No doubt coincidental to the forthcoming Space X IPO, Elon Musk (and other tech leaders) have been touting a novel answer to meet rising demand for data centres and energy access issues — send them to space and power them with an endless supply of solar energy. With terrestrial challenges around accessing power, community push-back and digital infrastructure as a military target, orbital data centres are being positioned as a clean, secure, and soon to be economically competitive alternative.  

However, there are several reasons why one would be sceptical about such claims. To pick a few:

• Construction. While ‘mini’ data centres designed to process satellite data may be feasible to fire into space in one shot, gigawatt-scale AI data centres would require hundreds of launches at a minimum, along with space‑based robotic assembly plants that don’t yet exist.
• Maintenance. Even sophisticated, largely automated hyperscale facilities on Earth have staff on site daily and there is a regular cycle of upgrade and replacement. The case for data centres in space requires sophisticated robot-based maintenance and an accessible set of spares and upgrades.
• Space debris. There are over 1.2 million space debris objects capable of ‘catastrophic damage’ in orbit, according to the European Space Agency’s 2025 Space Environment Report. That number is projected to increase exponentially, and with it the risk of collisions with and damage to orbital data centres.
• Radiative cooling. The problem of debris comes into sharp relief when you consider that to cool a space data centre, you rely on radiative cooling, as opposed to evaporative or convective cooling used on Earth. While convective cooling moves air or liquid past hot components and evaporative cooling dissipates heat through water evaporation typically in cooling towers, radiative cooling removes heat through infrared radiation. To cool a 5 GW data centre in space (the size proposed by Starcloud) you would need a radiative surface of square kilometres (and a similar size for the solar panels). That is a lot of surface area at risk of flying debris.

But won’t data centres be ‘greener’? On Earth, purely solar-powered data centres aren’t feasible due to… nighttime (in reality, batteries are used to smooth out power demand). But because the sun always shines in the right orbit, a space-based data centre could be entirely solar powered. Great, right?

Not necessarily. What’s missing in that equation is the small matter of having to send the thing into space in the first place. Put simply: do you save more CO₂e (equivalent tonnes of CO₂, the most prevalent greenhouse gas) through space‑based solar power than you expend through repeated launches?
 

Firing an AI data centre into space is climate madness. Use the money to build out more renewables.

Starcloud suggests they could get their kit into space in fewer than 200 launches. However, the numbers don’t add up. On Earth, modern AI servers are heavy and need reinforced structures to support them. A NVIDIA DGX H100/H200 system weighs around 130 kg and draws around 10 kW power. For Starcloud’s 5GW data centre you would need around 500,000 of these, so that’s circa 65,000 tonnes. SpaceX’s Starship can carry 100-150 tonnes per trip, so even using the maximum 150 tonnes at 85% packing efficiency, that’s 510 trips. And in reality, you aren’t only sending up the servers. By our calculations you’d need at least three times the mass for non-IT equipment, including the radiative cooling apparatus, some shielding and many other components. Now we’re looking at least 2000 trips.  

NASA has calculated that a Starship launch emits around 3500 tonnes of CO₂e. So, to deploy a single (albeit very large) 5GW data centre in space it would take 2000 trips, costing seven million tonnes of CO₂e - equivalent to around 1.7 million petrol cars on the road for a year. There are some ‘embodied emissions’ to account for too - for example - building the Starship (or rather twenty of them since each is (optimistically) expected to make 100 trips), but this is likely small compared to cumulative launch emissions.  

So how does that compare to the emissions saved? In the US, the grid is still somewhat ‘dirty;’ generating power relies on burning a lot of fossil fuels and therefore emissions are high. To run a 5 GW data centre on the US grid for a year, currently, you would be looking at around 13.5 million tonnes of CO₂e. More than all those launches. However, that is entirely down to the fossil fuel burned to generate electricity in the US. Compare that to a renewables-dominated grid like in the Nordics where lots of AI data centres are located and emissions will be close to zero.  

A space data centre wouldn’t come close to ‘paying back’ the carbon used to launch it within the useful service life of a GPU, which according to recent reports is only one to three years. Couple that with the $90 million to launch a single starship and both the financial and environmental arguments begin to unravel. It would cost $180 billion for the 2000 launches needed to deploy a 5GW data centre, equivalent to a quarter of the $690 billion invested in renewable energy globally last year. I know how I’d spend the money.

For further insights on how the ICT sector can accelerate the clean energy transition, please visit the Carbon Trust's ICT sector vision and solutions page.