The cloud does not exist. There is only a substrate.

Twenty years ago, in a meeting room, someone decided to call “cloud” the heaviest, most material thing humanity has ever built: servers, cables, turbines, mines, millions of tonnes of concrete and silicon, and human labour that is often brutal.

They chose the name of something light, that sits in the sky and has no address. It worked extremely well, because protests gather around what can be seen and touched: a 40,000-square-metre shed that drinks your well water. No one has ever taken to the streets against something called a “cloud”.

Out of that successful illusion comes Substrato, a map I am building with Relatronica: it lines up infrastructures, links, and impacts of what sits underneath and holds everything else up. A geology of technology.

What sits behind a prompt

Every time you write to an AI, a physical chain starts: a mineral extracted under conditions no one narrates; a chip from a handful of factories; a data centre packed with computers that never sleep; and the plant that feeds it power.

Substrato currently maps 817 data centres, 1,068 energy plants, and 17 extraction and production sites across 36 countries, with 25 documented links. The numbers are still small: the aim is a fine-grained, updated, verified map.

Why they use so much

A server turns almost all the electricity it receives into heat that must be removed or the machines shut down. Two cost lines — power to run, power and water to cool — and they never stop.

According to the IEA, in 2022 data centres, crypto, and AI together used about 460 TWh — roughly as much as all of France in a year. With the AI push we could pass 1,000 TWh by 2026: a second Japan added to the bill.

Global numbers slide past until you look at where they land. Let’s talk about home.

The cloud under your house

A Politecnico di Milano survey counted 49 active data centres in Lombardy; 33 are in the Milan metro area — nearly half of Italy’s 170–200 facilities. Between 2023 and 2025 another 83 were announced, for about €5 billion in investment.

Data centres now account for about 2% of national electricity demand. Estimates put them at 7–13% by 2035. On Italian consumption of about 300 TWh a year, that means going from a couple of million households’ worth to several million.

Each data centre asks for a connection often equal to a small city’s power. Requests are growing faster than the grid was designed for. In the Pavia area residents are already fighting new plants. Lombardy has passed a data-centre law that parts of the regional council see as written more to speed permits than to protect land.

That is what the word “cloud” does: if infrastructure is immaterial, there is nothing to regulate or debate. Conflict and harm, if they do not feel concrete, leave the spotlight.

Saying what we do not know

Big operators aggregate, round, declare climate neutrality for the whole group never for a single site, and mix offset credits with real consumption that happens here, where we live.

What we are trying to do with Substrato is put in writing what we know — and what we do not — with sources and notes.

A polished infographic promises zero impact without saying where, how much, with which water and which local grid. A figure with declared uncertainty can be checked, contested, corrected.

Why it concerns you

Artificial intelligence lives somewhere. Whoever decides where the next hyperscaler rises decides who gets water, who gets affordable energy, who lives next to continuous cooling noise. These are territorial — hence political — decisions, but you can only debate them after you have seen them.

That is why Substrato is a map, not a PDF to forget: a citizen, a journalist, a councillor can open a supply chain and form their own view. The cloud is a successful alibi. The substrate has coordinates, weight in tonnes, an electricity bill. Substrato is open source, with no industrial funders behind it.