‘The Cloud’ Is a Physical Place — and It's Thirsty and Hot

Every time someone tells me their photos, their messages, or their company’s data live “in the cloud,” I hear a small act of misdirection. The word is soft, weightless, almost spiritual — a wisp of vapour floating somewhere above us, untouchable and clean. But the data centre environmental impact behind that word is anything but ethereal. The cloud is a physical place. It is a warehouse, or a whole campus of them, packed wall to wall with hot computers that never sleep, drinking electricity and water around the clock. There is an old line among engineers that captures it perfectly: there is no cloud, just someone else’s computer. I would go further — there is no cloud, just someone else’s computer, in someone else’s building, on someone else’s land, cooled by someone else’s water and powered off everyone’s grid.

I want to pull that word down out of the sky and set it on the ground, where it belongs, because once you see the cloud as a physical thing, you start asking physical questions. Where is it? What does it eat? Who lives next to it? And who pays the bill when the meter runs?

What the word “cloud” hides

Marketing chose “cloud” for a reason. A cloud has no address. It has no smell, no hum, no fence, no transformer yard. It cannot be a nuisance to your neighbourhood because it isn’t anywhere. This is a beautiful thing to sell, because it lets a company move your entire digital life into buildings you will never see and bill you monthly for the privilege, all while the physical consequences stay comfortably out of frame.

Strip the branding away and here is what is actually there. A data centre is a large, windowless building filled with racks of servers — the same kind of machine as the laptop in front of you, only there are tens of thousands of them, and they run flat out, all day, every day. Servers turn electricity into computation and, as a by-product, into heat. A room full of them is essentially a room full of electric heaters that must be cooled continuously, or the machines cook themselves. Feeding the building is a spine of physical infrastructure most of us never picture: high-voltage power lines and substations, diesel generators for backup, banks of batteries, and thousands of kilometres of fibre-optic cable stitched under streets and oceans to carry the bits in and out.

None of that is weightless. All of it occupies land, draws power, and demands cooling. The “cloud” is one of the most concrete industries humanity has ever built. We just agreed not to look at it.

There is no cloud, just someone else’s computer — in someone else’s building, on someone else’s land, cooled by someone else’s water.

What these buildings actually consume

Once you accept that the cloud is a place, its appetite stops being abstract. A large data centre consumes three things in enormous quantity: land, electricity, and water. And with the current stampede into artificial intelligence, every one of those numbers is climbing steeply.

Land. These are big footprints, and they cluster. Companies want to build near cheap power, cool climates, fast fibre, and friendly tax regimes, so data centres arrive in packs. In India, hubs around Mumbai, Chennai, Hyderabad and the Delhi belt are filling with them as global platforms localise storage and race to host AI workloads. Each campus takes real estate off the table for housing, farming, or anything else a community might have wanted from that ground.

Electricity. This is the big one. Estimates suggest data centres already account for a meaningful and fast-rising share of global electricity demand, and AI is pouring petrol on that fire — training and running large models is far hungrier than the ordinary web ever was. The uncomfortable part, especially in India, is where those electrons come from. A grid still leaning heavily on coal means a “cloud” that runs substantially on fossil fuel. When a company boasts about a shiny new AI feature, it rarely mentions the extra generation quietly spun up to power it. I’ve written separately about AI’s energy consumption, and the trajectory there is the single most important number in this whole story.

Water. This is the consumption people find hardest to believe, so it’s worth saying plainly: many data centres cool themselves with water, evaporating enormous volumes to shed the heat their servers produce. Reportedly, a single large facility can draw on the order of a small town’s worth of water. In a country where taps in some cities run dry for parts of the year, siting thirsty server farms in water-stressed regions is not a footnote — it is a direct competition between a corporation’s cooling towers and a neighbourhood’s drinking supply. I’ve unpacked the numbers and the regional stakes in the hidden water cost of AI; here I only want you to hold the image: the cloud is thirsty, and it drinks locally.

The hardware doesn’t vanish either

There is a fourth consumption that outlives the electricity bill: the machines themselves. Servers are not eternal. GPUs and chips age out, get superseded by faster silicon, and are ripped from the racks — often after just a few years — to make room for the next generation. The AI boom is accelerating this cycle, because yesterday’s accelerator is worthless the moment a better one ships.

All that discarded hardware becomes e-waste — a stream of toxic, hard-to-recycle metal and plastic that, in India especially, too often ends up dismantled by hand in the informal sector, poisoning the people and soil that process it. The mining that produced those chips in the first place carried its own environmental cost, on the other side of the world, before the server ever powered on. So the footprint isn’t only the energy and water a data centre uses while running; it’s bookended by extraction at the start and disposal at the end. Follow any of those threads and you’ll find the same detail I keep — the burden lands somewhere physical, on someone specific.

This is enclosure, dressed up as innovation

Here is where I stop describing and start arguing, because I’ve seen this shape before. Power, water, land, clean air — these are a commons. They are the shared physical inheritance that no company created and everyone depends on. And what a data centre does, at its core, is draw down that commons for private profit.

The platform gets the revenue, the market value, the press release about being at the frontier of AI. The community near the campus gets the truck traffic, the strain on a shared aquifer, the transformer hum, and a claim on grid capacity that might otherwise have kept their own lights on. The costs are socialised — spread thin across the public grid, the local watershed, the atmosphere we all breathe — while the profits are privatised, concentrated in a handful of firms most of whose names you already know.

This is the capture pattern, and it is remarkably consistent. A new technology arrives wrapped in the language of the future. Its physical demands are quietly externalised onto the public. And by the time anyone tallies the true cost, the infrastructure is built, the dependence is locked in, and the beneficiaries are too large and too useful to challenge. I’ve traced this same manoeuvre across other industries in how technology gets captured; the cloud is simply its latest, and perhaps its most literal, expression — an enclosure not of a village green but of the electrons and water molecules we all share.

The platform gets the revenue and the frontier-of-AI headline. The neighbourhood gets the truck traffic, the strained aquifer, and the claim on a grid that might have kept its own lights on.

Who pays, and who profits

It’s worth being blunt about the two ledgers, because the gap between them is the whole point.

Who pays. The family whose borewell runs shallower each summer. The town whose grid was promised to residents and industry, now sharing capacity with a server farm. The informal worker breathing fumes off a dismantled circuit board. The public exchequer that hands out tax breaks and cheap land to attract the campus. And, diffusely, all of us, through a warming atmosphere that every extra coal-fired megawatt-hour makes a little worse.

Who profits. The global platforms whose apps, ads, and AI models run on that hardware, and who book the value while the physical costs stay off their balance sheet. When you see the asymmetry stated this plainly, the softness of the word “cloud” starts to look less like poetry and more like strategy.

What accountability could actually look like

I am not anti-data-centre. I write these articles on machines that live in one; the digital world I love is only possible because these buildings exist. The problem isn’t that the cloud is physical — everything is physical. The problem is that we’ve been encouraged to pretend it isn’t, and that pretence is exactly what lets the costs go unpriced and unwatched. Naming the building is the first act of accountability. Here is what the rest could look like.

• Clean power, for real. Data centres should be matched to genuinely additional renewable generation — new solar and wind built for the load — not creative accounting that claims green credits while the local grid burns more coal to keep up. In a coal-heavy grid like India’s, this is the difference that matters most.
• Transparent siting. Before a campus is approved, its projected water draw and power demand should be public, and weighed honestly against the region’s existing stress. A thirsty facility in a water-scarce district is a choice communities deserve to see made in the open, not discover during the next drought.
• Mandatory disclosure. Operators should report the electricity, water, and carbon behind their services, in numbers a citizen can check. Right now the exact figures are contested largely because the companies best placed to publish them don’t. Sunlight would settle the argument.
• Own the full lifecycle. Responsibility shouldn’t end when a server is unplugged. The firms that generate mountains of retired hardware should be on the hook for its safe recovery, not free to let it drain into an informal recycling economy that poisons the poorest.
• Give something back locally. A campus that draws on a community’s power, water, and land should return more than a handful of security jobs — grid investment, water restoration, real local benefit, so the ledger of who-pays and who-profits comes closer to balancing.

None of this is radical. It’s just what we’d expect of any heavy industry that touched the shared essentials of life — and make no mistake, that is what this is. The cloud is a factory. It runs hot, it runs thirsty, and it runs on a commons we all own but few of us are asked about.

So the next time an app cheerfully offers to back your life up “to the cloud,” picture the real thing. A vast, humming shed somewhere near a city you know, its cooling towers exhaling water into the sky, its power lines glowing off a coal plant over the horizon, its old servers waiting for the scrap heap. That’s the cloud. It was never in the sky. It was always right here, on the ground, next to someone — and the least we can do is stop looking away.