Dematerializing the “Cloud” Myth
For a time, saying cloud made people think AI was invisible and weightless. It made it seem that whenever you type a question into a chatbot, your words just float away into a digital sky, hiding the heavy real‑world data centers that actually power AI. That is not enough here. It has always been misleading, and the gap between cloud myth and real AI has only grown with every new AI mode.
Every prompt is backed by a physical building full of servers creating enormous heat and staggering amounts of fresh water to cool that heat. A new issue brief from the Pacific Institute, published in September 2026, confirms that water withdrawals by data centers vary enormously depending on cooling technology, climate, and management practices, but the overall trajectory is clearly upward as AI facilities multiply.
By 2025, American data centers built specifically for AI workloads were already pulling close to one trillion liters of water annually, according to an analysis of investor and media disclosures compiled by SpaceDaily. Food and Water Watch has projected that American AI operations could require up to 720 billion gallons of water every year by 2028 just to keep servers cool, alongside as much as 300 terawatt-hours of electricity. None of this water simply disappears without consequence. Up to 85 percent of what data centers use through evaporative cooling never returns to the local supply at all, which means it is effectively lost to the communities surrounding these facilities.
The Multi-Token Tax: Why Next-Gen AI Reasoning Is Chronically Thirsty
The newest wave of so-called reasoning models, built to think step by step before answering, has quietly multiplied this problem. These systems generate long internal chains of thought before producing a final response, and every one of those hidden tokens still demands real computation and real cooling. A study found that reasoning queries producing around 5000 output tokens raise energy use roughly 13 times higher than a standard short answer and that even a modest share of these longer queries, say 10% of daily traffic, can more than double a data center’s total energy consumption. Benchmark testing on seven different reasoning models found that thinking tokens made up anywhere from roughly 10% to nearly 90% of total output and that reasoning models consistently burned far more energy than non-reasoning models of a similar size, in one case using more than fourteen times as much power. Water use tracks energy use closely, since cooling alone can represent 30% to 40% of a facility’s total energy draw. Put simply, the more a model appears to think, the thirstier it becomes, and the industry’s current race toward ever more elaborate reasoning architectures is quietly scaling up water demand along with it.
The Sovereign Gridlock: Data Center Opacity vs. Civil Infrastructure Survival
What makes this crisis particularly difficult to manage is that almost nobody outside the companies themselves actually knows the real numbers. Many data center developers decline to report their water consumption at all, and some local governments have reportedly signed NDAs that prevent them from telling their own residents how much water a nearby facility is using.
In Wisconsin, environmental advocates sued the city of Racine in an effort to force disclosure of projected water use tied to a Microsoft data center campus, arguing that communities have a basic right to know what is being drawn from their water supply before construction even begins. Texas faces a similar information gap, since the state has historically lacked any systematic way of tracking how much water its data centers actually consume.
This opacity is colliding head-on with strained civic infrastructure. Research from Data Center Watch found that local opposition blocked or delayed $68 billion worth of data center projects in just the second quarter of 2026 alone, driven by anger over land use, noise, and resource consumption. In Virginia, one AI facility reportedly triggered a 76% spike in nearby residents’ electricity bills, which eventually forced the state to require hyperscale companies to cover the cost of the transmission infrastructure their own projects demanded.
The Geopolitical Siting War: From Terrestrial Water Stress to Space-Based Compute
Faced with rising resistance at home, tech companies have increasingly gone looking for water and power wherever regulations are loosest, often in places that can least afford to give either one up. Investigations have documented new hyperscale facilities concentrated in drought-stricken parts of Chile, Uruguay, and Mexico, where protests have erupted over the fear that corporate servers are competing directly with residents for drinking water.
Court documents revealed that one proposed Google facility in Cerrillos, Chile, could consume roughly 7 billion litres of water annually, enough to supply 80,000 people. A separate mapping analysis found that around forty percent of major American data center locations sit in regions already classified as high or extremely high-water stress. This terrestrial scramble has pushed some of the biggest players to look somewhere far less contested, which is outer space.
On October 1, 2026, Google launched a prototype satellite under Project Suncatcher, testing whether its Tensor Processing Units can survive in orbit, where sunlight is nearly constant and up to eight times stronger than on Earth’s surface. Nvidia has unveiled its own Vera Rubin Space Module for orbital data centers, and companies including Starcloud and a SpaceX and Nvidia partnership called Starmind are chasing similar ambitions. As one NPR report bluntly put it, space offers virtually unlimited free energy and, notably, no protesters.
Redefining Resource Policy in the Age of Artificial Intelligence
None of this means artificial intelligence should be stopped. It does mean the discussion about it has been seriously missing important parts. Looking at water and electricity as expenses instead of major factors has let companies build places in areas that have little water, not share basic information about how much water and energy they use with the people living near them, and now try to leave Earth to avoid responsibility. Real change would need companies to share information about how much water and energy they use at each place, rules about where to build that really consider how much water is available in an area, and a real talk about how much it really costs to keep making machines think longer and longer. It turns out intelligence is not all light. It leaves a mark on water. Ignoring that only makes the problem worse for the rivers, underground water sources, and people who can’t handle more costs.
If you want to submit your articles and/or research papers, please visit the Submissions page.
To stay updated with the latest jobs, CSS news, internships, scholarships, and current affairs articles, join our Community Forum!
Find our FREE quizzes and MCQs here!
The views and opinions expressed in this article/paper are the author’s own and do not necessarily reflect the editorial position of Paradigm Shift.
Hooria Akbar is an independent researcher and content writer with a strong interest in contemporary issues across AI, technology, society, and emerging trends. She writes research-informed articles, thought pieces, and blogs on a wide range of topics, aiming to present complex ideas clearly and engagingly for diverse audiences.
