IEEE Spectrum has poured cold water on Elon Musk’s space-AI prophecy where you put data centres in space.
For those not in the know, Musk has predicted that space will become the cheapest place to run AI within a few years.
However, IEEE has been crunching the numbers and reckons that Musk’s hype does not add up.
The report says SpaceX’s proposed million-satellite constellation would require enormous increases in launch capacity and satellite manufacturing capacity. It would still face ugly problems with cooling, radiation, maintenance, latency, orbital debris and interference with astronomy.
There are roughly 14,500 active satellites in orbit, and Musk’s Starlink constellation accounts for about two-thirds of them.
Both launch cadence and satellite manufacturing capacity would have to scale astronomically to deploy a million orbital data-centre satellites.
For context, there have been roughly 7,000 orbital launches throughout human history.
To lob a million satellites into low Earth orbit on SpaceX’s Starship, designed to carry up to 60 satellites per vehicle, would require 16,666 launches.
That assumes every one of those launches was devoted exclusively to satellite deployments which is unlikely. SpaceX launched a record 165 orbital missions in 2025.
Even at 10 times that cadence, the job would take a decade, assuming rockets, regulators and reality all behaved themselves.
At Starlink’s current pace of about 4,000 satellites a year, even a generous tenfold increase in capacity would still leave a 25-year slog. Short of a manufacturing revolution, the idea looks less like an AI gold rush and more like an expensive PowerPoint cult.
Then there is the technical issue of putting a data centre in space. Heat is another brutal problem, because dissipating it in space needs enormous radiators.
IEEE Spectrum editor Dina Genkina noted that the startup Starcloud has sent only one Nvidia H100 GPU into orbit, and its radiator was too weak to allow the chip to run at full power. A single 700-watt H100 would require about 1.4 square metres of radiator area.
A 100-megawatt data centre could require 2,500 radiators, each measuring 80 square metres.
