The idea arrives, as these things often do, dressed up as destiny. AI is gobbling power at a rate that makes even the most ambitious data centre plans look like a child’s Lego project. The grid is groaning, planners are dithering, and locals would rather not have another enormous humming shed parked near their village. So naturally, the industry has decided the answer is… space.
Yes. Data centres. In orbit. At first glance, this sounds like the sort of thing you hear at a conference just before lunch, when everyone is hungry and slightly suggestible. But it raises a fair question. Is this the future of infrastructure, or just a very clever way of not fixing the problems we already have?
Start with the basics. A data centre is, fundamentally, a very expensive way of turning electricity into heat. The building is almost irrelevant. Over its life, the real cost is power. AI has made this worse, pushing demand to levels that would have seemed faintly ridiculous not long ago. A single facility can chew through hundreds of megawatts, and it does so continuously, like a particularly greedy Labrador.
Then there’s cooling. All that computing produces heat, which must go somewhere. On Earth, that means water, airflow, and carefully chosen locations. It’s not glamorous, but it works. Except now we’re running out of the obvious places to put these things. Land with power, fibre, and planning approval is scarce. Even when you find it, you’re often looking at years before anything actually gets built. And none of it matters unless the site sits neatly inside a network, because a data centre on its own is about as useful as a motorway to nowhere.
So, inevitably, someone looked up and said: what about space? On paper, it’s brilliant. Unlimited solar energy. No grid constraints. No local councillors objecting to the view. Cooling? Easy. Space is basically one giant heat sink. No fans, no water, no complaints.
Unfortunately, there is a small catch. Getting anything into space is still ruinously expensive. Even with cheaper rockets, you’re paying thousands per kilogram. And a data centre is not a featherweight. Servers, shielding, structure, power systems, all of it adds up. Very quickly.
Then you have to design the thing. On Earth, if something breaks, you send a person with a screwdriver. In orbit, you send a prayer. Everything must work remotely, survive radiation, and keep going without anyone popping in to have a look. That’s not just difficult. It’s expensive in ways that spreadsheets struggle to capture.
But the real problem is simpler. Data has to move. A lot of it. Sending vast quantities of information back and forth between space and Earth introduces latency, bandwidth limits, and a reliance on ground infrastructure that doesn’t yet exist at anything like the required scale. Running heavy AI workloads in orbit and pretending it will feel seamless to users on Earth is, at best, optimistic.
So where does that leave us? Right now, firmly in the “interesting but impractical” category. The economics don’t stack up. The operational model is largely untested. And the risks are not trivial. Space is not a friendly place. Radiation damages hardware. Remember the warning about solar flares. Repairs are difficult. If something goes wrong during launch, you don’t get a second chance. You get a very expensive cloud of debris.
In time, some of this may change. Launch costs will fall. Satellite networks will improve. There are already sensible niche cases emerging, particularly where data is generated in space and can be processed there instead of being beamed back down. Military and specialist applications may well justify the cost.
But the idea of shifting mainstream computing off the planet? Not any time soon.
Which is slightly disappointing, because it would have been quite fun.
The more important point is what this says about Earth. If serious people are discussing putting data centres into orbit, it’s not because space is suddenly attractive. It’s because power, planning, and coordination on the ground are becoming unworkable.
In other words, we’re not trying to escape to space because it’s better.
We’re considering it because we’ve made things here unnecessarily difficult.
And that, frankly, is a far more interesting problem.
By Peter Zanatta
