Generative AI is artificial intelligence that produces new digital content. Whether designed to generate text, photos, videos, or code, generative AI is increasingly able to create high-quality material faster than humans.

The release of OpenAI’s Chat GPT, one of the earliest generative AI chatbots, was hailed as a digital milestone. Internet users suddenly had access to a tool that could write, translate, summarise and plan for them. Its potential appeared endless. In the week following its release in November 2022, ChatGPT passed one million user visits. Just three years later, weekly site visits have grown to 810 million.

Though it leads the pack, ChatGPT is not the only generative AI model whose usership has skyrocketed. Gemini, DeepSeek and Copilot are just a handful of other models that have experienced rapid growth since their respective releases. 

As these AI chatbots become more and popular, the scrutiny of them also grows. Digital safety advocates increasingly highlight the risks posed by the under-regulated new technologies. These warnings often focus on direct human impacts, such as human labour replacement or the spread of misinformation.

But what about some less apparent impacts that AI has on humans and the planet we inhabit? 

Water Use - A Hidden Environmental Cost

Online discussion around the environmental footprint of generative AI has intensified in recent months, with water usage frequently cited as a major concern. And for good reason. AI systems rely on vast amounts of water to cool the powerful computers housed in data centres.

Pre-AI, air-based cooling systems were used in data centres to stop computers from overheating. The larger computing power of AI, however, is much more energy intensive and therefore requires a more efficient cooling system.

Water-based cooling systems have been introduced as a solution, where water removes heat from coolant circulating through large computers. Though some water is reused after being piped through cooling towers, up to 80% evaporates. The demand for fresh water is therefore constant.

Global estimated water consumption by the AI sector is 560 billion litres per year, projected to increase to 1.2 trillion litres by 2030. This could spell trouble for countries aiming to rapidly expand their respective AI infrastructures. In the US, which has around 75% of the global AI computing capacity, water scarcity is an increasing issue around large AI data centres. A Bloomberg research team found that two-thirds of new data centres constructed since 2022 are located in areas that already have high water stress levels.

Energy Demand and Carbon Footprint

Water use is only one part of the environmental equation. Generative AI is also very energy intensive. Training large language models requires thousands of computers to run continuously. Even after training, responding to inputted user prompts, a process known as inference, uses significant amounts of electricity.

In 2024, global electricity use in AI data centres was 460 TWh - this is roughly the same as France’s entire annual usage. The International Environment Agency (IEA) projects that this will more than double by 2030. As over half of global AI is powered by fossil fuel-produced electricity, the carbon footprint of the industry is enormous, and will likely become greater over time.

The future energy mix of global AI depends on key national energy policies. Currently, the US and China are by far the largest hosts of AI data centres. The IEA predicts that most of the increased electricity demand from an expanding AI industry will be fulfilled by coal, in China, and natural gas, in the US.

The Material Costs

Obtaining the critical resources used in AI hardware also carries a significant environmental cost. The processors and specialised semiconductor chips that power generative AI are built from dozens of rare earth metals and minerals. Extracting these materials is highly energy intensive and the process often results in widespread polluting events. For every tonne of critical minerals extracted, roughly 2,000 tonnes of toxic waste is produced.

Mining operations often occur in regions with weak environmental safeguards. In China, the largest producer of rare earth minerals, polluted water bodies and contaminated soils are commonplace adjacent to mining hubs, and have been linked to health issues of local populations and notable declines in biodiversity.

The pace of innovation in AI fuels material consumption. Hardware quickly becomes obsolete, with computer chips being replaced every three to five years as newer, faster versions are developed. This massively contributes to the broader global e-waste crisis.

A 2024 study predicts generative AI alone could generate up to 5 million tonnes of e-waste by 2030. If a circular economy of AI chips and their component metals do not emerge, mass disposals of hardware will likely cause widespread pollution issues as hazardous metals and chemicals leach into soils and waterways.

Mitigating Future Impacts

Generative AI now appears to be an unavoidable feature of the digital future. Its rapid development has already surpassed the expectations of many experts. As investment into the sector accelerates, the scale and reach of generative AI are only likely to grow. The accompanying environmental impacts will also expand - these issues can no longer be treated as secondary concerns of the industry.

Excessive water consumption, enormous energy demand, and the mass extraction and disposal of critical minerals risk are increasingly pushing AI development into a dangerously unsustainable trajectory. This does not mean generative AI must or should be abandoned, but it does mean it does suggest tighter legislation is required. As generative AI becomes more deeply woven into our modern way of living, the question is no longer whether it will shape the future, but whether that future will be managed responsibly.