Could the Future of AI Data Centres Be in Space?
Artificial intelligence has a physical infrastructure problem.
Behind every apparently weightless interaction with an AI system sits an increasingly large collection of processors, servers, cooling equipment, electricity infrastructure and data centres.
The International Energy Agency estimates that electricity consumption from data centres worldwide could more than double to around 945 terawatt-hours by 2030.
That would be slightly more electricity than Japan consumes today.
As artificial intelligence becomes more capable and more widely used, an uncomfortable question is beginning to emerge: where are we going to find the energy and infrastructure required to keep scaling it?
One start-up has proposed an answer that sounds more like science fiction than infrastructure policy.
Put the data centres in space.
There Is Already an NVIDIA GPU in Orbit
Starcloud, a US start-up developing orbital computing infrastructure, launched its Starcloud-1 satellite in November 2025.
Inside it was an NVIDIA H100, one of the high-performance GPUs that helped power the generative AI boom on Earth.
This was not simply a symbolic payload.
Starcloud says the satellite subsequently ran a version of Google’s Gemini model and trained a small language model in orbit.
In other words, some of the same computing technology found inside terrestrial AI data centres is already performing AI workloads hundreds of kilometres above Earth.
The experiment is small. Starcloud-1 is a satellite, not an orbital hyperscale data centre.
But the ambition behind it is considerably larger.
Starcloud wants to establish whether substantial amounts of the computing infrastructure required by artificial intelligence could eventually be moved off the planet.
Why Would Anyone Put a Data Centre in Space?
The answer begins with energy.
Solar panels on Earth have an obvious limitation. Night arrives.
Clouds, weather and the atmosphere also affect the amount of solar energy that reaches them.
A spacecraft placed in the right orbit can receive sunlight for almost the entire day. Solar radiation is also stronger before it passes through Earth’s atmosphere.
Starcloud estimates that peak solar generation in space can be roughly 40 per cent higher than on Earth and that carefully selected orbits could provide solar arrays with a capacity factor above 95 per cent.
That could potentially provide enormous quantities of renewable electricity without competing for land or requiring the same terrestrial electricity-grid infrastructure.
Starcloud ultimately envisages orbital computing installations measured not in megawatts, but in gigawatts.
Then There Is Water
Data centres do not only consume electricity.
They also generate enormous amounts of heat.
On Earth, cooling that equipment can require substantial infrastructure and, depending on the cooling system and location, significant quantities of water.
That creates an increasingly difficult sustainability question, particularly when new data centres are proposed in regions where water supplies are already under pressure.
Space appears to offer an extraordinary alternative.
There is no need to consume fresh water through conventional evaporative cooling systems.
But describing cooling in space as “free” would be misleading.
A vacuum is actually a difficult place to remove heat because there is no surrounding air to carry it away through convection.
Instead, orbital data centres would need enormous radiators capable of transferring waste heat into space as infrared radiation.
The absence of water consumption is attractive.
The engineering required to make it work at data-centre scale is formidable.
What Does Space Actually Offer?
| Potential Advantage | Why It Matters | The Catch |
|---|---|---|
| Near-continuous solar energy | Certain orbits could provide almost constant access to sunlight. | Enormous solar arrays would need to be launched and deployed. |
| Stronger solar generation | Solar radiation is not reduced by Earth’s atmosphere. | Generating electricity is only useful if the entire system can operate reliably in orbit. |
| No freshwater cooling requirement | Could reduce pressure on terrestrial water supplies. | Waste heat still needs to be rejected through large radiators. |
| No competition for terrestrial land | Hyperscale facilities require substantial physical infrastructure on Earth. | Orbital infrastructure creates an entirely different environmental footprint. |
| Potentially massive scale | Space removes some of the physical constraints faced by terrestrial data centres. | Launch capacity, maintenance, radiation and cost remain significant challenges. |
The Economics Are Ambitious
Starcloud argues that orbital computing could eventually offer substantially lower energy costs than terrestrial data centres.
Its projections suggest energy costs could become up to ten times lower, largely because of continuous solar generation and reduced dependence on terrestrial energy infrastructure.
Those numbers should currently be treated as company projections rather than established economics.
The cost of operating a solar panel in orbit is one thing.
Getting thousands of tonnes of computers, solar arrays, radiators, communications equipment and structural components into orbit is another.
Reusable heavy-lift rockets could change that calculation dramatically.
But the economics of launching and operating hyperscale orbital data centres have not yet been demonstrated commercially.
For now, Starcloud has proved that powerful terrestrial AI hardware can operate in orbit.
It has not yet proved that putting a hyperscale data centre there is cheaper than building one on Earth.
Space Creates Its Own Problems
Moving computing infrastructure away from Earth does not make its environmental footprint disappear.
It changes the footprint.
Thousands of launches would consume fuel and produce emissions. Satellites and computing infrastructure would eventually reach the end of their operational lives. Hardware would be exposed to radiation. Components would need to survive launch vibration and extreme thermal conditions.
Then there is the increasingly serious problem of orbital congestion.
Low-Earth orbit is already filling with communications satellites, Earth-observation spacecraft and other infrastructure.
Gigawatt-scale computing installations would introduce another category of enormous orbital infrastructure.
That raises questions about collision risks, space debris, astronomy, regulation and who ultimately gets to occupy valuable orbital territory.
We could solve one infrastructure problem on Earth only to create another above it.
There Is Also a Latency Problem
Not every computing workload makes sense in space.
Moving enormous datasets between Earth and orbit introduces communications constraints. Terrestrial data centres already benefit from sophisticated fibre networks connecting them to users and to one another.
For some workloads, however, orbital computing could make considerable sense.
Satellites already generate huge quantities of information through cameras, radar and other sensors. Instead of transmitting all of that raw information back to Earth, an AI system could process it in orbit and send back only the useful results.
Wildfire detection, weather observations, agriculture and Earth monitoring are obvious examples.
The first serious market for orbital AI may therefore not be moving ChatGPT into space.
It may be processing space-generated data where the data already exists.
An Infrastructure Race Has Begun
Starcloud is no longer merely proposing the idea on a whiteboard.
The company has raised substantial investment to develop larger orbital computing systems and is working towards spacecraft containing considerably more powerful computing infrastructure.
Its experiment also arrives at an important moment.
The artificial intelligence industry is pouring extraordinary amounts of capital into chips, electricity generation, cooling systems and data-centre construction.
The AI race is quietly becoming an infrastructure race.
Whoever can secure abundant energy, computing capacity and efficient cooling may possess an important advantage in the next stage of artificial intelligence.
That makes space considerably more interesting.
The Bigger Question
The idea of putting data centres in orbit initially sounds absurd.
But many pieces of modern technological infrastructure would have sounded equally improbable before the economics changed enough to make them possible.
The question is therefore not simply whether an NVIDIA GPU can operate in space.
We already know that it can.
The real question is whether falling launch costs, increasingly powerful rockets, continuous solar energy and rising terrestrial energy constraints could eventually make space a sensible place to build computing infrastructure.
Perhaps they will.
Or perhaps orbital data centres will prove to be an extraordinarily expensive solution to a problem better addressed through more efficient chips, renewable energy, nuclear power and better-designed terrestrial data centres.
Either outcome makes Starcloud’s experiment worth watching.
Because the AI revolution is forcing us to confront a fact that is easily forgotten when interacting with software on a screen.
Artificial intelligence may live in the cloud, but the cloud still has to live somewhere.
The question now is whether some of it should live above us.