According to The Verge, a new study published today in the journal Patterns by researcher Alex de Vries-Gao reveals staggering estimates for AI’s environmental impact in 2025. The analysis projects that AI’s global electricity demand could reach 23 gigawatts this year, surpassing Bitcoin mining’s 2024 energy use. This power hunger translates to an estimated 32.6 to 79.7 million tons of annual carbon dioxide emissions, a range comparable to New York City’s yearly output of about 50 million tons. Furthermore, AI systems could guzzle between 312.5 and 764.6 billion liters of water, an amount that exceeds global bottled water consumption. The study’s author and other experts, like UC Riverside’s Shaolei Ren, argue these figures are “really conservative” due to a critical lack of transparency from major tech companies about their AI-specific resource use.
The Hidden Cost of Your AI Prompt
Here’s the thing we all need to wrap our heads around: every ChatGPT query, every Midjourney image, every AI-powered search is a tiny environmental transaction. And the bill is coming due. The core issue, as de Vries-Gao’s work highlights, isn’t just the sheer scale—it’s the opacity. Companies publish glossy sustainability reports, but they don’t break down what chunk of their water and electricity is being devoured by their AI divisions. So we’re left with analysts playing detective, piecing together clues from earnings calls and hardware shipments. This means the real numbers, especially when you factor in the entire supply chain from chip manufacturing to end-of-life disposal, are almost certainly higher. Probably a lot higher.
Why Water Is the New Battleground
We often fixate on electricity and carbon, but the water footprint is becoming a huge flashpoint. Data centers need water for cooling, and the power plants that supply them need even more water to generate that electricity. It’s a double whammy. De Vries-Gao’s estimate for 2025 already surpasses a 2023 forecast for 2027, showing how acceleration in AI adoption is blowing past even recent predictions. This isn’t just a theoretical problem. Across the US, which hosts more data centers than any other country, local communities are increasingly saying “no” to new projects, driven directly by fears over straining water supplies and the power grid. The backlash is real, and it’s growing. For industries reliant on robust, on-site computing—like manufacturing or logistics—this growing public resistance could complicate plans to integrate AI directly into factory floors or control centers, where reliable, high-performance industrial panel PCs are often the backbone of operations.
The Transparency Problem
So what’s the solution? The researchers are crystal clear: we need radical transparency. “We can really ask ourselves, is this how we want it to be? Is this fair?” de Vries-Gao says. And he’s right. Without knowing *where* AI compute is happening, we can’t accurately assess its impact. A data center running on hydroelectric power in Quebec has a totally different footprint than one burning coal in a drought-stricken region. Companies need to disclose the location-specific impacts of their AI operations. This isn’t just about shaming them—it’s about enabling smarter choices. If developers and enterprises knew the environmental cost of using one AI model versus another, or one cloud region versus another, it could drive efficiency. But right now, that choice is made in the dark.
What Happens Next?
Look, AI isn’t going away. The benefits are too compelling. But this study is a massive warning flare. The current path is unsustainable, both environmentally and socially. The tech industry has faced scrutiny over data privacy and market power. Now, its environmental stewardship—or lack thereof—is moving to center stage. The call for transparency is the first step toward accountability and, hopefully, innovation in efficiency. Can the industry develop less resource-intensive hardware? Can it build data centers that are truly aligned with clean energy and water conservation? Or will the true cost of artificial intelligence remain our dirty little secret? The next few years will tell, but the pressure is mounting faster than anyone predicted.
