Every time someone generates an image, drafts a document with an AI assistant, or asks a chatbot for directions, a data centre somewhere draws power, consumes water for cooling, and edges slightly closer to a hardware refresh that will eventually become electronic waste. Multiplied across billions of daily interactions, those individual costs add up to something the United Nations now considers urgent.
A report published in June 2026 by UN News and supported by the United Nations University found that data centres powering AI could consume 945 terawatt-hours of electricity annually by 2030. That figure is nearly triple the combined annual electricity use of Pakistan, Bangladesh, and Nigeria. The environmental costs of AI, the report concludes, are not a future problem to be managed with better efficiency. They are a present-tense burden, and one that is unevenly distributed.
Water, land, and waste
Electricity is only one dimension. Data centres require large volumes of water for cooling, and the water footprint extends into the energy production chain as well. The UN analysis projects that AI-related water consumption could reach the equivalent of the basic annual domestic needs of 1.3 billion people by the end of the decade. In regions already under drought stress, the siting of large AI facilities is becoming a genuine political and ecological conflict, not just an environmental footnote.
Electronic waste is the third pressure point. AI infrastructure depends on rapid hardware cycles, as the compute chips driving today’s frontier models become obsolete within two to four years. The report estimates that AI infrastructure could generate up to 2.5 million tonnes of e-waste annually by 2030. A large share of that waste will flow to regions with limited capacity to manage it safely.
The UN’s broader concern is geographic inequity. The economic benefits of AI are diffused globally, but the physical infrastructure is concentrated. Data centre clusters in Virginia, Ireland, Singapore, and a small number of other locations bear environmental costs that local communities did not choose and are not compensated for. In some countries, data centre electricity demand already represents a meaningful share of national grid load.
The other side of the ledger
The same report that documents AI’s environmental costs is careful to note that AI is also producing tools with genuine climate applications. Google DeepMind’s GraphCast model can produce a 10-day weather forecast in under one minute, dramatically faster than conventional numerical models. Faster forecasts mean faster extreme-weather warnings, which have direct consequences for disaster preparedness in vulnerable regions.
Climate tech investors are paying attention. Trellis identified 15 climate tech startups to watch in 2026, many of them using machine learning to optimize renewable energy grids, model carbon sequestration, and analyze satellite data for deforestation monitoring. The Columbia University Lamont-Doherty Earth Observatory described the situation as a genuine paradox: AI is simultaneously one of the fastest-growing contributors to energy demand and one of the most powerful analytical tools available to climate scientists.
What the industry is being asked to do
The UN’s recommendations center on transparency and accountability. Developers and operators should publicly report energy, water, and land use. Procurement decisions for AI infrastructure should factor in regional grid carbon intensity and water scarcity. Governments should require environmental impact assessments for large-scale data centre developments, just as they would for any major industrial facility.
Whether voluntary measures move fast enough to matter is the central question. The major cloud providers have made public commitments to run on clean energy by 2030. But those commitments were made before the current surge in AI workloads, and the timelines are already under pressure. SpaceX’s June 2026 reveal of the AI1, a 70-meter orbital supercomputer designed to run in space, suggests the industry is already looking at entirely new infrastructure models to escape the constraints of terrestrial power and cooling. It is not a solution to the current problem, but it is a signal of how seriously the infrastructure challenge is being taken. For more coverage of AI and climate, visit Mylistingo.
