For most of the history of computing, software managed to maintain a useful illusion: that it did not really live anywhere.
A program was code. A website was a URL. The cloud was, conveniently, a cloud. We spoke about digital products as though they occupied some weightless layer above the physical world, detached from weather, geography, rivers, power stations and temperature.
Artificial intelligence is making that illusion increasingly difficult to maintain.
Imagine someone saying that they intend to build their own ChatGPT and then move it to Greenland to keep it cool.
It sounds like a joke.
It is also an unexpectedly good description of one of the strangest problems facing modern computing.
An AI model may appear on a laptop as a clean little text box. Behind that box, however, is an industrial system. There are accelerators consuming electricity, racks producing heat, cooling equipment removing that heat, networks moving enormous quantities of data and buildings designed around machines whose appetite for power has become impossible to hide.
The more sophisticated the digital world becomes, the more physical it seems to get.
This is the peculiar reversal happening inside modern technology.
For decades, progress meant making the machine disappear.
Computers moved from dedicated rooms onto desks, then into pockets. Servers disappeared into the cloud. Storage became something users rarely had to think about. Software companies could grow enormous while their customers remained almost completely unaware of the physical infrastructure underneath them.
AI brings the machine back into the room.
Not literally, perhaps. The machine may be thousands of miles away. But suddenly its location matters.
How much electricity is available there?
How expensive is it?
Where does that electricity come from?
How much heat can the facility remove?
How much water does the cooling system require?
Can the electrical grid support another enormous computing campus?
Can new transmission infrastructure be built quickly enough?
And, eventually, an oddly primitive question appears beneath one of humanity’s most sophisticated technologies:
Is it cold there?
There is something wonderfully absurd about this.
We are building machines capable of generating software, analyzing proteins, producing images, translating languages and holding conversations across enormous bodies of human knowledge. Yet somewhere underneath all of that abstraction sits a stubborn thermodynamic fact.
The computers get hot.
Intelligence, once industrialized, produces heat.
That fact changes the geography of technology.
The traditional map of the software industry was drawn around talent, capital and networks of companies. The emerging map of AI infrastructure has additional coordinates: electricity, grid capacity, land, water, fiber and climate.
A place that once seemed peripheral to the technology industry can become interesting because it possesses something Silicon Valley cannot manufacture with code: abundant energy, available land or naturally cold air.
This does not mean that the future consists of dragging every AI cluster toward the Arctic. Geography is more complicated than temperature. Data centers require reliable power, connectivity, skilled maintenance, supply chains, political stability and enormous amounts of infrastructure. A cold coastline is not automatically a good place to build a computing empire.
But the direction of the question is fascinating.
The cloud is acquiring a climate.
For years, the technology industry trained us to think upward. Everything was becoming virtual, distributed and detached from place.
AI forces us to look downward again: at concrete floors, electrical substations, cooling pipes, transmission lines and the ground beneath the data center.
Perhaps this is one of the great ironies of artificial intelligence.
The closer machines come to imitating something as abstract as thought, the more brutally material their existence becomes.
A chatbot feels almost immaterial. Ask it a question and words appear. There is no smoke, no engine noise, no visible machinery. The interface gives no indication that somewhere, processors are switching billions upon billions of times while an industrial cooling system carries the resulting heat away.
The conversation is weightless.
The infrastructure is not.
And this may eventually influence where parts of the computing economy are built.
The nineteenth century placed factories near coal, rivers, ports and railways. The twentieth century built industries around oil, highways and electrical grids. The twenty-first century may discover that intelligence has its own geography.
Not because intelligence itself needs mountains, rivers or winter.
Because the machines producing it do.
So the idea of taking an artificial intelligence north to keep it cold is funny only for a moment.
Then you realize that we have built something so computationally hungry that climate itself can become part of the architecture.
The cloud was never really a cloud.
It was always a building somewhere.
Now the building is getting hot.