You’re probably using AI to draft emails, generate code, or summarize long reports. It feels weightless. It feels clean. But every single prompt you enter triggers a physical reaction in a massive data center somewhere, burning through electricity and drinking thousands of gallons of water.
United Nations Secretary-General António Guterres just took the stage at London Climate Action Week to tell the tech industry that the party's over. He launched the AI Environmental Transparency Initiative, demanding that artificial intelligence companies stop hiding their carbon, water, and land footprints. "No more hidden costs," Guterres said. "It is time to come clean."
The real story here isn't just that AI uses energy. We already knew that. The real story is the sheer scale of the deception and the looming crisis that tech giants are trying to outrun.
Why Your AI Search Drinks More Than You Do
When you search for something on a traditional search engine, the server looks up an index. It's fast, and it's relatively cheap. When you ask an AI model to generate a response, the server has to run billions of mathematical calculations to predict the next word.
A June 2026 report from the UN University Institute for Water, Environment and Health puts hard numbers on this reality. By 2030, AI data centers are projected to consume around 945 terawatt-hours of electricity every year. That is more power than the combined annual electricity usage of Pakistan, Bangladesh, and Nigeria. We're talking about more than half a billion people's worth of energy, sucked up by rows of silicon chips.
But the tech industry loves talking about carbon offsets and green energy targets because it distracts from a much scarcer resource: freshwater.
Data centers get hot. To keep the chips from melting, operators rely on massive cooling towers. The UN report reveals that by the end of this decade, AI's water footprint could match the basic annual domestic water needs of 1.3 billion people. Right now, about 30% of the electricity consumed by data centers globally still comes from coal, according to the International Energy Agency.
The Net Zero Mirage
Go to the sustainability page of any major tech company, and you’ll find ambitious promises. They claim they'll be net-zero or carbon-negative by 2030. They buy renewable energy certificates (RECs) to match their power usage on paper.
It's mostly a shell game.
Data centers need power 24 hours a day, 7 days a week. Wind and solar power don't work that way. When the sun goes down, these facilities draw power from whatever is on the local grid—which usually means gas or coal. To make matters worse, US utilities are bracing for a $1.4 trillion capital surge to upgrade grids just to feed these centers by 2030. Projects like Meta’s gas-powered Hyperion campus in Louisiana prove that when the grid can't supply enough green energy, tech firms will gladly burn fossil fuels to keep the servers humming.
The tech elite are banking on nuclear energy or future breakthroughs to save them. Microsoft has signed deals to back nuclear power, and others are looking into small modular reactors. But those solutions are a decade away. The AI boom is happening right now, and the emissions are spiking today.
What Needs to Happen Next
Guterres isn't saying we should ban AI. He acknowledged that the technology can help optimize energy grids, model climate mutations, and find efficiencies we can't see. But we can't use a tool to solve the climate crisis when the tool itself is burning down the house.
If you are a tech leader, a developer, or an enterprise buyer, the current hands-off approach is a liability. Relying on vague corporate sustainability reports won't cut it anymore. True accountability requires immediate, systemic shifts in how we build and deploy software.
- Demand Granular Model Disclosures: Stop accepting blanket "carbon neutral" claims from LLM providers. Require API vendors to disclose the exact carbon intensity per 1,000 tokens generated, broken down by the time of day the inference occurred.
- Audit the Infrastructure Supply Chain: Move data workloads away from grids heavily reliant on fossil fuels. Prioritize hosting in regions like Iceland or parts of Scandinavia, where geothermal and hydro power provide genuine, 24/7 renewable baseload energy.
- Optimize for Efficiency over Size: Stop automatically deploying the largest available model for simple text classification or routing tasks. Use small, fine-tuned, open-source models that run at a fraction of the computational and energy cost.
- Enforce Local Water Accountability: Before signing contracts with cloud providers, evaluate the water scarcity of the data center's location. Push vendors to transition away from evaporative cooling toward closed-loop liquid cooling systems that don't deplete local aquifers.
The era of treating the environment as an unpaid line item on a corporate balance sheet is ending. Regulators are already moving, with the European Union instructing tech firms to align infrastructure with climate goals. If the industry doesn't volunteer its metrics honestly, governments will extract them through heavy penalties.
