A new UN study has named Hong Kong’s data centres as some of the most carbon-intensive in the world, blaming the city’s heavy dependence on a fossil-fuel-powered energy grid. A government data centre in Cheung Sha Wan. Photo: Googlemaps. The report , titled “Environmental Cost of AI’s Energy Use,” examined the global carbon, land and water impacts of the infrastructure powering AI, saying that by 2030, data centres could consume 945 terawatt-hours. That is “nearly triple the combined annual electricity use of Pakistan, Bangladesh, and Nigeria, countries collectively home to more than 650 million people,” according to a UN press release. Lamma power station. File photo: Tom Grundy/HKFP. “Indonesia, India, and Hong Kong (SAR) are among the most carbon-intensive grids with carbon footprints 62%, 51%, and 43% higher than the global average, respectively. Poland and Mainland China rank lower with carbon intensities at 30% and 21% higher than the global average,” the UN University’s Institute for Water, Environment and Health said in a report on Wednesday.
In comparison, the carbon footprint of electricity in the US, Germany, and Italy is 18 per cent, 24 per cent, and 32 per cent below the global average, respectively.
Energy in Hong Kong is 67 per cent derived from fossil fuels, 32 per cent from nuclear and just 1 per cent from renewables, the report said. Energy sources for countries and territories across the world. Photo: UN. There is also a water footprint for cooling heat-intensive data centres, as well as a land footprint. “AI-related water consumption could equal the basic annual domestic needs of 1.3 billion people by the end of the decade, while its land footprint may exceed 14,500 square kilometres – roughly twice the size of the Jakarta metropolitan area,” the UN said.
However, Hong Kong was ranked among the lightest for water and land consumption, mostly because its energy mix does not rely on renewable energy sources, which require large amounts land. The environmental cost of data centres. Photo: UN. As a trade and logistics hub, with around 300 internet service providers, Hong Kong remains a prime location for data centres. Its telecommunication networks connect to 12 external submarine optical fibre cable systems, with more under construction, according to the city’s Digital Policy Office.
The government is building a new 110,00 square metre data facility in Sandy Ridge, 90 per cent of which will be dedicated to data centres, according to a government press release in March.
Daily AI use, not training
The UN report said that day-to-day use of AI models accounted for around 80 to 90 per cent of total energy demand, as opposed to just model training. It cited the case of ChatGPT, which was processing around 2.5 billion prompts per day, with image generation requiring a thousand times more energy than a simple text query.
“China’s DeepSeek, launched in January 2025, attracted more than 20 million daily active users within three weeks, and had about 125 million monthly active users by mid-2025,” the report said. An aerial view of Alibaba’s Zhangbei data centre cluster in Hebei, China. Data sources: Epoch AI; Sentinel-2 false-colour imagery, February 2026. Photo UN. According to the Digital Policy Office website , “data centre operators are all striving to enhance energy efficiency , so as to reduce their power consumption, their operating expense and also their environmental impacts.”
It cites existing policies by the Electrical and Mechanical Services Department (EMSD), which set rules for ensuring the energy efficiency of buildings and regulate the use of fresh water in cooling towers for air conditioning systems. The EMSD’s Green Data Centres Practice Guide lays out initiatives for efficient data centre design, procurement, operations and disposal, whilst also promoting the use of assessment tools to measure environmental impacts.