Do you need ChatGPT to find a recipe?
AI data centers consumed 448 TWh of electricity in 2025, ranking 11th globally; projected to reach 6th place by 2030 with emissions matching UK's annual output. Water consumption could reach 9.3 trillion liters by 2030, equivalent to entire sub-Saharan Africa's annual needs; ChatGPT alone processes 2.5 billion daily requests.
- Data centers consumed 448 TWh of electricity in 2025, ranking 11th globally; projected to reach 6th place by 2030
- Water consumption could reach 9.3 trillion liters by 2030, equivalent to sub-Saharan Africa's annual needs of 1.3 billion people
- ChatGPT processes 2.5 billion requests daily, consuming 383 GWh annually—enough to power 3 million sub-Saharan Africans
- AI market projected to grow from $189 billion in 2023 to $4.8 trillion by 2033
- Most AI data centers located in US, China, and EU; environmental costs of mining and waste fall on developing nations
UN report warns AI's explosive growth is consuming water equivalent to 1.3 billion people's annual needs and electricity matching France's usage, urging transparency and behavioral changes to mitigate environmental crisis.
A United Nations report released this week carries a stark warning: the artificial intelligence industry is consuming water at a scale that rivals the annual needs of 1.3 billion people, and its electricity demand is climbing toward a rank that would place it among the world's top energy consumers. The findings arrive as the AI market itself is projected to balloon from $189 billion in 2023 to $4.8 trillion by 2033—a expansion that will only deepen the environmental footprint.
The numbers are substantial enough to demand attention. Data centers that power AI systems and other digital services consumed 448 terawatts-hours of electricity in 2025, a figure that would rank them eleventh globally in energy use, just behind France. By 2030, that consumption is expected to climb to roughly 945 terawatts-hours annually, pushing data centers into sixth place worldwide. The carbon emissions from that electricity use alone would equal approximately 399 million tons—exceeding the entire net emissions of the United Kingdom in 2025.
Water consumption presents an even more acute concern. The UN estimates that data centers could draw 9.3 trillion liters of water annually by 2030, an amount equivalent to the yearly freshwater needs of sub-Saharan Africa's 1.3 billion residents. ChatGPT, the most visible AI application, processes around 2.5 billion requests daily and consumes roughly 383 gigawatt-hours of electricity per year—enough to power nearly three million people in sub-Saharan Africa for a year. Video generation by AI systems proves especially thirsty; a single short video can demand as much electricity as hundreds of still images.
Kaveh Madani, director of the UN University Institute for Water, Environment and Health, framed the findings as preliminary. "What we're showing here is probably only the tip of the iceberg," he told the AFP. The institute's report calls for mandatory transparency from AI developers and providers, demanding they disclose the environmental footprint of both model training and user-facing responses. Governments are urged to require standardized environmental reporting from AI companies and to site new data centers away from regions already facing water scarcity.
The burden of this expansion, however, falls unevenly across the globe. Most specialized AI data centers cluster in the United States, China, and the European Union, yet the environmental costs of mining the raw materials and processing the resulting waste concentrate heavily on developing nations that reap few of the economic benefits. This geographic mismatch—prosperity concentrated in wealthy nations, environmental damage dispersed across poorer ones—underscores what researchers describe as a digital divide with material consequences.
The UN's recommendations extend to user behavior. Miriam Aczel, a researcher at the institute, posed a simple question: "Do you need ChatGPT to find a recipe?" Internet searches powered by AI can consume up to ten times more energy than conventional searches. The report urges people to reserve AI tools for tasks that genuinely require their capabilities, rather than reaching for them out of habit or convenience. Madani emphasized that the report is not anti-AI but rather a call for proactive oversight. "We must monitor its impacts so we can mitigate them, so we can control them before it's too late," he said. The window for establishing guardrails, the UN suggests, is narrowing as the industry accelerates.
Citas Notables
What we're showing here is probably only the tip of the iceberg.— Kaveh Madani, UN University Institute for Water, Environment and Health
We must monitor its impacts so we can mitigate them, so we can control them before it's too late.— Kaveh Madani
La Conversación del Hearth Otra perspectiva de la historia
Why does the UN frame this as urgent now, rather than waiting for more complete data?
Because the trajectory is clear and accelerating. They're watching a technology that doubles in capability every few years while consuming resources at an exponential rate. Waiting for perfect data means waiting until the problem is already embedded in infrastructure.
The report mentions this is "probably only the tip of the iceberg." What's hidden beneath?
The full lifecycle costs. Right now they're measuring electricity and water at the data center. But they're not fully accounting for the mining of rare earth minerals, the manufacturing of chips, the cooling systems, the disposal. And as AI gets embedded in more devices—phones, cars, sensors—those costs multiply.
Why does it matter that data centers are concentrated in three regions?
Because it creates a moral hazard. The companies profiting from AI are in wealthy nations with stable water supplies and reliable grids. The countries bearing the extraction and waste costs are often in regions already stressed by drought and poverty. It's a transfer of environmental burden.
The report suggests users avoid AI for simple tasks. Is that realistic?
Probably not at scale. But it's not really about individual recipe searches. It's about signaling that convenience isn't the only value that matters. If enough people start questioning whether they need AI for something, it changes how companies think about deployment.
What would real accountability look like?
Mandatory disclosure of environmental costs the way companies disclose financial costs. Governments setting hard limits on water use in data-center regions. And pricing that reflects the true environmental cost, not just the electricity bill. Right now, the market doesn't see water depletion as expensive.
Does the UN think AI itself is the problem, or how we're using it?
They're careful to say it's not anti-AI. The technology itself isn't inherently wasteful. It's the assumption that every task benefits from it, and the lack of any friction in deployment. If using AI cost something visible—in water, in carbon—people would use it differently.