (GenAI) such as ChatGPT. While the former is introduced mindfully in some sectors for efficiency gains and data management, the latter has spread uncontrollably in all aspects of our daily lives. In only three years, AI tools shifted the landscape in which businesses operate with a vast majority of organisations reporting the use of AI in at least one business function (Figure 1).

Figure 1 - McKinsey & Company

AI vs GenAI

While AI is widely expected to reduce operational costs, automate repetitive tasks, enhance profitability, and generally contribute to key sectors such as agriculture, health, and energy, GenAI currently plays a different role. The core difference lies in functionality: traditional AI is primarily designed to process large amounts of data and automate tasks, whereas GenAI is more accessible and is commonly used as a creative tool to generate text, images, and videos.

Another key distinction between traditional AI and generative AI relates to their potential societal impacts. While the benefits of traditional AI are often considered alongside its drawbacks, the broader implications of GenAI remain debated. Currently, its widespread global use raises questions about the amount of resources it needs to function, largely seen as disproportionate.

AI use in 2025

Microsoft reports that in 2025, one in six people worldwide used GenAI tools with regions such as the United Arab Emirates, Singapore, Norway, New Zealand, and France leading the way. In the United States alone, about 82% used GenAI weekly and 46% used it daily. As Figure 2 shows, almost every country in the European Union reported an increase in the use of AI in 2025.

Figure 2- Eurostat

In other words, with hundreds of billions of dollars poured into the development of AI, its use and deployment are expected to continuously increase worldwide.

Environmental Footprint of AI

However, as demand for AI continues to rise, so too does the demand for the resources required to support it. As such, there are currently 1189 large hyperscale data centres globally with an additional 504 in planning and construction stages. These massive data centres are energy intensive as they are the place where AI model training and deployment happen. They are home to servers, storage systems, networking equipment, batteries and backup power generators, and other additional equipment. Like any electronic device, these facilities heat up when in use, and at the scale these centres’ function, they require large cooling and environmental control systems.

Therefore, the accelerating growth in investment of these data centres driven by the increase demand for AI is driving up the demand for electricity. The increase is mostly met by fossil fuels and consequently driving up CO₂ emissions dramatically. Therefore, AI is estimated to have emitted more greenhouse gases than the whole of New York City, or about 80 million tonnes in 2025 alone.

Besides electricity, AI needs a considerable amount of fresh water to operate its cooling and environmental control systems. For instance, a mid-sized data centre can consume water as much as a small town would. Larger ones can see water consumption go up as high as 5 million gallons of water per day – in other words, the equivalent of a city of 50.000 people. Besides the direct use of water for operations, AI also needs water for indirect electricity generation and for the manufacturing process of the hardware. Some reports estimate that AI’s water consumption in 2025 could be equal to the global annual consumption of the bottled water industry. Estimates also show a possible increase of 870% in water use in the following years to meet the demand of new facilities being built.

Yet, in the US, two-thirds of new data centres since 2022 either have been built or are under construction, in areas where water stress is already an issue. This poor planning means that certain communities have faced water challenges such as disrupted water supplies, hindering their quality of life.  

Policy Landscape and Regulations

Although it is difficult to forecast precisely the environmental footprint of GenAI as tech companies rarely disclose such data, it is apparent that the resources needed are absurd. And yet, 2025 trends indicate that the use of AI will be increasing rapidly in the coming years, despite a lack of robust regulation and policy frameworks to safeguard water security for communities located near large data centres. Equally, insufficient attention is being paid to ensuring that national net zero targets are not undermined by the growing reliance on fossil fuels to meet rising electricity demand. Some projections suggest that AI-related energy demand could increase by as much as 25 times compared to current levels.

Call-to-action

Considering these facts, it is crucial that everyone use AI, especially GenAI, responsibly and wisely. Integrating GenAI into business functions will drive up the carbon footprint of an organisation significantly and could undermine the achievability of net zero targets (see our guide on managing AI’s footprint in an organisation). Accordingly, membership bodies such as trade associations and professional bodies have the responsibility to raise awareness to their members on the responsible use of AI and its associated environmental costs.

If you are a membership body and you are unsure what this looks like for you, CAFA has guidance, 1:1 support, and resources that you can use for free. join here for access.