One of the side effects of the dramatic rise of AI is its thirst for both power and water. Data centres that house chips designed for AI workloads (such as GPUs from Nvidia) consume considerable power. In the USA, 5% of all electricity is consumed by data centres, compared to 2% in Europe and 1% in China. In Virginia, data centres account for a quarter of all electricity demand. The AI boom has caused huge levels of investment in data centres, with PwC estimating over $5 trillion of investment in new data centres by 2030. Goldman Sachs estimates $8.5 trillion by 2030. All those cat videos eat up electricity.
Data centres also consume water. Many data centres use evaporative cooling systems, where water evaporates to remove heat from servers and associated equipment. There is also indirect water use, since electricity generation, particularly from certain thermal power plants, can consume water. Finally, there is the role of water in chip manufacturing. Semiconductor fabrication requires large quantities of extremely pure water, although this is a separate lifecycle impact from operating AI models.
How much water is used? US data centres (not just AI-related ones) consumed 66 billion litres of water in 2023, while a 2025 Greenpeace report estimated that global data centres consume 175 billion litres of water annually. In addition, US data centres indirectly consume an estimated 800 billion litres of water through the power stations that supply them, around 12 times their direct consumption. For context, the US figure of 66 billion litres (17 billion gallons) is around 0.12% of total US public water supply (which is 14 trillion gallons annually), but if you include the indirect use of water then this rises to 1.63% (including 1.51% from indirect water consumption). Golf courses in the US use about 500 billion gallons of water, around 3.57% of the volume of the public water supply. Note that power stations and most golf courses draw from rivers, wells, or recycled water rather than tap water. Agriculture consumes 70% of global freshwater withdrawals.
It is important to understand that a data centre in a water-rich region creates a different environmental challenge from one located in a desert location like Arizona. A 2026 International Energy Agency (IEA) report found that 40% of AI-dedicated data centre capacity is in areas with high baseline water stress. In Northern Virginia, a major data centre hub, local water consumption by data centres jumped from 1.13 billion gallons in 2019 to 1.85 billion gallons in 2023. You may reasonably wonder why on earth someone would choose to build a data centre in the desert, given the water issue. It is because such areas offer cheap, abundant land, and often allow reliable renewable energy such as solar power in a sunny environment.
Conscious of public backlash, some companies are making an effort to be more sustainable. Google replenished 78% of its freshwater consumption in 2025, and plans to raise this to 120% by 2030. Microsoft came up with a data centre design in 2024 that should eliminate the need for water consumption for cooling. Certainly, closed-loop cooling systems that use dry cooling can greatly reduce water consumption, though they may increase electricity demand. Putting data centres in cool areas (think Alaska or Iceland) reduces cooling requirements, but may introduce network latency issues. As the IEA notes, reducing water consumption of data centres typically has a trade-off in higher electricity needs, and power generation is another bottleneck for data centres.
According to the International Energy Agency, electricity consumption from AI-focused data centres is expected to triple between 2025 and 2030, equivalent to annual growth of approximately 25%. Overall data-centre electricity consumption is expected to roughly double over the same period. It is likely that the demand for AI data centres will not in itself cause a global water crisis, but it may very well aggravate existing water shortages.
Bear in mind that the same data centre growth will stress electricity grids too. A single large 1 GW data centre operating continuously is roughly equivalent to the electricity consumption of three million UK households. In July 2026, the UK Office of Gas and Electricity Markets (Ofgem) reported that applications for new electricity-demand connections had risen from 41 GW to 125 GW between November 2024 and June 2025, with data-centre projects accounting for at least 80 GW of the applications. Some of these requests may never materialise, but this is likely to put pressure on all aspects of the electricity infrastructure, including transmission networks, which may need to be upgraded. There are steps that can be taken to minimise the problems, such as building dedicated new generating capacity (power stations), locating data centres near power hubs where possible, and using on-site battery technology to reduce the pressure on the grid. However, building extra gas-fired generation can relieve grid shortages but will increase carbon emissions. It can be seen that water consumption, power consumption and carbon emissions are inextricably linked in the world of expanding AI data centres.







