Please, No, Please, No — A Plea for Nature in the Age of Percolating AI

Some fears deserve to be spoken the way a child speaks them — without sophistication, hands folded, twice for emphasis. Please, no. Please, no. That is the correct prayer for our moment, as artificial intelligence percolates into every crack of human life — every phone, every office, every farm, every school — and sends its bill, quietly, to the one account that cannot refuse payment: the Earth's.

This is not a plea against AI. It is a plea against a particular carelessness — the assumption that intelligence can be poured infinitely into the world without asking what it is poured through. Because the answer, today, is: through power plants, through rivers, through mines, through land. And each of those deserves its own refrain.

Please, no — not the air

Begin with the largest number. Data centres already draw around 1.5% of all electricity generated on Earth — roughly the consumption of a mid-sized industrial nation — and the International Energy Agency expects this to more than double by 2030, with AI as the primary driver. Behind every effortless chat reply stands a warehouse of processors running hot, day and night. Training a single frontier model consumes gigawatt-hours; but the deeper weight is inference — billions of small queries a day, each trivial, together enormous. A drizzle, falling forever, is a flood.

And here is the fact that should end all complacency: the companies leading the AI race have watched their own emissions rise — by a third, by nearly half — in the very years they had pledged to cut them. Their climate targets, announced with fanfare, are being quietly missed, one data-centre groundbreaking at a time. When the richest, most engineering-proud institutions on the planet cannot keep their carbon promises under AI's hunger, the plea writes itself. Please — not the air. Not another decade of gas turbines spun up to make chatbots faster. The atmosphere has no delete key.

Please, no — not the water

The quieter cost is the one that should trouble India most. Data centres drink. To cool their racks, they evaporate millions of litres of clean, treated water — often in exactly the regions already counting every drop. The chips themselves are born thirsty: a single semiconductor fabrication plant consumes water on the scale of a small city, purified to a fineness rivers never know. Communities from Arizona to Chile have already stood up to ask why their aquifers should subsidise the world's queries. And now the data-centre wave is arriving on our own map — Hyderabad, Vizag, Chennai — cities courting server farms as trophies while their summer taps already run uncertain. By all means, bring the investment. But please — not the water. Not groundwater traded for gigaflops in a country where farmers still watch the sky. Every project should answer one question before the first brick: whose water, and what returns to them?

Please, no — not the land and what lies beneath

Intelligence, it turns out, is heavy. It is copper and lithium and cobalt and rare earths, clawed from the ground, refined in furnaces, shipped across oceans. It is thousands of acres flattened for server halls and substations and transmission corridors. It is e-waste — yesterday's accelerators, obsolete in three years, leaching into soil somewhere the buyers of intelligence will never visit. None of this is unique to AI; all of it is accelerated by AI, because the buildout is the fastest infrastructure race since the railways. Railways, too, were progress. They also taught us what progress does when nobody speaks for the forest in its path. Please — not the land, not carelessly. Mine, but restore. Build, but site wisely. Retire hardware into recycling chains, not riverbanks.

Please, no — not the trap of efficiency

Here the plea must grow sophisticated, because the comforting answer — "the chips are getting efficient!" — hides a trap with a name: Jevons paradox. When coal engines became efficient, Britain burned more coal, not less, because efficiency made steam cheap enough for everything. AI is racing down the same slope: the energy per query has collapsed, and precisely because of that, queries are exploding — into toys, ads, spam, idle amusement. Efficiency without restraint does not shrink the footprint; it multiplies the feet. So please — no worshipping efficiency alone. The question is not only "how little does each use cost?" but "which uses are worth the cost at all?" A civilization that burns a forest to generate entertainment has not computed wisely, however efficient its chips.

The other side of the ledger — and the honest forecast

A fair plea must admit the defence, and AI's defence is real. It is one of the few technologies with a credible claim to repaying its environmental debt. It already balances power grids and sharpens renewable forecasts; it cut cooling energy dramatically in the data centres it inhabits; it is accelerating the discovery of better batteries, catalysts, and materials; it tightens the climate models we navigate by. Its hunger for clean, constant power is doing what decades of lobbying could not — pushing tech capital into restarted nuclear plants, small modular reactors, and solar contracts of historic size. Used deliberately, the tool could become the gardener.

So here is the honest forecast, without comfort or panic: a J-curve. For the next three to five years, the costs win — emissions from the buildout will outrun the offsets, water and grid stress will become local flashpoints, more climate pledges will bend. Around the turn of the decade, the curve can bend back — if efficiency, clean-power buildout, and genuine scientific payoffs converge. And in the long run, the outcome is not destiny but decision. If AI meaningfully accelerates fusion, storage, or carbon removal, it will rank among the best environmental investments our species ever made. If it mostly manufactures distraction while running on gas, it will be remembered as a luxury the atmosphere paid for. Both endings remain open. The next ten years choose between them — and the damage done in the gap is the damage that lasts.

What the plea asks

"Please, no" is not a policy. So let the refrain end in demands simple enough to check. Power every new data centre with clean energy — matched hour by hour, not laundered through certificates. Disclose water use publicly, site cooling away from stressed basins, and return treated water to the communities that lent it. Price carbon into compute so the true cost appears on the invoice, not on the glacier. Recycle the hardware. Publish the footprint of training runs the way food publishes calories. And reserve the greatest computational feasts for the greatest human needs — medicine, energy, climate, learning — not merely the most profitable appetites.

None of this stops AI. All of it disciplines the manner of its percolation — which is all nature has ever asked of any of our revolutions. Fire, farm, factory, and now this: each time, the Earth's request was the same. Take what you need. Restore what you take. Do not make me the silent line item on the bill.

Please, no — not this time. This time, let us prove that the most intelligent thing we ever built taught us, at last, to be intelligent about the only home we have.