Recently I have been blogging about why we absolutely have to leave our beloved planet one day (one of my most popular posts in recent times so thank you very much all of you), in the light of the SpaceX IPO and Elon's Musk's vision to get us out of the solar system and beyond (I still think he is a James Bond villain, but that is another story for another day). I then outlined what Elon Musk says that he is aiming to achieve before focussing on his Starlink satellite service which he hopes will finance much of his vision - with more than a little help from the new investors thanks to the IPO. So now it is time to move on to orbital data centres, the next aspect that is also relatively immediately achievable.

An orbital data centre is essentially a GPU supercomputer placed on a satellite in low‑Earth orbit (LEO). Instead of building a terrestrial data centre on land, with grid power, cooling systems, and long construction timelines, companies like SpaceX propose launching solar‑powered compute satellites that run AI workloads from space. Even though the idea sounds like it was born in science fiction, the business logic is surprisingly grounded.

Incidentally, there is nothing wrong with futurist science fiction as it shows us what can be and from a technological perspective that is fascinating. The ethics are a kind of take it or leave it depending on your specific interest and that is where the story telling comes in.

Such orbital data centres are specifically not for communications or imaging as are most satellites currently in orbit; rather they are for computation and will be largely made up of GPU clusters / AI accelerators, large solar arrays (SpaceX’s AI1 uses ~70‑metre wings generating 120–150 kW), deployable radiators to dump heat into space (1,400 W/m² cooling density) and probably most significantly Starlink connectivity for low‑latency access to Earth-based customers. By adding to existing and functioning technology it makes this next step not only logical but feasible too.

To give this even more perspective SpaceX has filed for up to one million such satellites, aiming for 1 terawatt of orbital compute by 2027–2028. We are not talking about even a decade away.

There are significant advantages to putting such infrastructure into space. First and foremost Earth’s power grid is the bottleneck. In major data‑centre hubs, securing a new high‑power grid connection can take 5–7 years. SpaceX can launch orbital compute in 18 months, avoiding grid delays that cost companies $350M+ in foregone revenue. Furthermore and most obviously cooling is easier in cold, cold space. Radiators can dump heat far more efficiently than terrestrial cooling towers. While this is perfectly reasonable and will be an offset against global warming I personally wonder if such heat should be directed into either hot water or power generation. The clincher is that it increasingly becoming more and more affordable. The costs are falling for Falcon deliveries and in the longer term given Starship’s full reusability launch costs could fall to as little as $20/kg which is about 1% of current costs.

To give that nominal $20 some perspective. Ryanair currently charge $15 per kg for excess luggage.

So how will this work in practice?

Musk has outlined three models which I will explain logically and in order.

Model A will offer a premium service for external customers and in particulare for those who cannot wait for terrestrial grid capacity. CapEx is estimated at $310M per MW, with 3.5–4.5 year payback. For a project of this scale those numbers are beautiful. SpaceX believes that the service it offers will benefit in particular AI labs needing immediate scale,  financial firms needing low‑latency inference and governments wanting sovereign compute capacity.

Secondly (Model B), which is probably the most significant of all the models, it will provide internal infrastructure for Musk’s Companies, such as SpaceX, Tesla, and xAI, all of which have massive demand for such a service. xAI’s Memphis supercluster is one of the largest GPU installations on Earth and Tesla’s FSD and Optimus robotics generate continuous training workloads, all of which need to be faciliated. Such Data Centres contained within SpaceX’s orbital constellation will primarily serve internal demand, not external cloud customers. In doing so SpaceX will be able to guarantee independent utilisation, free of terrestrial constraints and it will fit neatly in to Musk's organisational vertical - simply by keeping it "internal" and not looking for vendors and the problems that may arise with them.

This brings us neatly to Model C in which Musk will seek to centralise and contro the whole stack with a kind of space monopoly. His companies, from start to finish will control each aspect, from launch (Starship) to the final delivered product(s) in space and lets not forget that this will include manufacturing (Starlink factory), networking (Starlink), AI models (xAI) and the creation of semiconductors (Terafab project). This will make SpaceX's domiance almost unassailable and could create a $1.4 trillion orbital compute business.

The impacts, assuming it happens, will be massive and could go as far to reshape AI and global Infrastructure. This in itself may challenge the way we do a lot of things and how we view the whole world With AI growth outpacing Earth’s ability to build data centres space-based ones have to be the alternative. Back here on Earth demand for data centres and more importantly the electricity to power them could reach 4% of global consumption by 2035. SpaceX is targetting10,000 Starship flights per year by 2040 (just let that sink in for a minute!) and with such capacity it will massively outperform anything earth based as it will create and enable planetary-scale AI, no longer reliant on grid coverage, to power among other things. autonomous vehicles, humanoid robots (Skynet and the Terminators ! ? ! ) and global AI agents.

At this point the viability of such development is very high, the moral questions of how much power it will give SpaceX and ultimately Elon Musk is another matter. Power is often a benign force, how it is utilised is not. It might not quite be time to call in James Bond yet, but it might come to that...

As always stay safe and will my friends.