What We Burn to Speak to Machines
Inside the Global AI Buildout: Power, Water, and Sacrifice Zones
Years ago, I found myself in Las Vegas walking into a data center with a political leader I was working for at the time (Yes, I used to work inside politics). Before we even reached the building, men in black tactical gear stood at attention, assault rifles slung across their chests. It felt less like visiting a piece of “important infrastructure” and more like approaching a border checkpoint into some fortified authoritarian nation-state.
What struck me most was the sheer size of the place. I had no real mental template for it. The building stretched on and on, a sealed industrial glacier. And inside: the roar of machines, the thrum of endless servers stacked like metallic lungs, and the cooling systems.. wow. They were monstrous, almost geological in scale, great steel organs wheezing to keep the silicon from melting under its own labor.
And this was before artificial intelligence. Before the explosion of GPUs and chatbots, before the rise of multimodal video models or the billion-parameter behemoths people now casually summon on their phones. Back then, these places already felt too large, too hungry, too armored for comfort, particularly when they are surrounded by residential homes.
Walking through those halls, I remember thinking: This is what the cloud looks like when it touches the ground.
Not soft, not weightless. A place that exists somewhere, even if our language insists that our digital lives float placelessly above the world.
Today, AI companies want us to imagine intelligence materializing effortlessly from light and air. They tell us it comes from “the cloud,” as if it were some neutral, atmospheric resource… as if we weren’t directing torrents of energy, rivers of water, and an army of minerals into these systems.
Clouds don’t need armed guards. Clouds don’t drink millions of gallons of freshwater a day. Clouds don’t demand gigawatts of electricity just to keep themselves from overheating.
AI doesn’t live in the sky. It lives in buildings like the one I visited. It lives in deserts and floodplains, in drought-stricken states and fragile biomes, in the midst of communities who never asked to host it.
In other words, AI lives in places. And those places are being reshaped, sometimes quietly erased, to keep our illusion of weightlessness alive.
This is the truth we must begin with. The cloud is a metaphor. But the consequences are not.
The Scale We Refuse to See
When people talk about AI, they often speak in abstractions: “models,” “agents,” “the cloud,” “digital transformation.” But the infrastructure behind AI is not abstract at all. It is concrete. Literal concrete, poured by the ton. It is steel, wire, copper, compressors, water pipelines, diesel tanks, and transmission lines humming with electricity pulled from distant rivers, mountains, and mines.
If we peel back the metaphor, what we find is staggering.
The United States alone now houses data centers that consume roughly as much electricity as many entire nations. And they are expanding at a pace that even utilities admit they can’t keep up with. One analysis suggests that nearly half of the world’s data-center electricity consumption occurs in the U.S. The new AI-optimized facilities are larger, hotter, thirstier, and far more power-hungry than their predecessors.
And outside the U.S.?
The explosion is even more dramatic.
Across Mexico, Brazil, Malaysia, Indonesia, Kenya, South Africa, and Ireland, hyperscale data centers are rising like industrial cathedrals. These are not the modest server rooms of the early internet era. These are multi-million-square-foot complexes built with the assumption that the future will demand more computation, more inference, more generation, more energy. Always more.
In Querétaro, Mexico, thirty-two data centers are planned in a region already experiencing historic drought. In Dublin, protests erupted when communities realized that cloud campuses were drawing down water reserves and overwhelming an aging electrical grid. In Brazil’s northeast, renewable energy intended to serve local people is being devoured by new AI facilities built to handle data from North America and Europe. In Malaysia and Indonesia, officials are already warning of grid instability. In Kenya and Nigeria, rolling blackouts are colliding with the arrival of foreign-owned data megacomplexes.
This is a global land rush, except the land is not the prize. The water is. The electricity is. The minerals are. The stability of local grids is. And the silence or political weakness of local communities is.
What’s emerging is a planetary pattern. AI’s growth does not follow human need.
It follows cheap power, weak regulation, low political resistance, tax incentives, and geographical convenience to undersea fiber cables.
And the geography tells a story that innovation propaganda tries very hard to hide:
the story of sacrifice zones.
Because for every brightly lit stage presentation about the future of intelligence, there is a community living in the shadow of a hyperscale facility. There is a watershed being tapped. There is an ecosystem losing resilience. There is a grid pushed to failure. There is a river warmed or drained. There is a valley threatened by mining for copper or lithium or uranium.
AI in the Numbers
A brief accounting of the water, energy, and materials flowing into the AI revolution.
Water
By 2027, AI data centers could consume 6.4 trillion liters of freshwater annually. That’s enough to fill 2.8 million Olympic swimming pools.
Microsoft already draws 42% of its water from regions officially classified as “water-stressed.”
Google reports 15% of its data-center water use occurs in “high water scarcity” areas, and that number is rising.
A single hyperscale AI complex can require millions of gallons per day just for cooling.
Water withdrawals for AI compete directly with human drinking water, agriculture, and the ecological flows that keep rivers alive.
Energy
Global data-center power consumption is projected to reach ~1,000 TWh/year by 2030, or roughly equivalent to the electricity use of Japan.
AI will likely account for 40–60% of this surge, making it one of the fastest-growing energy loads on the planet.
U.S. data centers alone already consume as much electricity as entire nations, and expansion pressures utilities to revive fossil fuel generation.
Coal plants in the U.S. and Europe are being kept alive, or reactivated, specifically to feed AI server clusters.
Nuclear power is being revived at an unprecedented scale to meet AI’s 24/7 baseload demand, triggering a new wave of uranium mining on Indigenous land.
AI is not “efficient.” It is becoming one of the most energy-intensive technologies humanity has ever built.
Materials and Mining
To meet AI’s copper needs alone, the world may need to mine as much copper in the next 25 years as in all of human history to date.
AI chips require 10–15× more energy and water to manufacture than standard chips.
Server lifespans are shrinking to 3–5 years, accelerating the world’s fastest-growing toxic waste stream: e-waste.
Each AI accelerator chip contains layers of rare metals that are currently impossible to recycle economically.
New mining and processing demands fall overwhelmingly on the Global South: the Atacama Desert (lithium), the Congo (cobalt), Indonesia (nickel), the Navajo Nation and Kazakhstan (uranium), Chile and Peru (copper).
The “intelligence” of AI rests on an explosion of extraction, massive new wounds on land, watersheds, and communities.
The Heresy of Limits and the Fantasy of Reform
There comes a moment in every civilization when it must confront the question it has been running from: What are the limits? Not the limits of technology but the limits of the planet. Not the limits of imagination but the limits of extraction, of water, of soil, of energy, of communities that refuse to disappear quietly.
We are living in that moment now.
The AI industry speaks in the language of boundlessness. More compute. More capability. More speed. More scale. A future where every interaction is mediated by a model. A world optimized, predicted, inferenced, and computed into submission.
But the Earth speaks a different language. Dry riverbeds. Drained aquifers. Mining scars visible from space. Communities forced to fight for drinking water that will be diverted to server-farm cooling loops. Uranium dust drifting across the lands of people who have already buried too many relatives.
To speak of limits in an age like this feels almost heretical. But limits are not the enemy of life; they are the condition that makes life possible.
Indigenous leaders around the world have been trying to teach this for centuries.
Deep ecologists have joined that chorus. Every watershed, every forest, every carbon cycle is already telling us the same story.
The world is finite. The cost of denial is collapse.
Yet here is the part most commentators refuse to say out loud. The people who are profiting from AI’s expansion know all of this, and they do not care.
Why would they?
The companies building hyperscale data centers receive tax incentives, water-rights guarantees, energy discounts, and political protection. They have the ear of presidents and prime ministers. Their lobbyists ghostwrite legislation. Their PR teams shape the narrative. And their financial power dwarfs that of most nations.
To believe that this system will reform itself is, frankly, a fool’s errand.
Nothing in history supports the idea that entrenched power voluntarily restrains itself.
Not plantation owners. Not mining barons. Not oil companies. Not tech monopolies.
Every meaningful change, from labor protections to civil rights to environmental laws, was won because people made demands that power could no longer ignore.
Frederick Douglass said it best:
Power concedes nothing without a demand. It never did and it never will.
And we must be honest enough to admit what that implies:
If we want limits… real limits, ecological limits, democratic limits, moral limits, they will not be granted. They must be imposed. By communities who refuse to sacrifice their water. By Indigenous nations asserting sovereignty. By cities and regions that refuse to host data centers in drought zones. By social movements willing to challenge the mythology of “inevitable technological progress.”
This is where the cultural shift becomes critical.
It is not enough to resist the worst harms of AI’s expansion; we must also unlearn the worldview that makes these harms seem reasonable or inevitable. The worldview that says growth is god. That efficiency is virtue. That more technology is always better, safer, wiser. That the Earth is simply a backdrop for human ambition.
We need a culture that honors restraint more than acceleration. A culture that values place more than abstraction. A culture that remembers the intelligence of land, water, and community is older than any algorithm.
Our survival depends on it.
It is the radical act of saying “The world has limits. And we choose to live within them.” Not because we are small, but because we understand that respecting limits is the essence of belonging to a watershed, to a biome, to a community, to a future.
Power will not hand us this shift. But the Earth is demanding it. And if history teaches anything, it is that when people demand loudly enough, fiercely enough, with enough solidarity and imagination, even the largest empires can be made to kneel.
How many lives must be lost? Rivers sucked dry? Ecosystems destroyed? What needs to happen before we stand up and demand that it end?
I ask myself this question every day. Is my writing enough? Are my volunteer hours restoring damaged ecosystems enough? I don’t know. But I’m trying like hell. And I know many of you are too. I’m encouraged every day when I open up Substack and see what others are doing and writing about.
Thank you all, and thank you for your support here on Substack. It makes all the difference. We’ll keep writing about the problems and how we can respond here at Collapse Curriculum.
Sources:
MIT Technology Review.
O’Donnell, James & Casey Crownhart. “We did the math on AI’s energy footprint. Here’s the story you haven’t heard.”
May 20, 2025.Mongabay.
McGovern, Gerry & Branford, Sue. “AI data center revolution sucks up world’s energy, water, materials.”
November 14, 2025.
Lawrence Berkeley National Laboratory (LBNL).
United States Data Center Energy Use Report.
U.S. Department of Energy, 2024.
(Key numbers: U.S. data centers ~200 TWh/year; AI-specific servers 53–76 TWh; projected AI load 165–326 TWh by 2028.)International Energy Agency (IEA).
Electricity 2024 / Energy Demand from AI.
Paris: IEA, 2024–2025.
(Key numbers: global data centers ~536 TWh in 2025; trending toward ~1,000+ TWh by 2030.)Goldman Sachs Research.
Global Data Center Power Demand Forecast.
2025.
(Key numbers: AI to drive ~165% global data-center power increase by 2030.)Shaolei Ren et al., University of California Riverside.
AI’s Water Footprint Research.
2023–2024.
(Key numbers: AI data centers may require 6.4 trillion liters of freshwater annually by 2027.)United Nations University.
Global E-Waste Monitor.
2024.
(Key context: AI server lifespan 3–5 years; toxic waste from accelerators non-recyclable.)