The arithmetic of this ambition has shifted because of Starship. If SpaceX’s heavy-lift rocket can collapse launch costs by an order of magnitude, the economic penalty of putting a server in orbit may finally fall below the rising cost of keeping one on the ground. However, the engineering and regulatory hurdles remain formidable. Managing a million-satellite fleet would push orbital congestion to a breaking point, significantly heightening the risk of the Kessler Syndrome”a cascade of debris that could render certain orbits unusable. Furthermore, space is a hostile environment for the delicate GPUs required for AI; radiation hardening adds weight and cost, while the lack of on-site maintenance means a single hardware failure is permanent.
There is also the matter of governance. A space-based compute sovereign would operate largely outside the tax and environmental regimes of nation-states, raising thorny questions about data residency and the monopolisation of orbital shells. For now, the proposal should be viewed with a healthy measure of scepticism. It is as much a speculative hedge against the physical limits of Earth as it is a technology roadmap. Whether this represents a genuine shift in infrastructure or merely a high-stakes play for “orbital real estate” before regulators close the window, it confirms one reality: the AI boom has become an infrastructure problem so large that the planet itself is starting to feel too small to contain it.
