The trend – The rapid expansion of AI is placing an unprecedented strain on global natural resources, particularly energy, water and land use. Over the past few weeks we have seen a flurry of reports highlighting the compounding environmental costs of AI data centres – and how these costs are vastly outpacing current regulatory frameworks. From exacerbating local drought to shifting public sentiment against hyperscale infrastructure, the physical footprint of the cloud has emerged as a critical friction point. Read on to explore how the industry and policymakers are responding.

‍

Warning signals – AI’s environmental footprint is rapidly expanding, according to a UN report. It highlights that electricity used to power AI and data centres could double to 945 TWh annually by 2030, while AI-related water consumption could reach 9.3tn litres matching the yearly domestic needs of 1.3. billion people by the end of the decade. Meanwhile, the land footprint of data centres’ electricity demand is expected to expand from 6,900 sq km in 2025 to more than 14,500 km by 2030 and annual e-waste from AI infrastructure could rise to 2.5 million mt by the same year.

‍

Community backlash – At the local level, the sheer volume of water required to dissipate server heat is exacerbating existing ecological crises. An investigation by The Guardian detailed how data centre expansion is directly compounding localised droughts and straining municipal water systems across the US, noting that large facilities can use up to five million gallons of water a day. Illustrating this dramatic shift in public sentiment is a Gallup poll that revealed a majority of Americans would now prefer to live next to a nuclear power plant over an AI data centre, citing strains on local resources including water and energy consumption.

‍

Corporate responses – Technology firms are in the race to create more sustainable data centres. For example, Amazon, Google, Meta and Microsoft have joined an Elemental Impact-led initiative to support 10 start-ups developing advanced cooling technologies, low-carbon building materials and energy storage. Projects will receive between $500,000 and $5m each through 2027. Microsoft’s CEO Satya Nadella has also claimed that the company’s next-generation AI “superfactory” in Wisconsin, US, has been designed to use only as much water as a single restaurant does in a year thanks to a cutting-edge cooling loop that only needs filling once.

‍

Setting the standard – Recognising the need for clarity, nine of the world’s biggest built environment and sustainable finance organisations recently launched the Greening AI Data Centres Coalition (GADCC). The goal is to create an internationally aligned framework of environmental and social standards for data centres spread across energy, carbon, water, waste, biodiversity and community impact. It will also support the establishment of credible green finance instruments such as green bonds and sustainability-linked loans for data centre investments that meet its standards.

‍

Extreme relocations – To escape terrestrial resource constraints entirely, the industry is exploring radical geographical shifts. In a groundbreaking demonstration project, the world’s first wind-powered underwater data centre has officially commenced operations in China. By submerging server clusters in marine environments, operators can exploit natural oceanic cooling to eliminate freshwater consumption while running on offshore renewable energy. While experts note potential risks to marine ecosystems such as disturbing sediments or heating seawater, they said these are likely manageable with further monitoring.

‍

To the stars – Taking this logic to the absolute limit, the Carbon Trust recently investigated the viability of launching data centres into space. Its analysis points out that such data centres are being touted as clean, secure and potentially economically competitive. However, there are major barriers to overcome to make this a reality. Construction and maintenance will require sophisticated robots, while radiative cooling will need surfaces in the region of square kilometres. Furthermore, the growing amount of space debris – a trend we recently reported on – brings collision risks. Multiple launches in the hundreds or thousands will create millions of tonnes of CO2 emissions.

‍

Under the microscope – Managing AI’s environmental impact requires moving beyond isolated efficiency metrics. As the sustainable finance sector works to build original rating systems and accurate scoring methodologies, establishing thematic frameworks that measure absolute resource consumption will be vital. Identifying which operators are genuinely mitigating their footprint – rather than simply displacing it – will be the defining challenge for those assessing long-term material risks in the tech sector.