Keeping a data centre cool on Earth is a surprisingly large part of the job. Cooling can swallow anywhere from a tenth to a third of a facility’s entire energy budget, and the thirstiest sites drink millions of litres of water to do it. Move that data centre into orbit and the whole approach collapses. There is no air to blow across the servers, no cooling towers, and no river to dump the heat into. The only way left to get rid of waste heat is to radiate it away as infrared light into the blackness of space, through vast panels. That single fact shapes everything about the idea of computing in orbit.
Why cooling is such a big deal on Earth
Every computer turns electricity into heat, and a data centre is thousands of them packed together. Left unchecked, that heat would quickly cook the machines, so a large share of the effort and expense goes not into computing but into carrying the heat away.
The bill is real. Cooling commonly accounts for something in the range of a tenth to a third of a data centre’s total energy use, and often more in hot climates. Much of that cooling relies on water, evaporated to carry heat off just as sweat cools skin. A single one-megawatt facility can get through tens of millions of litres a year, and the largest operators consume water by the billions of litres, a growing source of tension where supplies are tight.
How Earth sheds heat, and why none of it works in space
On the ground, heat has easy ways to escape. It moves by conduction into cooler materials, by convection as air or water flows past and carries it off, and by evaporation in cooling towers. Every one of those methods depends on having a surrounding fluid, air or water, to hand the heat to.
Space has neither. The vacuum around a spacecraft is not cold air; it is very nearly nothing at all. With no air to convect into and no water to evaporate, and nothing to conduct heat into, the familiar cooling methods simply stop functioning. You cannot fan away heat that has nowhere to go.
The only exit is radiation
There is one mechanism that still works in a vacuum, and only one: thermal radiation. Every warm object glows in infrared, giving off energy as light, and in space that glow is the sole route for heat to leave. It streams away as invisible infrared into the cold sky and does not come back.
Spacecraft already live by this rule. The International Space Station and ordinary satellites carry radiator panels, plumbed with fluid loops that ferry heat from the interior out to the panels, which then shine it away. A data centre in orbit would need exactly the same trick, but on a far grander scale, because it would be trying to shed the heat of a small power station.
Why the radiators are the hard part
Here is the catch that turns a neat idea into an engineering wall. Radiating heat is a relatively feeble way to get rid of it, and how much a panel can shed drops sharply as it cools. Electronics can only run so hot, and at those modest temperatures a radiator has to be enormous to reject a serious amount of power.
For a data centre drawing megawatts, the radiator panels could end up larger than the solar arrays feeding it, and engineers working on the concept describe heat rejection, rather than power or computing, as the real limiting factor. It gets harder still because the panels are bathed in sunlight and in heat reflected off the Earth, both of which they must be angled and engineered to avoid soaking up. In space, cooling is not a footnote to the design. It largely is the design.
Why anyone is trying anyway
Given all that, why bother? Because the other side of the ledger is tempting. In the right orbit the Sun never sets, offering nearly constant power. There is no land to buy, no local water to drain, and no electricity grid to overload, at a moment when the demand for computing, driven by artificial intelligence, is rising faster than terrestrial infrastructure can comfortably supply.
The interest is no longer only theoretical. The European Space Agency has run a feasibility study, called ASCEND, into orbital data centres. A startup, Starcloud, launched a satellite in late 2025 carrying a high-end AI chip to operate in space, and Google has begun exploring a space-based design of its own. These are early demonstrations and studies, not working orbital data centres, and it is worth keeping that distinction clear.
What to watch
The path forward runs through small steps: a handful of processors in orbit, then larger power budgets, then the deployable radiators that any serious scaling will demand. The physics is not in doubt. In space you trade Earth’s air and water for sunlight and radiator panels, and the question is whether the thermal engineering and the launch costs can ever make that trade worthwhile.
What to watch is whether heat rejection can be scaled without the radiators growing impossibly large, because that, more than the computing itself, is where the idea will succeed or stall. For now, the blackness of space stands ready as the ultimate heat sink, reachable only through panels quietly glowing in infrared.
