Artificial intelligence (AI) is not an eco-friendly industry - it currently requires a lot of water for its cooling systems and is already estimated to be using up at least 1% of the Earth's total electricity generated. Given Malta's climate commitments, what is the country doing to mitigate this growing technology's environmental harm?
Following announcements of a €100 million investment in digitalisation and AI, the upcoming launch of free, nationally certified AI courses, and a €4 million initiative to roll out Microsoft Copilot across the entire public service - all within the past year - Malta appears to be going all-in on AI. Despite AI being the talk of the town, its contributions to carbon emissions and water scarcity are seldom mentioned.
When contacted by this newsroom, leading Maltese experts - AI Professor Alexiei Dingli from the University of Malta and the Malta Digital Innovation Authority's (MDIA) CEO Kenneth Brincat - said that these environmental externalities need to be "actively managed" till AI technologies evolve to become smaller, more efficient, and resultantly more eco-friendly.
Brincat, as the CEO of the authority in charge of Malta's National AI Strategy, told this newsroom that according to recent OECD analysis, "advanced AI systems - particularly those relying on large-scale compute - have environmental impacts linked to energy consumption, water usage for cooling, and associated carbon emissions".
More AI, more burning of fossil fuels
So, in what ways is AI, so far, harming the natural environment?
Reports and academic papers note that not only is the AI industry majorly fuelled by fossil fuels like coal and natural gas, but servers within the data centres that help large language models (LLMs) like ChatGPT, Claude, Perplexity, and so on, answer people's everyday queries use enormous amounts of water to keep their systems from overheating.
AI data centres are facilities featuring the necessary hardware to run AI workloads - this includes generating AI media, running AI assistants and chatbots, and training new AI models. These data centres can single-handedly cover hundreds of acres in space, with some alone being significantly larger than the city of Valletta.
With the USA and China presently being the two largest countries in the data centre market, what powers the AI industry is heavily dependent on their commitments to renewable energy sources, or lack thereof.
Most AI data centres in the USA are powered by natural gas, with this fossil fuel being responsible for 40% of their generated electricity. Renewable sources cover just under a quarter (24%) of their power, nuclear energy covers 20%, and coal the remaining 15% of their energy.
According to Pew Research, US data centres alone consumed 183 terawatt-hours (TWh) of electricity in 2024, equivalent to over 4% of the USA's total electricity consumption that year. This electricity consumption is forecast to increase by 133% by 2030 and reach levels near the industry's global consumption in 2024, to around 426TWh.
Meanwhile, a whopping 70% of the energy generated to run AI data centres in China derives from coal. Renewables then cover a fifth (20%) of their electricity, and nuclear power is responsible for almost a 10th (10%) of their total share.
The IEA also declared last year that electricity generated by fossil fuels is only expected to begin slowing down after 2030, once nuclear energy and renewables grow more prominent in these regions.
According to the International Energy Agency, 460 terawatt-hours (TWh), that is, 460 billion kWh or 460,000GWh, of electricity was generated to supply AI data centres in 2024. By comparison, in 2024, Malta supplied a total of 3,106.1GWh of electricity that same year - 148 times less. Electricity generated from EU renewables in 2024 totalled at just three times as much as the global AI industry's electricity demand. The IEA projects that at base case, global electricity generation to supply data centres will grow to over 1,000TWh in 2030 and virtually triple to 1,300TWh in 2035.
Hundreds of billions of litres of water used to continually cool AI systems every year
On top of being mostly powered by brown energy, the hardware within AI data centres is prone to overheating and hence requires constant water cooling to keep functioning. Recently published academic research indicates that AI systems may already have an annual carbon footprint equal to New York City's, as well as a water footprint that already rivals the global annual consumption of bottled water.
It is estimated that AI systems used up between 312.5-764.5 billion litres of water in 2025, leaving a carbon footprint of between 32.6-79.7 million tonnes of carbon dioxide.
The high strain placed on AI systems means that data centres require constant cooling to keep themselves from overheating. This infrastructure is so powerful that air cooling is not sufficient; they need water to constantly keep temperatures down.
In addition, the more advanced a given AI prompt is, the more water is consumed - this extends to content generation.
Not all water is adequate for these cooling systems. Dingli told this newsroom that it's important for water within these cooling systems to not have any corrosive elements, as this may damage hardware over time. Because of this, potable water is often used.
This is contributing to accelerated water depletion and global water scarcity issues. With millions of gallons being used every day in these data centres, nearby towns have been experiencing water shortages; some affected residents have had their water supplies disrupted since these centres were established.
On 20 January this year, the United Nations declared that the world has entered an era of "global water bankruptcy", thus going a step beyond the long-warned "global water crisis". In updating this status, the UN University Institute for Water, Environment and Health noted that many regions around the world are experiencing persistent shortages, "whereby water systems can no longer realistically return to their historical baselines".
This status update was made after it was made apparent that many regions around the world are experiencing persistent shortages, "whereby water systems can no longer realistically return to their historical baselines".
According to Eurostat (August 2025), Malta is the most vulnerable from the four EU countries experiencing "water stress". Eurostat defines that a country experiences water stress when its annual water resources are below 1,700 cubic metres per inhabitant. Malta has around just 100 cubic metres of water per inhabitant - the lowest mark from all EU countries. Other EU member states characterised by water stress are Poland, Czechia, and Cyprus. Across all EU countries, each inhabitant has between 8,000-9,000 cubic metres of water resources on average.
While there are no such data centres in Malta, thus leaving Malta safe from their direct environmental effects, Dingli noted that Maltese are contributing to these environmental issues the more they use AI services.
It's like the start of the industrial revolution, things will improve - Dingli
Dingli told this newsroom that the present environmental effects of the AI revolution are comparable to the beginning of the industrial revolution, in that the first inventions of the late 17th century were much worse pollutants than their future iterations.
Dingli stated that the future of AI lies in smaller models, "not in the models we have today". He said that these technologies will become "smaller and much more efficient" through an ongoing push for AI models to run on mobile devices and smaller computers.
Till humanity reaches this stage, he said that certain solutions can be achieved if data centres are constructed with some additional planning. For instance, he referenced how some companies are throwing some of the water they use back into rivers, while some data centres in northern Europe are using their massive heat output to keep nearby villages warm.
CEO Brincat told this newsroom that despite these environmental consequences, with AI now being a foundational layer of digital infrastructure around the globe, "the focus should not be 'AI versus sustainability' but how AI is designed and deployed responsibly, with its environmental footprint measured and mitigated".
In this regard, the intersection between technology and ESG is a key research and policy priority to the MDIA, he added. Brincat commented that Malta's participation in the OECD's Global Partnership on AI (GPAI) is enabling us to actively contribute in these kinds of discussions on responsible and sustainable AI.
"If we want, as a country, to enhance our technology infrastructure, we need to also mitigate the climate challenges that there are, because unfortunately, technology is not that climate conscious and we'll have to think about having such infrastructure which mitigates such risks and such challenges - because the challenges are there," Brincat added.
So, does all this mean that AI and environmental sustainability don't go together? Not quite. While these concerns certainly exist, AI technologies can help other industries become more sustainable, for example, farming systems can optimise their water output for irrigation, improve waste separation, support conservation efforts, and more.
With AI continuing to prosper, evolve, and solidify its place in everyday life, world leaders unilaterally agree that its benefits to boost productivity outweigh its direct cons towards Mother Earth.
So, until the AI industry evolves to become more eco-friendly, perhaps it would be wise to refrain from chatting with LLMs, like ChatGPT, in times of boredom.